The NYU Journal of Intellectual Property and Entertainment Law is proud to present Volume 7 Issue 2 of the Journal. While PDFs of the individual articles may be found accompanying their respective posts, you may view and download a PDF of the complete issue here
A Judicial ‘Supplement’ to Advertising Law: The Fourth Circuit’s GNC Decision and Policy Implications for the Dietary Supplement Industry
By Gia Wakil*
Download a PDF version of this article here
Introduction
Every home has a story, as the old adage goes, and the story in this Note is set in the modern-day vitamin cabinet. Few items are as ubiquitous in American homes as the pill bottle. From vitamins like A and C, or minerals like iron and zinc, the supplements found in a single household typically run the full alphabet. The majority of adults in the United States take one or more dietary supplements occasionally, or even every day,[1] with sales to American consumers exceeding $35 billion per year.[2] Americans rely on supplements to ameliorate nutritional deficiencies or to maintain their health. And yet, despite the prevalence of supplements, concerns abound about their efficacy and the veracity of claims supporting the use of supplements. Oftentimes, the prospect of shorter colds, stronger nails, or improved general health justifies the price of the “gamble” in the minds of many consumers.
It was in similar, health-conscious hopes that consumers in the In re GNC Corp. [3] case had reached for bottles of glucosamine-chondroitin, two common ingredients in joint health specialty products.[4] The labels touted that the mixtures “promote[] joint mobility and flexibility” and provide “[c]linical strength for daily long-term use.” GNC allegedly had, and on some labels cited to, a reasonable scientific basis for these claims, though its legitimacy was tarnished in the face of mounting scientific evidence indicating otherwise.
After years of medical controversy examining whether glucosamine-chondroitin is effective, lawsuits against a variety of its manufacturers and sellers popped up across the United States.[5] Perhaps the consumers were tired of “gambling” on glucosamine and thought that the products did not live up to their billing. More likely, plaintiffs’ lawyers had been following the debate, and they believed there were issues to be asserted in the form of consumer class actions.[6] Multiple false advertising lawsuits against GNC commenced between March and December 2013, with allegations of consumer deception across California, Florida, Ohio, New York, Illinois, Pennsylvania, and New Jersey.[7]
The controversy surrounding truthfulness in supplement advertising has only grown louder in recent years, and for good reason. Over the last fifteen years, sales of vitamins, minerals, and nutritional and herbal supplements — which, together, comprise the dietary supplement industry — have surged. In December 2013, McKinsey & Company estimated the global value of the supplement market to be $82 billion, with a high concentration of that value in the U.S.[8] High levels of sales appear to correlate with high consumer confidence. The majority of U.S. adults — 68 percent — take supplements, and nearly 85 percent of those consumers express “overall confidence in the safety, quality and effectiveness of dietary supplements.”[9] Competition in the supplement industry is fierce and no one company accounts for more than a five percent share.[10]
Unbeknownst to many consumers, the laws governing supplement labeling and advertising are notoriously tricky to navigate. The Food and Drug Administration (FDA) issues rules and regulations regarding supplement labeling, marketing, and safety,[11] but there are significant limitations to FDA oversight. For example, the manufacturer or seller does not need to prove that a claim is accurate to the FDA’s satisfaction before it appears on the product,[12] and the agency reviews substantiation for claims periodically, as resources permit.[13] The Federal Trade Commission (FTC) polices health and safety claims made in advertising for dietary supplements,[14] but its oversight is similarly limited by the availability of resources.[15]
High sales, patchy government supervision, and dubious marketing practices have created a “perfect storm” on the litigation frontier. False advertising in the consumer class action context is a rapidly growing area of law.[16] GNC is a rich case study for exploring issues related to truth in advertising, particularly when dealing with scientific controversy. In GNC, the Fourth Circuit held that the existence of a single expert in agreement with a claim bars the statement from being found literally false.[17] Since Plaintiffs did not challenge the validity of GNC’s study that supported the claims found on bottles — even after an opportunity to amend their complaint — they could not prove that glucosamine-chondroitin was not beneficial to joint health. In other words, at least one reasonable expert or study is sufficient to carry the claim on the bottle, even if the vast weight of scientific evidence suggests otherwise. Furthermore, a “battle of the experts” did not necessitate a jury trial, nor did it forestall judgment in the supplement manufacturer’s favor.
This Note argues that the Fourth Circuit’s holding in GNC is the preferred solution to the truth-in-advertising dilemma that is rampant in the supplement industry, particularly in periods of scientific controversy. Scientific evidence is neither static nor consistent; there can be two pools of reasonable evidence on either side of a medical controversy, and the minority position may well be proven right. Reserving judgment for juries beyond the pleadings stage would overestimate juries’ abilities to resolve highly technical scientific controversies. Both consumers and retailers benefit from consistent interpretation of consumer protection measures, and the Fourth Circuit has articulated a workable standard for these various state law claims.
The argument unfolds in three parts. Part I surveys the landscape of advertising laws, with heightened focus on the Lanham Act and state consumer deception statutes. Supplement advertising should be treated as a carve-out amidst this patchwork of laws, and I will argue that the Fourth Circuit’s ruling, if appropriately limited, does no damage to the existing doctrines. Part II offers a detailed description of the GNC case, both at the district and appellate court levels. This background serves to illustrate why the unanimous GNC judgment is an interesting solution to a difficult truth-in-advertising question. Part III assesses the policy implications of the Fourth Circuit’s decision. It begins by addressing criticism of the GNC ruling, which was presented in an Amicus Brief submitted by a prominent group of law professors. It then responds to the criticism provided in the Amicus Brief and counters with the merits of the Fourth Circuit’s holding. The Conclusion reviews the argument and demonstrates why the analysis presented is persuasive.
Part I: The Legal Landscape of the Supplement Industry
A. Sources of Truth-in-Advertising Law
There is a patchwork of federal, state, regulatory, and common law sources of truth-in-advertising law, in addition to industry-specific trade associations that advise dietary supplement manufacturers and retailers. Both the FTC and FDA[18] can initiate their own investigations into false advertising and labeling of dietary supplements. Their overlapping jurisdiction and shared enforcement responsibilities are explained further in Section I.B., infra. However, since actual oversight by these bodies is notoriously constrained, consumers and business competitors have often resorted to alternative forms of legal redress. Companies may challenge claims made by their competitors under the federal trademark law, Lanham Act § 43(a), discussed in Section I.C., infra.[19] While the federal cause of action is not available to consumers, the Lanham Act remains instructive in understanding state deception laws that are available to consumers.[20] The relationship between the Lanham Act and state consumer protection laws is explored further in Section I.D., infra. Lastly, the dietary supplement industry finds guidance from self-regulatory associations: the Council for Responsible Nutrition (CRN) is its primary trade association,[21] and the Better Business Bureau’s National Advertising Division (NAD) offers alternative forms of dispute resolution. [22]
B. The FTC-FDA Regulatory Regime
Congress has entrusted matters of food, drug, and supplement policing to both the FTC and FDA,[23] which have overlapping jurisdiction and work together to ensure consistency in consumer products.[24] “In 1971, the agencies issued a memorandum of understanding under which the FTC assumed primary responsibility for advertising and the FDA assumed primary responsibility for labeling of food, medical devices, and cosmetics.”[25] Advertising in the FTC’s domain includes print and broadcast ads, infomercials, catalogs, and similar direct marketing materials; labeling refers to the information panels on product itself, such as the nutritional content and manufacturer address.[26]
There is an enormous disparity between what the law permits and what the agencies enforce. Dietary supplements have posed a unique challenge to the joint FTC-FDA working relationship. Neither food nor drug, dietary supplements occupied a “liminal” regulatory category for much of the 20th century.[27] By the 1990s, however, Congress overrode attempts by the FDA to regulate supplements more extensively. The resulting legislation paved the way for the modern explosion of supplement products on the American market.[28]
1. FTC Regulation of Supplement Advertising
The FTC is the federal consumer protection agency charged with safeguarding consumers against “unfair and deceptive trade practices,” under the authority granted to it in Section 5(a) of the Federal Trade Commission Act. It uses two investigative methods to challenge advertising it finds to be false: either compelling information from a potential defendant (for example, in the form of a subpoena), or requesting voluntary cooperation (in the form of an access letter). Notably, consumers do not enjoy the same authority in compelling disclosure of a corporate advertiser’s studies.
The FTC has articulated its own standard for actionable false advertising. The 1972 Pfizer doctrine[29] is an unsubstantiated theory of liability: no dissemination (of an ad claim) without prior substantiation. For an advertisement to be substantiated, the advertiser must have had a “reasonable basis” for its advertising claims before they are disseminated. For health or safety claims, which include representations made in connection with dietary supplements, the Commission has typically required a relatively high level of substantiation. In such cases, the “reasonable basis” must be “competent and reliable scientific evidence,” typically defined as “tests, analyses, research, studies, or other evidence based upon the expertise of professionals in the relevant area, that has been conducted and evaluated in an objective manner by persons qualified to do so, using procedures generally accepted in the profession to yield accurate and reliable results.”[30] The FTC reaffirmed this standard in its Policy Statement Regarding Advertising Substantiation, and there are numerous representative cases applying the doctrine.[31]
2. FDA Regulation of Supplement Labeling
Unlike drugs, dietary supplements do not need FDA approval before being legally marketed in the United States. Under the Dietary Supplement Health and Education Act (“DSHEA”) of 1994,[32] the FDA’s role in regulating the dietary supplement industry is constrained. Under DSHEA, a supplement company is responsible for determining that its products are safe and that any representations made are substantiated by adequate evidence of their truthfulness. A company does not have to provide the FDA with the evidence it relies on to substantiate safety and effectiveness unless specifically requested. Instead, the FDA has relied on a disclosure theory of consumer protection, requiring firms to identify the product as a dietary supplement and include a “Supplement Facts” panel that identifies each ingredient contained in the product.[33]
C. The Lanham Act
Congress enacted the Lanham Act (codified at 15 U.S.C. §§ 1051-1127 (2012)) in 1946. It serves both as the federal trademark law and as a primary source of truth-in-advertising regulation. Section 43(a) of the Lanham Act provides that “any person who believes that he or she is or is likely to be damaged” by a false or misleading description or representation of fact may sue.[34] The Trademark Revision Act of 1988 substantially broadened the scope of section 43(a) to cover a company’s false representations about itself and others.[35]
Section 43(a) is the core legal protection for those injured by false advertising. The Supreme Court has interpreted standing under section 43(a) to be limited to business competitors; consumers are excluded from filing suits under the Lanham Act. In Lexmark International, Inc. v. Static Components, Inc., the Supreme Court explained that standing to sue for false advertising under the Lanham Act requires pleading “injury to a commercial interest in sales or business reputation,” and that injury must be “proximately caused by the defendant’s misrepresentations.”[36] The injury cognized by Lexmark is “unfair competition” through false or misleading advertising, rather than consumer deception or confusion. Accordingly, consumers’ interests do not fall within the “zone of interests” protected by section 43(a)(1)(B).[37] As a result, plaintiffs in these lawsuits are typically commercial competitors of the alleged false advertiser.
Since remedy under the Lanham Act is unavailable to non-competitors, consumers often rely on state consumer protection statutes to bring cases against advertisers. Since these state statutes effectively police the same mischief, federal common law remains instructive in understanding the state law cause of actions.[38]
Section 43(a)(1)(B) establishes liability for two modes of advertising: advertising that is false and advertising that is misleading. Actionable representations include statements made about one’s own goods or services, as well as the commercial disparagement of others. These two categories of advertising, as well as the specific violations that they encompass, are next considered.
1. Two Modes of False Advertising
Courts are tasked with determining whether representations are either false or misleading. Both incur liability under section 43(a), but there are important doctrinal and pleading distinctions. The analysis begins with five basic elements: (i) whether the alleged misrepresentation is false or misleading, as opposed to non-actionable puffery,[39] (ii) whether a plaintiff can demonstrate actual deception or capacity for deception as a result of the falsehood, (iii) whether the falsehood or misleading information is material to a consumer’s decision to buy,[40] (iv) whether a plaintiff was harmed, either by direct diversion of its sales or by a lessening of the goodwill associated with its products, and (v) whether the falsehood or deception occurs in interstate commerce.[41] These elements are conjunctive, and a plaintiff must meet all five to trigger liability.
i. Literal Falsity
Literally false representations communicate one unambiguous message, either verbally or visually, that is untrue or unsupported. These representations violate the Lanham Act without proof of consumer deception; instead, courts presume that the buying public has received the false message.[42] Consequently, plaintiffs in these cases can avoid the time and expense of preparing consumer surveys regarding the ad’s message. The burden of proof in falsity cases is on the plaintiff, and falsity is assessed on the basis of scientific testing or related types of extrinsic evidence.[43]
A classic example of a literally false representation comes from Coca-Cola.[44] In 1982, Tropicana featured athlete Bruce Jenner in a television commercial for orange juice, squeezing juice out of an orange directly into a Tropicana carton while saying, “[i]t’s pure, pasteurized juice as it comes from the orange.” In fact, the juice was heated and, in some cases, frozen before packaging. Further, the juice did not in fact come “pasteurized” straight from the orange. The Second Circuit granted preliminary injunctive relief against the “blatantly false” statement. The plaintiff did not have to make a showing that the advertisement would mislead the consuming public.
ii. Literal Falsity by Necessary Implication
A court may evaluate an advertisement for falsity even if the representation is implied by context, rather than stated directly. This category of false advertising is referred to as false by necessary implication. Actionable advertisements of this type convey one unambiguous message. While the message is conveyed implicitly, the meaning of the message is clear and unequivocal. Ultimately, if the impression left on the viewer conflicts with reality, it is effectively treated as if that false impression was directly stated.
The Second Circuit embraced the false-by-necessary-implication doctrine for the first time in Time Warner Cable, Inc. v. DIRECTV, Inc.[45]— a case which provides an excellent illustration of this type of actionable falsity. A series of DIRECTV commercials featured celebrities touting the merits of the satellite service provider’s 1080i high definition (HD) transmission. A commercial from the “Source Matters” campaign concluded with William Shatner stating that “settling for cable would be illogical.” This statement was made in the context of surrounding text (“amazing picture quality of […] DIRECTV HD”) and images (a very clear DIRECTV picture and a far inferior picture from an anonymous second provider, though cable was obviously targeted). Without extrinsic evidence, the district court determined that Mr. Shatner’s assertion (“settling for cable would be illogical”) “could only be understood as making the literally false claim that DIRECTV HD is superior to cable HD in picture quality.”[46] In reality, however, DIRECTV and cable HD’s picture quality were equivalent. The Second Circuit upheld this ruling for Time Warner, concluding that the impression left on the viewer conflicted with reality.[47]
2. Misleading Advertising
The second category of actionable false advertising claims are, while not literally false, deemed misleading to consumers. Such ads contain representations of fact about a product or service that are ambiguous or tend to deceive consumers. The statements may be literally true, and yet have the tendency to mislead. The language of section 43(a) deems merely misleading representations equally as objectionable as those that are literally false, explicitly rendering unlawful a “false or misleading description of fact, or false or misleading representation of fact.”
There are a number of important differences between false and misleading advertising, even though both types of representations violate section 43(a). Although the categories are doctrinally distinct, the evidentiary differences are more significant in practice. Literal falsehoods — those that are false on their face — are actionable without proof of consumer deception, as in the Coca-Cola and DIRECTV cases previously considered. In a case of misleading advertising, plaintiffs must present extrinsic evidence to establish consumers’ perception of the implied falsehood.[48] Such proof typically includes consumer surveys, direct consumer testimony, or consumer comments received in the ordinary course of business.
In Vidal Sassoon, Inc. v. Bristol-Myers Co.,[49] the Second Circuit addressed literally false as well as misleading claims. This case provides a clear example of the difference between literal falsity and misleadingness. In a television commercial, a spokesperson states that “in shampoo tests with over 900 women, Body on Tap got higher ratings than Prell for body. Higher than Flex for conditioning. Higher than Sassoon for strong, healthy looking hair.” The literally false claim is that the tests involved over 900 women; in fact, only two-thirds of testers were actually adult women. The claims regarding higher satisfaction were deemed misleading based on a consumer survey assessing the message of the ad. Consumers surveyed about the commercial said that they thought each tester had tried at least two of the named products in order to compare their quality. Since this was not the actual testing procedure, the court held that consumers were deceived by the ad’s claims, and that the claims were actionable as misleading.
D. State Consumer Protection Statutes
Every state has a consumer protection law that prohibits deceptive practices.[50] Many of these statutes take the form of a “little” FTC Act, incorporating the language of Section 5 of the statute for which they are named.[51] These statutes provide the basic protections for consumers engaging in thousands of transactions across the United States, prohibiting “unfair methods of competition or unfair deceptive trade practices” as well as “all forms of fraudulent, deceptive, and unfair acts.” While many statutes track the federal guideline, they nevertheless may vary in form and strength from state to state.[52] State attorneys general and consumers may bring suits pursuant to these acts.
As mentioned infra, the state consumer deception statutes at issue in GNC were interpreted in accordance with the great body of federal common law surrounding false advertising and the Lanham Act. There remains a significant difference in the standing requirements: in a Lanham Act lawsuit, the plaintiff is typically a competitor; in putative consumer class action lawsuits, representatives must allege that they were personally deceived to have standing.[53] Despite these differences in standing, the laws have been interpreted relatively congruously. The mischief to be corrected, in both cases, is false advertising and unfair competition. However, there is no requirement that each state’s consumer deception law track the Lanham Act or federal case law.
Part II: The GNC Case and the Glucosamine-Chondroitin Problem
A. Factual Background
General Nutrition Corporation (GNC), a national nutritional products retailer, has manufactured and sold a line of joint health supplements[54] for years. These products, which list glucosamine, chondroitin, and various other compounds as the primary active ingredients, are marketed collectively under the “TriFlex” brand. Although the products differ slightly in their total combination of ingredients, they advertise similar claims[55]; essentially, the TriFlex brand improves the health, comfort, and function of joints. The label for one product, TriFlex Fast-Acting, included a ”[c]linically studied” establishment claim: a “12-week multi-center, randomized, double-blind, placebo controlled study of 60 adults [. . .] taking 250 mg/day of the GNC TriFlex Fast-Acting Blend” proved that the product was “shown to improve joint comfort and function,” in addition to promising 20% improvement in joint function and 25-30% improvement in joint flexibility.[56] GNC produced a similar line of products for Rite Aid, which claimed to “promote joint health” and “help[. . .] rebuild cartilage and lubricate joints.” In compliance with FDA guidelines, all of the TriFlex and Rite Aid products included the disclaimer: “This statement has not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease.”
Beginning in March 2013, class action lawsuits against GNC and Rite Aid popped up across the United States. Plaintiffs in California, Illinois, Florida, New York, New Jersey, Pennsylvania, and Ohio alleged violations of an array of state consumer protection, deceptive practices and/or express warranty statutes in regard to the glucosamine-chondroitin products.[57] In December 2013, the pending lawsuits were consolidated by the Judicial Panel on Multidistrict Litigation and, pursuant to 28 U.S.C. § 1407, transferred to the U.S. District Court for the District of Maryland.[58]
B. The Complaint
In the Consolidated Amended Complaint, Plaintiffs alleged that the “overwhelming weight of high quality, credible and reliable studies demonstrate that glucosamine and chondroitin…do not provide joint health benefits” (emphasis added).[59] They stated that the inefficacy of these supplements was “generally recognize[ed]” by the scientific community.[60] In support of their allegations, Plaintiffs cited to thirteen studies released between 2004 and 2013.[61] In the studies, researchers concluded that (1) glucosamine and chondroitin did not reduce symptoms for osteoarthritic users or chronic joint pain sufferers (who, Plaintiffs alleged, were an appropriate proxy for non-arthritic users), and (2) MSM, another compound found in the TriFlex products, did not provide pain or joint symptom relief for osteoarthritic consumers. Notably, the Plaintiffs did not provide any testing of GNC’s particular products or combination of ingredients. Instead, they relied on the “vast weight of competent and reliable scientific evidence” (i.e., the cited studies) to infer that the “ingredients in GNC’s TriFlex Products do not work as represented” and that the “representations [were] false.”[62]
Defendants GNC and Rite Aid filed a Motion to Dismiss,[63] arguing that the studies did not test the products (or specific combination of ingredients) at issue,[64] that no representations were made with regard to improving osteoarthritis symptoms, and that osteoarthritis patients were not an appropriate proxy for non-arthritic users, among other deficiencies. Regarding the osteoarthritis issue, Defendants specifically cited to the FDA disclaimer on all of the labels, which state that the products are “not intended to diagnose, treat, cure, or prevent any disease.”[65]
C. Decision Granting Motion to Dismiss
In spite of Plaintiffs’ allegations, District Court Judge J. Frederick Motz dismissed the Complaint with leave to amend. He found that Plaintiffs’ allegations regarding the “vast weight of scientific evidence” — in light of GNC’s study to the contrary — could not support the single conclusion that the claims are false.[66] Under the Twombly/Iqbal plausibility standard, there was a “fatal flaw” in the allegations of the CAC: it did not allege that “experts in the field” were prepared to testify that, on the basis of the existing scientific evidence, any reasonable expert would conclude from the cited studies that glucosamine and chondroitin are ineffective in non-arthritic consumers.[67] The “mere existence of a ‘battle of the experts’” was not proof of falsity, but was rather to be “expected.”[68] The following excerpt from Judge Motz’s Memorandum Decision provides an important glimpse into the Court’s reasoning:
“Disagreements between experts, even under the ‘reasonable degree of scientific certainty’ standard, are to be expected. In my judgment, however, the fact that one set of experts may disagree with the opinions expressed by other qualified experts does not ipso facto establish any violation of the applicable consumer protection laws. If there are experts who support what defendants say in their advertisements, the advertisements are not false and misleading . . . unless the clinical trial relied upon by defendants was itself false and/or deceptive.”[69]
In this passage, the Court referenced the “reasonable basis” standard that marketers must adhere to when preparing advertising claims (discussed later in Section III, supra). Stated briefly, if scientific evidence points in different directions, the reasonable basis standard will allow for inconsistent messaging. In a footnote to this excerpt, the Court addressed whether this is a burden of proof issue and concluded that it is not, based on the nature of the advertising claims in the case.[70] The Court contrasted the example to a product liability case, in which plaintiff must establish that a product is defective. There, Judge Motz said, it would be “entirely appropriate for a jury to decide the defect issue” based on expert testimony that, to a reasonable degree of scientific certainty, a product is defective.[71] This distinction is enormously significant for procedural and substantive reasons, reserving for juries their fact-finder role beyond false advertising cases.
By granting the Plaintiffs leave to amend their complaint, Judge Motz offered Plaintiffs an opportunity to revive their claim for literal falsity. He specifically instructed that Plaintiffs must plead that the clinical trial relied on by GNC (1) did not exist at all, (2) exists but did not support any of GNC’s representations about TriFlex, or (3) exists and supports the assertions on TriFlex Fast-Acting’s bottle, but was not conducted in an appropriately scientific manner.[72] Only under such a pleading could the Court infer, on the face of the complaint, that the products are ineffective as to non-arthritic users. Otherwise, if at least one expert supported what GNC and Rite Aid said in their ads, the advertisements could not be false.[73] Plaintiffs could only file an amended complaint if they could do so in accordance with Fed. R. Civ. P. 11, which requires sufficient due diligence to avoid sanctions.[74] Absent such a pleading, Plaintiffs’ claim of falsity would not be facially plausible.
Six days after the Motion to Dismiss was granted, GNC’s counsel sent plaintiffs a letter contending that “qualified experts” support the TriFlex and Rite Aid products’ claims.[75] Plaintiffs rejected the opportunity to amend, and instead filed a motion for reconsideration on the basis of the existing complaint.[76] Judge Motz denied the motion and reiterated the rationale presented in the Memorandum Decision. Regarding policy, he added one additional reason to his earlier holding: It would be unfair to consumers who wish to gamble on glucosamine and chondroitin if lay juries could effectively ban the sale (or artificially raise its pricing) simply because evidence of their effectiveness is inconclusive.[77] Plaintiffs appealed to the Fourth Circuit.
D. Plaintiffs’ Appeal
After oral argument before a three-judge panel, the Fourth Circuit unanimously affirmed the District Court’s decision in favor of GNC.[78] The Fourth Circuit inexplicably held that “plaintiffs must allege that all reasonable experts in the field agree that the representations are false.”[79] In other words, a single expert in disagreement bars a statement from being literally false. In the Court’s words:
“[I]n order to state a false advertising claim on a theory that representations have been proven false, plaintiffs must allege that all reasonable experts in the field agree that the representations are false. If plaintiffs cannot do so because the scientific evidence is equivocal, they have failed to plead that the representations based on this disputed scientific evidence are false.”[80]
Under the Fourth Circuit’s test, “the question of falsity hinges on the existence (or not) of scientific consensus.”[81] Scientific experts may — and often do — disagree about the truthfulness of a statement, but equivocalness is not falsity.[82] Regarding Plaintiffs’ argument that a “battle of the experts” could not be resolved on the pleadings, the Court retorted:
“When litigants concede that some reasonable and duly qualified scientific experts agree with a scientific proposition, they cannot also argue that the proposition is ‘literally false.’ Either the experts supporting the Companies are unreasonable and unqualified (in which case, there is no real battle of the experts to begin with) or they reflect a reasonable difference of scientific opinion (in which case the challenged representations cannot be said to be literally false).”[83]
Indeed, by characterizing the dispute as a “battle of the experts,” the Court held that Plaintiffs (inadvertently) conceded that “a reasonable difference of scientific opinion exists as to whether glucosamine and chondroitin can provide the advertised joint health benefits.”[84] The Fourth Circuit also responded to Plaintiffs’ concern that manufacturers might hide behind so-called experts in proffering dubious marketing claims, relying on the Federal Rules of Evidence to ensure relevant and reliable scientific testimony.[85]
Part III: Legal and Policy Considerations
A. Law Professors’ Amicus Brief, with Rebuttal
After the Circuit Court handed down its GNC decision, sixteen prominent law professors submitted an Amicus Brief for the Plaintiffs in support of rehearing.[86] The Amici describe their interest in the case as “promoting truth in advertising, which protects consumers and promotes fair competition.”[87] In the eight-page brief, the law professors[88] supported Plaintiffs’ petition for rehearing and criticized the Fourth Circuit’s decision. The Amici argued that the Fourth Circuit “erred when it disregarded binding precedent” to produce a ruling with adverse procedural and substantive consequences. They further argued that questions of falsity cannot be resolved on the pleadings; instead, the fact finder (presumably a jury) should evaluate competing experts and pools of scientific evidence to determine the truth of a claim.
The brief is problematic in numerous respects. Specifically, the Amici (i) misstate the legal standards in falsity cases, (ii) misstate the nature of “binding” precedent, (iii) mischaracterize the nature of scientific evidence, and (iv) fail to appropriately consider the impact of the suggested alternative on jury instructions. This Note argues that the Amicus Brief, as well as the alternative outcome that Amici support, is a less persuasive answer to the truth-in-advertising dilemma presented by the GNC case. Instead, the Amici’s criticism offers insight into the utility and strengths of the Fourth Circuit’s logic. This section presents the Amici’s arguments and responds to points addressed in their brief.
1. Inaccurate Legal Standards
The Amicus Brief begins with a “Summary of Argument” that describes the purported standard in literal falsity cases: “Literal falsity is about how an advertisement is received by consumers. The adjectives ‘literal,’ ‘explicit,’ and ‘implicit’ (and falsity ‘by necessary implication’) describe consumer reaction to a message, which is either proved by evidence such as surveys or presumed as a matter of law.”[89] The italicized portions of this excerpt are problematic and, pun intended, literally false. First, literal and explicit claims are not assessed on the basis of consumer perception; rather they are accepted as literal messages. Second, literal and explicit claims are evaluated based on underlying substantiation, such as scientific testing, regardless of the message received by consumers. These important doctrinal and evidentiary distinctions are reviewed in Section I and, briefly, below.
As mentioned in Part I, infra, two modes of advertising are liable under Section 43(a): advertising that is false, and advertising that is misleading. Within the umbrella of false advertising, courts have generally recognized two types of falsity: (i) advertising that is literally false (based on a clear and explicit misrepresentation of fact), and (ii) advertising that is false by necessary implication (based on a claim that, considered in context, necessarily and unambiguously implies a message). Neither case requires surveys or other proof of consumer deception. Amici’s emphasis on “consumer reaction to a message” and “surveys” is misplaced, as both of those issues are irrelevant to a finding of literal falsity.
Amici then describe an example of a potentially misleading representation: “Vitamin B7 can remedy hair loss.” Since implied messages can convey multiple messages to the buying public, evidence of consumer confusion is required to determine which message consumers are receiving. While a discussion of misleading advertising can be helpful for comparison purposes, it is not relevant in the GNC case. First, the TriFlex claims at issue are literal and explicit: for example, “promotes joint mobility and flexibility” and “protects joints from wear and tear of exercise.” Second, there was no survey evidence of the messaging received by consumers, which would be required to sustain a cause of action for misleading representations.[90]
There are subsequent discussions regarding advertising that is false by necessary implication.[91] While actionable, this mode of deception is also irrelevant to the GNC case. The claims at issue on the TriFlex label are clear and explicit. Advertisements that are false by necessary implication do not involve an explicit statement of fact, but rather an unambiguous assumption that follows from a claim. Consider a comparison to the claims at issue in DIRECTV v. Time Warner Cable, discussed in Part I.C., infra. In that case, the court considered other words and images within the context of the statement that “settling for cable would be illogical;” the necessary implication was that settling for cable would be illogical because DIRECTV offers superior picture quality (it did not). The clear and explicit TriFlex claims simply do not fit into this category of consumer deception.
2. Misstatement of Precedent
The Amicus Brief continues with a citation to “binding precedent,” namely C.B. Fleet Co. v. SmithKline Beecham Consumer Healthcare, L.P., which the Fourth Circuit courts allegedly “disregarded” in dismissing the GNC case. In the opening paragraph of the Amicus Brief’s “Argument,” Amici state: “This Court has repeatedly held that both inquiries involve questions of fact. C.B. Fleet Co. v. SmithKline Beecham Consumer Healthcare, L.P., 131 F.3d 430, 434 (4th Cir. 1997) (“Whether an advertisement is literally false is an issue of fact.”).” This is both a mischaracterization of the holding as well as the case’s precedential value.
Amici appropriately classify C.B. Fleet as a literal falsity case.[92] At issue were “improved design” and “comparative superiority” claims for SmithKline’s new model of a feminine hygiene product. The Amici presumably provide this quotation in support of the alternative outcome they propose: factual disputes, such as a battle of the experts, should be reserved for juries. However, this proposition is conceptually distinct from the “issue of fact” identified by the C.B. Fleet Court. In C.B. Fleet, the Fourth Circuit was reviewing the district court’s determinations of falsity on appeal; the quote provided by Amici stands for the proposition that district courts, in their fact-finding role, are given deference in reaching such conclusions.[93] Furthermore, there is a citation following the C.B. Fleet reference — omitted without acknowledgment in the Amicus Brief — to L&F Prods. v. Procter & Gamble Co., 45 F.3d 709, 712 (2d Cir. 1995). That underlying authority from L&F Products states as follows: “The district court’s determination with respect to facial falsity is a finding of fact which we review for clear error.”[94] In both C.B. Fleet and L&F Products, the “issue of fact” is raised with respect to the deference due to district courts, rather than juries’ roles in resolving factual disputes.
In fact, the definition of falsity in C.B. Fleet does not conflict with the determination made in GNC. However, there remains a crucial factual difference between the cases. As the GNC Court noted, plaintiffs were given leave to amend the Consolidated Amended Complaint to plead that GNC’s study (1) did not exist at all, (2) existed but did not support any of GNC’s representations about TriFlex, or (3) existed and supported the assertions on TriFlex Fast-Acting’s bottle, but was not conducted in an appropriately scientific manner. The Court specifically instructed that plaintiffs could not rely on the “vast weight of the evidence” to prove falsity. In C.B. Fleet, plaintiff met this burden, challenging the scientific reliability of SmithKline’s testing procedures. In GNC, by contrast, plaintiffs twice relied on the “weight of the evidence” without attacking GNC’s testing – even after a second bite at the apple.
Although the Amici state that the Fourth Circuit has “repeatedly held that [the falsity inquiry] involve[s] questions of fact,” only C.B. Fleet is offered as support. One might reasonably expect a string cite supporting such an assertion. While the Fourth Circuit cemented a new rule for the meaning of literal falsity in supplement cases, its definition is hardly inconsistent with “binding” and “disregarded” precedent. In fact, the issue presented in GNC was a matter of first impression before the Court. While Amici’s proposal is a permissible solution to the GNC question, it is neither required by precedent nor, as this Note argues, able to achieve a better outcome on policy grounds. The Fourth Circuit’s ruling is a novel but befitting solution to truth-in-advertising questions in the dietary supplement industry.
3. The Nature of Scientific Evidence
The most persuasive portion of the Amicus Brief comes in the form of a hypothetical regarding the medical cause of ulcers. Amici present the following scenario: In 1910, a doctor would have said that stress and diet caused stomach ulcers, and that bacteria did not. We now know that this is untrue; the bacterium H. pylori — not stress and diet — causes many stomach ulcers. [95] The Amici use this illustration to argue that, despite expert support for a claim, it may be false. The fear is that corporate defendants may insulate themselves from lawsuits by hiding behind one unreliable study. This illustration lends support to Amici’s claim that experts are not infallible, and perhaps that juries serve as a valuable check on expert elitism.
Understood another way, the ulcer hypothesis could be marshaled in support of the Fourth Circuit’s ruling. Indeed, scientific knowledge is always a work in progress, and our understanding of medical issues is constantly evolving in the face of new research and methodology. Consider an alternative analysis of the ulcer illustration. In 1910, the majority of doctors, based on our understanding of ulcers at the time, would have said that the claim that bacteria cause ulcers is literally false. In the decades that followed, medical explanation for ulcers changed in light of evolving research. By 1950, the weight of scientific evidence would indicate that bacteria do, in fact, cause ulcers. Today, it is scientific truth that H. pylori is the culprit behind stomach ulcers — a proposition that would have been literally false in 1910 if the “vast weight of scientific evidence” set the standard in falsity cases.
The point is that the minority opinion in a scientific debate may turn out to be correct. The GNC holding corrects for the mistaken understanding that a minority position is wrong simply because it is not supported by a majority of scientists at that time. In a 1910 jury trial, the jury would likely — and incorrectly — have discounted the minority position because it would not meet the preponderance of the evidence standard. Then, the question becomes, why should we let a jury decide such matters? In the ulcer case, how and why would a jury have reached a better result?
The Fourth Circuit recognized a key difference between claims that, in light of the existing scientific evidence, are inherently false, as opposed to claims that are reasonably debatable. The GNC standard is most receptive to the changing nature of scientific evidence and techniques. Until consensus among duly qualified and reasonable experts emerges around the veracity of a claim, it would simply be premature to rule out the minority opinion as false. The ulcer anecdote, perhaps inadvertently, stands for this proposition.
Amici would likely retort that the GNC rule too easily insulates corporate advertisers, hiding behind a single study, in literal falsity cases. However, as the GNC Court noted, plaintiffs remain protected due to the rules of evidence and by alternative causes of action. As Judge Floyd explained, “Plaintiffs remain protected from dubious experts by the Federal Rules of Evidence, which ‘ensure that any and all scientific testimony…is not only relevant, but reliable.’” The importance of relevant and reliable scientific testing cannot be overstated, since plaintiff must prove that no reasonable scientific expert could find merit in the advertiser’s claims. The Fourth Circuit is careful to rule out quackery as a source of substantiation. If no reasonable study exists in support of a proposition, then the plaintiff meets their burden of proof in literal falsity cases. Additionally, Judge Floyd noted that “[p]laintiffs who cannot meet the burden of proving literal falsity may avail themselves of a claim regarding misleading representations.” This is helpful to plaintiffs who cannot or do not challenge the scientific legitimacy of a claim, but consider the messaging to consumers to be deceptive. Plaintiffs may then rely on survey evidence to establish consumer deception or confusion regarding an ad claim.
4. Jury Instructions
Finally, a crucial issue is not addressed in the Amicus Brief, but is relevant to the alternative outcome that Amici support: if the GNC case were assigned to a jury, what would the jury instructions look like? Juries undeniably serve an important fact-finding role in courtrooms across the United States, but play no part in the GNC story. Amici aptly draw attention to this procedural predicament, but serious questions remain if a panel of jurors would have reached a better outcome in the glucosamine case.
GNC is procedurally significant because the Court ruled, on a motion to dismiss, that a literal falsity case could be resolved on the pleadings. The Fourth Circuit articulated a very specific pleading standard with respect to plausibility in falsity cases. To recall, future plaintiffs must allege that no reasonable study (or scientist) exists or supports the challenged advertising claims; reliance on the so-called “vast weight of evidence” is insufficient to survive a motion to dismiss. In so ruling, the Fourth Circuit held that a jury would not resolve conflicting disagreements among experts.
One could say, as Plaintiffs argued, that a battle of opinions among qualified and competent experts creates a genuine issue of material fact, and that juries can handle highly technical scientific evidence. The involvement of juries seems even more urgent when considering the resource limitations of the FTC and FDA. GNC’s resolution at the plausibility stage usurped the jury of its central province,[96] and corporate defendants can more easily evade liability from the FTC, FDA, and courts alike.
While plaintiffs’ argument prevails in principle, it has less purchase in practice. When there is reasonable evidence on both sides of a scientific controversy, why should six jurors decide if a product cannot be sold or advertised? Reserving judgment for juries beyond the pleadings stage would overestimate juries’ abilities to resolve highly technical scientific controversies. However, if plaintiffs plead issues related to the credibility or reliability of the defendant’s studies (which the GNC plaintiffs did not), a jury’s intervention seems far more befitting. In such case, if the evidence on one side is unreliable, there is no “battle of the experts” to begin with, and we can trust juries to side with the (only) party bearing substantiation. GNC preserves this crucial distinction beyond the pleadings stage.
Lastly, the GNC holding should be cabined to the dietary supplement area. Judges, academics (including Amici), and others would likely agree with this assertion. The GNC ruling does not rob the jury of its fact-finder role in cases involving product liability, wrongful death, and related matters. The District Court specifically addressed this concern, distinguishing the question in GNC from other burden of proof issues. In the words of Judge Motz:
“I have considered whether the issue is one of burden of proof, and I have concluded that it is not. Rather, the basis of my holding lies in the nature of the claims asserted by plaintiffs that rely upon the falsity, deceptiveness, or unfairness of defendants’ advertisements. In contrast, for example, in a product liability case in which a plaintiff must establish that a product is defective, it would be entirely appropriate for the jury to decide the defect on the basis of expert testimony that, to a reasonable degree of scientific certainty, a product is defective.”[97]
As is evident in the Court’s language, the decision is meant to be narrowly construed, and this message appears to have been received by other courts. As of February 2018, the GNC holding has not been applied outside of advertising cases.
B. Additional Merits of the Fourth Circuit’s Decision
There is useful criticism of the GNC ruling from Amici and others.[98] On the one hand, the GNC ruling seems like a massive victory for corporate defendants — and a massive loss for consumer plaintiffs — on both substantive and procedural grounds. GNC arguably imposes a high hurdle on consumer plaintiffs in literal falsity cases; without alleging perfect scientific consensus or attacking defendant’s study, plaintiff’s case can no longer survive a motion to dismiss. But this is only part of the story.
Prior to the GNC decision, the question of what constituted false advertising during a period of scientific controversy was unanswered by the case law. Despite the prevailing criticism, the Fourth Circuit’s position is a persuasive solution to a previously unresolved legal dilemma. The arguments that follow address the policy merits of the GNC standard for literal falsity in supplement cases. Specifically, this Section argues that (i) the Fourth Circuit’s standard fits neatly into the patchwork of pre-existing consumer protection measures, namely those of the FTC and FDA, (ii) the decision promotes fair competition and preserves consumer choice, and (iii) the res judicata effect in these cases is limited, preserving opportunities for future plaintiffs to take a bite at the proverbial apple as scientific knowledge evolves.
1. FTC, FDA, and Fourth Circuit Parity
The falsity standard articulated in GNC is consistent with other consumer protection measures, namely, the FTC and FDA’s approaches to truth-in-advertising. The Fourth Circuit’s holding reflects the policy rationales underlying the FTC’s prior substantiation doctrine and the FDA’s disclosure regime. Both agencies, which are most directly involved in regulating consumer communications about supplements, would not take issue with the TriFlex ads.
Under the FTC’s advertising standards, the TriFlex claims are truthfully advertised. The FTC’s prior substantiation rule, discussed in Section I.B., states that an advertiser must have a “reasonable basis” for making objective claims and must “possess substantiation, prior to running the ad, for affirmative product claims.”[99] GNC cited to a particular study in its possession on the bottle of one of the TriFlex products. Plaintiffs nowhere alleged that GNC did not have a reasonable basis for making the joint health improvement claims; they failed to allege this deficiency even after the District Court granted them leave to amend their complaint. While the FTC has investigated other glucosamine-chondroitin products, the author has no information indicating that the FTC acted (or would need to act) with respect to the TriFlex claims.
GNC was also compliant with the FDA’s supplement labeling requirements, since each TriFlex label included the ingredients in each bottle, the standard disclaimer, and other required information. In broader perspective, the Fourth Circuit’s ruling is consistent with the FDA’s position on supplement labeling and advertising. Consider the evolution of the FDA’s approach to regulating the supplement industry. The 1906 Federal Food and Drugs Act used the label as a tool to empower consumers to have a much better conception of what they were putting into their bodies. In the 1930s, the FDA confronted the question of allowable representations in the face of scientific uncertainty, further extending the “informative labeling” approach: representations could be made, even if not reflective of the consensus of scientific opinion, as long as there was reliable scientific basis behind them. In such cases, the statements would have to be qualified with a disclaimer that reliable scientific opinion differed on such claims.[100] The goal was to preserve freedom of consumer choice, with the agency’s primary concern being the safety of the products. The Fourth Circuit’s ruling follows the hands-off approach that the FDA has taken towards safe and properly labeled dietary supplements.
Sources of truth-in-advertising law are too numerous to recount here, but the Fourth Circuit’s position achieves congruity across axes. Both the District Court and the Fourth Circuit interpreted the state consumer protection laws similarly; the fact that the laws come from different states is a distinction without a difference.[101] The overall objective of these statutes remains the same: protecting consumers from false and misleading advertising. It seems reasonable that a federal court seeking guidance on false advertising issues would look to the Lanham Act on regulating advertising and consumer communications. In view of these considerations, an ad that is false under the Lanham Act should likewise be deemed false under a state consumer deception statute. Furthermore, both industry and consumers benefit from consistent interpretations of state consumer protection statutes, which provide adequate notice to both parties about pleading requirements and possible defenses.
2. Fair Competition and Consumer Choice
Over the last decade, dozens of false advertising lawsuits have been brought against purveyors of glucosamine-chondroitin. The claims typically reflect the conventional wisdom about improving joint health and flexibility, just as GNC’s TriFlex did. These cases across the United States are undoubtedly expensive for advertisers to defend. Ultimately, the price is borne by the consumer who continues to “gamble” on glucosamine because, in her subjective opinion, it works. Both GNC decisions consider this economic dimension, limiting litigation beyond the motion to dismiss stage for cases in which there are two legitimate medical understandings of a supplement’s efficacy. Furthermore, from a policy perspective, it seems unadvisable to “put out of business” an approach for which there is a sound medical view.
As Judge Motz aptly noted in his Memorandum Decision, “It is unfair to consumers who wish to gamble that glucosamine and chondroitin may be effective if lay juries can effectively ban the sale of glucosamine and chondroitin simply because the evidence of their effectiveness is inconclusive.” Satisfied consumers are free to purchase the products, and dissatisfied consumers are, of course, free to seek alternatives. Unlike drugs, which operate under completely different advertising standards, consumers are not required to take supplements. The perspective of jurors adds no value to this highly individual decision. The Court continued: “After all, damage awards and even the cost of defending against high stakes litigation has the effect of increasing the cost of glucosamine/chondroitin pills or, potentially, driving the pills from the market. Should those who choose to purchase the pills have to pay more for them (or be deprived of the opportunity to purchase them at all) when the science is uncertain merely because juries disagree with their own judgment about the pills’ efficacy?”[102]
In addition to preserving consumer choice, the GNC ruling respects that whether the supplements work is an enormously subjective inquiry. In another case involving glucosamine-chondroitin products, the court neatly summarized the dilemma:
“The health and comfort of joints is probably influenced by a number of variables. Did [plaintiff] keep all of them constant, adjust for ones that can’t be kept constant (like aging), and then somehow have her cartilage and joints examined? Did she keep precise records of how much [glucosamine-chondroitin] she took, why she took it, and just how long she took it for? Can she document what her physical condition was before and after she took [glucosamine-chondroitin]? Probably not. What’s more likely is that she took [glucosamine-chondroitin] casually and just didn’t feel much better, but that makes her own claims just as speculative as she alleges [the supplement’s] benefits are.”[103]
We could, as Amici argue, send these cases to juries and let them decide if the minority view does not hold water. But should we? The Court asked and answered this very question, stating: juries serve as “a proper check on expert elitism. However, in this case the question is not whether the views of jurors should prevail over the views of asserted experts and judges. Rather the question is whether the views of jurors should prevail over the views of those who choose to purchase glucosamine/chondroitin pills. What is ‘democratic’ in one instance may be tyrannical in another. . . .”[104]
Finally, there is a hidden narrative beneath these lawsuits. At first glance, it may seem that consumers are fired up about wasting money on useless products. However, a closer look at the court dockets suggests an alternative story: many of the glucosamine-chondroitin lawsuits seem to be brought by the same lawyers, not consumers. Bonnett, Fairbourn and Denlea & Carton have been heavily involved in various glucosamine-chondroitin product lawsuits. The results have been lucrative for law firms. In the Plaintiffs’ Memorandum supporting its Motion for Reconsideration, counsel noted “at least three nationwide settlements against manufacturers of similar joint relief supplements are in various stages of review.”[105] Should law firms be able to line their pockets from consumers’ purses? The FDA, FTC, state attorneys general, and corporate competitors are better safeguards of truth-in-advertising than plaintiffs’ firms would care to admit.
3. Res Judicata Effects
Critics of the GNC ruling may rest assured that claim preclusion in false advertising cases is limited to the named plaintiff(s). The res judicata effect does not bind anyone to the result aside from the person or persons who brought the suit. If a party sued GNC for its TriFlex claims in Texas today, nothing is settled, and the court is well within its discretion to hear the case. It is particularly important to preserve these opportunities for litigation as scientific knowledge evolves and the conventional wisdom shifts.
On the other hand, the limits of the res judicata effects in these cases reveal a troubling pattern. As mentioned previously, there have been dozens of lawsuits across the United States involving glucosamine-chondroitin supplements, many of which were brought within the last ten years. These cases, which have not utilized the GNC pleading standard, have resulted in disparate verdicts and settlements. For example, the glucosamine-chondroitin cases in California have come out all over the place: it is entirely possible that a Los Angeles jury would find the claims to be truthful, and a jury in San Francisco would deem them falsely advertised. As a result, glucosamine products may bear dissimilar advertising claims depending solely on the location where the product is purchased. This result seems arbitrary and premature, given the conflicting scientific evidence on the claims at issue.
Conclusion
During periods of scientific controversy, how should advertisers, regulators, and courts address truth-in-advertising dilemmas? As the GNC case has shown, this is a complex and interesting issue that is worth examining. The Fourth Circuit’s answer to this question is a fair and reasonable standard, relying on expert consensus to establish literal falsity. While respected academics and courts have greeted the Fourth Circuit’s ruling with caution, advertisers and consumers may be more satisfied with its outcome. Corporate defendants are once again reminded of the prior substantiation requirement in articulating advertising claims and can, with assurance, hang their hats on reasonable scientific testing. Consumers enjoy the continued availability of supplements that, in their subjective opinions, provide symptom relief – without paying for the extra costs associated with consumer class actions. Consumer plaintiffs are also given clear guidance on surviving a motion to dismiss, as the Fourth Circuit neatly articulated the pleading requirements in alleging literal falsity.
The novelty and significance of the GNC decision are only beginning to be understood, but it offers compelling policy advantages to proposed alternatives. The Fourth Circuit’s ruling fits into the preexisting patchwork of consumer protection measures, announcing similar principles to the FTC’s substantiation theory and the FDA’s disclosure theory of liability. The solution is well suited to the nature of scientific evidence, which is continually changing in response to new research and technology. It is hard to argue that, if qualified scientists disagree about the efficacy of a dietary supplement, lay juries would achieve a better outcome in determining truth from falsity. Elevating the opinions of six jurors over experts and consumers does not provide any more certainty in arriving at the truthful result.
Questions remain about what impact, if any, the GNC decision should have on false advertising law. The author hopes that this unanimous decision by a panel of respected appellate judges will receive thoughtful consideration. In broader perspective, the standard does no damage to the existing falsity doctrines, and can be carefully confined to the dietary supplement space. The new standard articulated in GNC may seem like a big pill to swallow, but it is an effective remedy to the truth-in-advertising question that plagued the dietary supplementary industry.
[1] National Institutes of Health, Dietary Supplements: What You Need To Know, U.S. Dep’t of Health & Hum. Servs. (June 17, 2011), https://ods.od.nih.gov/HealthInformation/DS_WhatYouNeedToKnow.aspx.
[2] See National Institutes of Health, Multivitamin/mineral Supplements: Fact Sheet for Health Professionals, U.S. Dep’t of Health & Hum. Servs. (July 8, 2015), https://ods.od.nih.gov/factsheets/MVMS-HealthProfessional/ (“In 2014, sales of all dietary supplements in the United States totaled an estimated $36.7 billion.”).
[3] See Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505 (4th Cir. 2015), reh’g denied Ct. Order Den. Mot. for Reh’g and Reh’g En Banc (July 27, 2015), ECF No. 47.
[4] There is a well-publicized association between glucosamine-chondroitin and joint health. An article on WebMD states that the natural glucosamine in our bodies “helps keep up the health of your cartilage – the rubbery tissue that cushions bones at your joints.” As we age, levels of the natural compound begin to drop, which leads to the gradual breakdown of the joint. WebMD reports that there is “some evidence” that glucosamine sulfate supplements help counteract this problem, though experts “aren’t sure how.” There are a plethora of glucosamine supplements advertising joint health benefits, such as Osteo Bi-flex and Nature’s Bounty glucosamine-chondroitin compound. The widespread use of glucosamine-chondroitin has even reached our pets, with supermarket giant Trader Joe’s and others proffering a line of the supplement for dogs. Is Glucosamine Good for Joint Pain?, WebMD (Jan. 17, 2018), https://www.webmd.com/vitamins-and-supplements/supplement-guide-glucosamine.
[5] The glucosamine-chondroitin lawsuits are too numerous to list here, but are discussed later in Section III.B. See, e.g., Lerma v. Schiff Nutrition Int’l, Inc., No. 11-cv-1056 (S.D. Cal. filed May 13, 2011); Padilla v. Costco Wholesale Corp., No. 11-C-7686 (N.D. Ill. filed Oct. 28, 2011); Quinn v. Walgreen Co., No. 12-cv-8187 (S.D.N.Y. filed Nov. 9, 2012). According to TruthInAdvertising.org, at least nine class action lawsuits had been filed by October 2013 claiming that companies were falsely marketing the health benefits of glucosamine supplements. Consumers Claim This Joint Is Not Jumping, Truth in Advertising Organization (Oct. 22, 2013), https://www.truthinadvertising.org/consumers-claim-joint-jumping/. Some of the lawsuits have resulted in settlements. See, e.g., McCrary v. The Elations Company, LLC, No. 13-cv-00242 (C.D. Cal. filed Feb. 07, 2013); Pearson v. NBTY, Inc., No. 1:11-cv-07972 (N.D. Ill. filed Nov. 09, 2011). Additionally, there is no indication that these lawsuits will stop being filed. As of December 2017, glucosamine-chondroitin is still a hot topic in false advertising. According to a consumer class action blog, plaintiffs’ firms have commenced investigations into Osteo Bi-flex, Schiff Move Free and Glucosamine, Walmart’s Spring Valley Glucosamine, and Nature Made Triple Flex, among others. Tracy Colman, Different Brands of Glucosamine Chondroitin May Be Falsely Advertised, Top Class Actions (Dec. 21, 2017), https://topclassactions.com/lawsuit-settlements/lawsuit-news/828289-different-brands-of-glucosamine-chondroitin-may-be-falsely-advertised/.
[6] It is the author’s suggestion that the facts of these cases offer support for the latter. Section III.B. offers commentary on this aspect of the glucosamine lawsuits.
[7] Procedurally, these cases are as complex and interesting as the ruling itself. Ultimately, according to the Fourth Circuit’s order, there were five GNC plaintiffs (Howard, Toback, Lerma, Calvert, and Gaatz) and three Rite Aid plaintiffs (Flowers, George, and Gross). GNC, 789 F.3d at 509 n.1. The underlying lawsuits were consolidated in the U.S. District Court for the District of Maryland. The consolidated cases are docketed as: Howard v. GNC Corp., No. 14-cv-00002 (D. Md. filed Jan. 3, 2014); Toback v. GNC Holdings, Inc., No. 14-cv-00122 (D. Md. filed Jan. 14, 2014); Lerma v. GNC Corp., No. 14-cv-00120 (D. Md. filed Apr. 18, 2013); Calvert v. GNC Corp., No. 14-cv-00123 (D. Md. filed Jan. 16, 2014). The Rite Aid product case was Flowers v. Rite Aid, No. 14-cv-00465 (D. Md. filed Feb. 18, 2014). Flowers was effectively a lawsuit against GNC, as the products at issue were manufactured for Rite Aid by GNC, and GNC was contractually obligated to indemnify Rite Aid for the claims at issue in the litigation. Consolidated Amended Complaint at ∂ 2, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No.20. To the best of the author’s knowledge, GNC did not contest this assertion. There are multiple related lawsuits that were resolved separately, such as Galvin v. GNC, No. 14-cv-00810 (D. Md. filed Mar. 21, 2014). Although Brown v. GNC Corp., No. 13-05890 (N.D. Cal. filed Dec. 19, 2013), was also transferred to the district court by the Multidistrict Litigation Panel, the Consolidated Amended Complaint does not include Yvonne Brown (the plaintiff in that action) among the named plaintiffs. GNC, 789 F.3d at 509 n.2. Galvin was voluntarily dismissed in October 2015, and therefore did not proceed with the other consolidated cases.
[8] Warren Teichner & Megan Lesko, Cashing in on the booming market for dietary supplements, McKinsey & Company Marketing & Sales Insights (Dec. 2013), https://www.mckinsey.com/business-functions/marketing-and-sales/our-insights/cashing-in-on-the-booming-market-for-dietary-supplements.
[9] Council for Responsible Nutrition, The Dietary Supplement Consumer: 2015 CRN Consumer Survey on Dietary Supplements, CRN USA (2015), http://www.crnusa.org/CRNconsumersurvey/2015/.
[10] Country Report: Vitamins in the US, Euromonitor International (Nov. 2017), http://www.euromonitor.com/vitamins-and-dietary-supplements-in-the-us/report. This insight into the brutally competitive landscape offers some explanation for GNC’s insistence on keeping its joint health tag line. Marketing and advertising claims, such as “promotes joint health,” are valued shortcuts for consumers as they navigate supplement shelves. The alternative is a sea of supplements that are labeled and identified solely by their active ingredient(s), like a bottle that simply reads “Vitamin D” or “Glucosamine-Chondroitin” on the front, without further description of health benefits. A consumer who is unfamiliar with glucosamine-chondroitin may not buy the product, were it not for the “promotes joint health” byline. The explanation provides quick information and incentive to buy.
[12] FDA 101: Dietary Supplements, U.S. Food & Drug Admin. (Nov. 6, 2017), http://www.fda.gov/ForConsumers/ConsumerUpdates/ucm050803.htm.
[13] Id.
[15] For more information about the FDA and FTC’s roles in policing supplement labeling and advertising, see Section I.B., infra.
[16] Kenneth A. Plevan, Recent Trends in the Use of Surveys in Advertising and Consumer Deception Disputes, 15 Chi.-Kent J. Intell. Prop. 49, 61 (2016) (commenting on the “recent explosion of consumer deception lawsuits brought as putative class actions, filed by private plaintiffs under state consumer protection laws . . .”); see also Theodore V.H. Mayer, Recent Developments and Current Trends in United States Class Action Law, 826 PLI/Lit 313, 325 (May 24, 2010) (citing Federal Judicial Center, The Impact of the Class Action Fairness Act of 2005 on the Federal Courts 4 (4th interim report, 2008)) (“Among the most remarkable trends from the period between 2001 and 2007 [was] . . . the continuing growth of consumer-protection or consumer-fraud class actions, which increased by more than 150 percent and now account for 20 percent of all federal class actions.”) Private plaintiffs have increasingly availed themselves of consumer protection statutes. According to one study of over 17,000 reported federal district and state appellate decisions, “[b]etween 2000 and 2007 the number of [consumer protection act] decisions reported in federal district and state appellate courts increased by 119%. This large increase in CPA litigation far exceeds increases in tort litigation as well as overall litigation during the same period.” See Searle Civil Justice Institute, State Consumer Protection Acts: An Empirical Investigation of Private Litigation Preliminary Report (Dec. 2009). An additional concern is the “double litigation” frontier: “There is nothing to prevent a private litigant from filing suit against a consumer product advertiser or manufacturer after a federal regulatory agency, such as the Federal Trade Commission, takes action against the same company and obtains full monetary redress for consumers.” Dana Rosenfeld & Daniel Blynn, The “Prior Substantiation” Doctrine: An Important Check On the Piggyback Class Action, 26 Antitrust 1, 68 (Fall 2011). Rosenfeld and Blynn state that there is an emerging trend of plaintiffs filing class action complaints that are “virtually identical to or rely heavily upon” FTC complaints or FDA warning letters. Id.
[17] Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505, 515 (4th Cir. 2015).
[18] Plaintiffs may seek redress from the Federal Trade Commission under the FTC Act, 15 U.S.C. §§ 41-58. Barton Beebe, Trademark Law: An Open-Source Casebook, Part IV, 2 (Jul. 20, 2017), http://tmcasebook.org/wp-content/uploads/2017/08/BeebeTMLaw-4.0-Introduction.pdf.
[19] Lanham Act § 43(a)(1)(B) is codified at 15 U.S.C. § 1125(a)(1)(B).
[20] The state consumer deception statutes are also known as “little” or “baby” FTC Acts. See Beebe, supra note 18.
[21] Denise E. DeLorme, Jisu Huh, Leonard N. Reid & Soontae An, Dietary supplement advertising in the US: A review and research agenda, 31 Int’l J. of Advert. 547, 555 (2012).
[22] See Beebe, supra note 18. See also Joshua M. Dalton & Jared A. Craft, What You Should Know About NAD False Advertising Claims, Law360 (Jan. 4, 2013), https://www.law360.com/articles/403099/what-you-should-know-about-nad-false-advertising-claims.
[23] Note that the FDA and FTC are agencies of different stature. Unlike the FTC, the FDA is not an independent entity of the U.S. government. The FDA is nestled under the Department of Health and Human Services and funded by the Department of Agriculture.
[24] Advertising Dietary Supplements, Consumer Healthcare Products Association, http://www.chpa.org/DS_Advertising.aspx (last visited Mar. 23, 2017).
[25] Rebecca Tushnet, Advertising and Marketing Law 1289 (3rd ed. 2014).
[26] U.S. Food & Drug Admin. Guidance for Industry: A Food Labeling Guide, (revised Jan. 2013), https://www.fda.gov/Food/GuidanceRegulation/GuidanceDocumentsRegulatoryInformation/LabelingNutrition/ucm2006828.htm.
[27] Mark Nichter & J.J. Thompson, For my wellness, not just my illness: North Americans’ use of dietary supplements, 30 Culture, Med. & Psychiatry 175, 176 (2006). Unlike drugs, supplements can be readily purchased without a prescription in a wide variety of brick-and-mortar stores and online.
[28] 15 U.S.C. § 45 (2012).
[29] Pfizer, Inc., 81 F.T.C. 23 (1972).
[30] Lesley Fair, Federal Trade Commission Advertising Enforcement, Fed. Trade Comm’n (revised Mar. 1, 2008), https://www.ftc.gov/sites/default/files/attachments/training-materials/enforcement.pdf; see, e.g., Brake Guard Products, Inc., 125 F.T.C. 138 (1998); ABS Tech Sciences, Inc., 126 F.T.C. 229 (1998).
[31] See, e.g., Removatron International Corp., 111 F.T.C. 206 (1988), aff’d 884 F.2d 1489 (1st Cir. 1989) (“adequate and well-controlled…clinical testing” is required to substantiate claims for hair removal product); Schering Corp., 118 F.T.C. 1030 (1994) (consent order) (tests and studies relied upon as “reasonable basis” must employ appropriate methodology and address specific claims made in the advertisement).
[32] Dietary Supplement Health and Education Act of 1994, Pub. L. No. 103-417, 108 Stat. 4325 (codified in scattered sections of 21 U.S.C.).
[33] Questions and Answers on Dietary Supplements, U.S. Food & Drug Admin., https://www.fda.gov/Food/DietarySupplements/UsingDietarySupplements/ucm480069.html (last updated Nov. 29, 2017).
[34] In its current form, section 43(a)(1)(B) states: (1) Any person who, on or in connection with any goods or services, or any container for goods, uses in commerce any word, term, name, symbol, or device, or any combination thereof, or any false designation of origin, false or misleading description of fact, or false or misleading representation of fact, which — . . . (B) in commercial advertising or promotion, misrepresents the nature, characteristics, qualities, or geographic origin of his or her or another person’s goods, services, or commercial activities, shall be liable in a civil action by any person who believes that he or she is or is likely to be damaged by such act. 15 U.S.C. § 1125(a)(1)(B) (2018).
[36] 134 S. Ct. 1377, 1395 (2014).
[37] Id. at 1389; see also J. Thomas McCarthy, McCarthy on Trademarks and Unfair Competition § 27.30 (5th ed. 2017).
[38] See Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505, 514 (4th Cir. 2015) (noting that, although consumers cannot invoke the protections of the Lanham Act, the “considerable body of federal common law construing the Act is instructive in construing the state laws at issue here”).
[39] Puffery is a safe harbor for advertisers who proffer exaggerated and unsupported (or, perhaps, unsupportable) claims. It is a non-actionable carve-out from the false advertising provisions of the Lanham Act, since the statements are typically so exaggerated that no reasonable consumer could be deemed to rely on them. To fall out of the test for false advertising, statements of puffery are not considered to be statements of fact; consequently, a plaintiff would fail to meet the first element of the test. Courts have different definitions of puffery. In Time Warner Cable, Inc. v. DIRECTV, Inc., the Second Circuit described one party’s internet ads as “inaccurate descriptions” of the television service, but “so grossly distorted and exaggerated that no reasonable buyer would take them to be accurate depictions.” 497 F.3d 144, 159 (2d Cir. 2007). In United Industries Corp. v. Clorox Co., “[p]uffery is ‘exaggerated advertising, blustering, and boasting upon which no reasonable buyer would rely and is not actionable under § 43(a).’” 140 F.3d 1175, 1180 (8th Cir.1998) (quoting Southland Sod Farms v. Stover Seed Co., 108 F.3d 1134, 1145 (9th Cir. 1997)). In other iterations, the puffery doctrine has protected purveyors of “The Best Beer in America,” In re Bos. Beer Co., 198 F.3d 1370 (Fed. Cir. 1999), “The Most Advanced Home Gaming System in the Universe,” Atari Corp. v. 3D0 Co., No. C 94-20298 RMW (EAI), 1994 U.S. Dist. LEXIS 8677 (N.D. Cal. May 16, 1994), and “Better Ingredients. Better Pizza,” Pizza Hut, Inc. v. Papa John’s Int’l, Inc., 227 F.3d 489 (5th Cir. 2005).
[40] Under National Basketball Ass’n v. Motorola, Inc., the test for materiality is whether the statement “misrepresent[s] an inherent quality or characteristic of a product.” 105 F.3d 841, 855 (2d Cir. 1997).
[41] Schick Mfg., Inc. v. Gillette Co., 372 F.Supp.2d 273, 276 (D. Conn. 2005).
[42] See Tushnet, supra note 25, at 259; see also Coca-Cola Co. v. Tropicana Prods., Inc., 690 F.2d 312, 317 (2d Cir. 1982) (“[T]he Court may grant relief without reference to the advertisement’s [actual] impact on the buying public.”), abrogated on other grounds by Fed. R. Civ. P. 52(a)(6), as recognized in Johnson & Johnson v. GAC Int’l, Inc., 862 F.2d 975, 979 (2d Cir. 1988).
[43] Consumer surveys are not valid supporting evidence in a falsity case, discussed further in Section I.B.2. The GNC case fits neatly within the question of literally false advertising, as the TriFlex ads and packaging presented explicit, unambiguous statements, and no evidence of consumer deception was presented. See generally GNC, 789 F.3d 505.
[44] See generally Coca-Cola, 690 F.2d 312.
[45] See Time Warner Cable, Inc. v. DIRECTV, Inc., 497 F.3d 144, 158 (2d Cir. 2007).
[46] Id. at 152.
[47] See id. at 158. Note that the Second Circuit ruled on numerous commercials and Internet ads in this case, and some of the district court’s opinion was reversed.
[48] Where an advertisement is literally true but misleading, the advertisement “has left an impression on the listener that conflicts with reality[;]” with proof of consumer confusion, the representations are considered misleading. See id. at 153.
[49] 661 F.2d 272 (2d Cir. 1981). See also 2-7 Gilson on Trademarks § 7.02 (2017).
[50] Carolyn L. Carter, Consumer Protection in the States: A 50-State Report on UDAP Statutes, 5 (Feb. 2009), http://www.nclc.org/images/pdf/udap/report_50_states.pdf.
[51] 2-7 Gilson on Trademarks § 7.02 (2017).
[53] See Kenneth A. Plevan & Angela Colt, Consumer Surveys: Certification, Bloomberg BNA Prod. Safety & Liab. Rep. 4 (Sept. 12, 2016).
[54] See Consolidated Amended Complaint at ∂ 26-35, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014) ECF No.20. For reference, the four GNC products at issue, according to the Plaintiffs, were TriFlex, TriFlex Sport, TriFlex Fast-Acting, and TriFlex Complete Vitapak. The six Rite Aid products at issue are Rite Aid Glucosamine/Chondroitin, Rite Aid Natural Glucosamine/Chondroitin, Rite Aid Glucosamine Chondroitin Advanced Complex, Rite Aid Glucosamine Chondroitin, Triple Strength + MSM, Rite Aid Glucosamine Chondroitin + MSM, and Rite Aid Glucosamine Chondroitin Advanced Complex with HA. The court does not distinguish between the GNC and Rite Aid brands, nor does it distinguish between the individual products at issue.
[55] See Consolidated Amended Complaint at ∂ 26-37, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. Jun. 20, 2014) ECF No.20; see also In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014). The claims on the labels, as presented by Plaintiffs and GNC, are as follows. The GNC TriFlex Dietary Supplement label contains “Maximum strength now with hyaluronic acid” and “Promotes joint mobility and flexibility.” The GNC TriFlex Sport label contains: “Protects joints from wear and tear of exercise,” “Maximum strength joint comfort for active individuals,” and “Clinical strength for daily long-term use.” TriFlex Fast-Acting label contains: “Now with a joint cushioning blend including hyaluronic acid and vitamin C,” “Maximum strength, fast-acting support – works in days,” and “Clinical strength for daily long-term use.” Lastly, the TriFlex Complete Vitapak label contains: “Maximum strength, fast-acting joint comfort – works in days,” “Rebuilds cartilage and lubricates joints,” and “Supports natural anti-inflammatory response.” The Rite Aid Glucosamine/Chondroitin Dietary Supplement label contains “helps rebuild cartilage and lubricate joints.” Each label contains an FDA disclaimer: “This statement has” or “these statements have” “not been evaluated by the Food and Drug Administration. This product is not intended to diagnose, treat, cure, or prevent any disease.” It appears that the only product to include a “here’s proof” claim is TriFlex Fast-Acting. There is a reference to a scientific study that supports GNC’s representations: “Clinically studied doses of glucosamine and chondroitin combined with MSM and a proprietary herbal blend, which is shown to improve joint comfort and function. In a 12-week multi-center, randomized, double blind, placebo controlled study of 60 adults, subjects taking 250 mg/day of the GNC TriFlexTM Fast-Acting Blend showed statistically significant improvements in measures of joint function and joint flexibility within 30 days compared to subjects on placebo.”
[56] See Defendant’s Memorandum in Support of its Motion to Dismiss Class Action Complaint, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014) ECF No.38; see also Rebecca Tushnet, Fourth Circuit Destroys Literal Falsity, Rebecca Tushnet’s 43(B)log (June 30, 2015), http://tushnet.blogspot.com/2015/06/fourth-circuit-destroys-literal-falsity.html. The GNC study was not published or otherwise publicly available, and there is currently no law requiring such disclosure.
[57] Some of the state consumer protection laws at issue include the California False Advertising Law, § 17500, et seq. (“FAL”), the California Unfair Competition Law, Cal. Bus. & Prof. Code § 17200, et seq. (“UCL”), the Florida Deceptive and Unfair Trade Practices Act, Fla. Stat. §§ 501.201, et seq. (“FDUTPA”), Illinois Consumer Fraud and Deceptive Business Practice Act, 815 Ill. Comp. Stat. 502/1, et seq. (“ICFA”), the New York Consumer Protection From Deceptive Acts and Practices Law, N.Y. Gen. Bus. Law § 349, et seq. (“NYGBL”), Ohio Rev. Code Ann. § 1302.26, the New Jersey Consumer Fraud Act, N.J.S.A. 56:8-1, et seq. (“NJCFA”), Pennsylvania Unfair Trade Practices and Consumer Protection Law, 73 Pa. Stat. Ann. §§ 201-1, et seq. (“PUTPCPL”).
[58] In re GNC Corp. TriFlex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-120, 2014 U.S. Dist. LEXIS 84184 at *1 (D. Md. June 20, 2014).
[59] Consolidated Amended Complaint at ∂ 38, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014) ECF No.20.
[60] Id.
[61] The studies can be found in the Consolidated Amended Complaint at ∂∂ 39-48, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014) ECF No.20.
[62] Consolidated Amended Complaint at ∂ 32, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014) ECF No.20.
[63] Since this case is considered procedurally significant, it is important to say a word about the legal standard on a motion to dismiss. A Rule 12(b)(6) motion to dismiss challenges the legal sufficiency of a complaint. Fed. R. Civ. P. 12(b)(6); see also Presley v. City of Charlottesville, 464 F.3d 480, 483 (4th Cir. 2006). The court has drawn a line between the “mere possibility” and “plausibility” that a defendant has acted unlawfully; plaintiff must meet the latter to survive a motion to dismiss. Ashcroft v. Iqbal, 556 U.S. 662, 678 (2009). The issue is whether plaintiff has stated enough factual matter, rising above the speculative level, to warrant a claim for relief. To defeat a 12(b)(6) motion, a complaint’s factual allegations must be sufficient to “raise a right to relief above the speculative level.” Bell Atl. Corp. v. Twombly, 550 U.S. 544, 555 (2007). The complaint must contain “sufficient factual matter, accepted as true, to state a claim for relief that is plausible on its face.” Ashcroft, 556 U.S. at 678. The “mere possibility of misconduct” is insufficient to avoid dismissal. Id. at 679. Likewise, “[l]abels and conclusions” and “naked assertion devoid of further factual enhancement” will fail to show that plaintiff is entitled to relief. Id. at 678. Ultimately, Plaintiffs’ bare assertions were insufficient to establish that the representations are false.
[64] Notably, similar and even predicate cases challenging the efficacy of glucosamine-chondroitin had been dismissed on similar grounds. See, e.g., Toback v. GNC Holdings, Inc., No. 13-80526, 2013 U.S. Dist. LEXIS 131135, at *16 (S.D. Fla. Sept. 13, 2013) (“Plaintiff’s allegations regarding the inefficacy of glucosamine and chondroitin simply fail to address the efficacy of the TriFlex Vitapak’s multifarious composition in promoting joint health . . .”); Eckler v. Wal-Mart Stores, Inc., No. 12-727, 2012 WL 5382218, at *6 (S.D. Cal. 2012) (plaintiff’s studies did not assess the “overall formulation that’s behind the representations at issue,” and so “the Court would be left with no facts from which to infer that [defendant] is liable for false advertising.”).
[65] Defendants’ Memorandum in Support of their Motion to Dismiss Consolidated Amended Complaint at 5, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No.25.
[66] The Consolidated Amended Complaint repeatedly described the TriFlex advertising as “false, misleading, and reasonably likely to deceive the public.” Consolidated Amended Complaint, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No.20. However, to plead that an advertisement is misleading, the allegation must be supported by evidence of consumer confusion. Since Plaintiffs did not provide any evidence of consumer confusion, the Court appropriately limited its analysis to a claim of literal falsity.
[67] Memorandum at 7, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 38.
[68] Id.
[69] Id. (emphasis added).
[70] Id. at n.2.
[71] See id.; see also Memorandum at 4 n.4, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 51. The significance of this carve-out for advertising cases is discussed further in Part III, supra.
[72] Memorandum at 7, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 38. An obvious issue, then, becomes whether Plaintiffs could so allege without appropriate discovery. Judge Motz addressed this question in his second Memorandum Decision, denying the motion for reconsideration: “[I]f plaintiffs can specify discovery requests that would aid them in alleging the above facts, they should file a motion setting forth the discovery that they request. Presumably, however, if plaintiffs’ experts are of the view that no reasonable expert would reach the conclusion reached by the expert upon whom defendant relies, they are already, by virtue of their asserted expertise, in possession of the relevant factual information.” Memorandum at 5, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 51.
[73] Memorandum at 4, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 51.
[74] Memorandum at 8, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 38.
[75] Exhibit 1 to Plaintiffs’ Memorandum of Law in Support of Motion to Correct Mistake of Law Pursuant to F.R.C.P. 60, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 44-1.
[76] See Plaintiffs’ Motion to Correct Mistake of Law Pursuant to F.R.C.P. Rule 60, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 43; Plaintiffs’ Memorandum of Law in Support of Motion to Correct Mistake of Law Pursuant to F.R.C.P. 60, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 44.
[77] Memorandum at 4, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 51; see also Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505, 512 (4th Cir. 2015).
[78] There are some important differences between the District Court’s Memorandum Decision and the Fourth Circuit’s Order. First, the Court included alternative grounds for affirming the District Court, noting that Plaintiffs “failed to allege that all of the purportedly active ingredients in each product are ineffective at promoting joint comfort, health, and flexibility.” Second, the Fourth Circuit disagreed with Judge Motz that specific formulations of the GNC and Rite Aid products needed to be tested to assess the truth of the labels’ representations; instead, at the motion to dismiss stage, the scientific studies in the CAC could render the claim facially plausible. Lastly, the court found that the factual dispute regarding the effectiveness of glucosamine-chondroitin for non-arthritic users was not appropriate for resolution on a motion to dismiss. The Fourth Circuit declined to adopt the latter grounds for affirming the District Court’s Order. See Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505 (4th Cir. 2015).
[79] See GNC, 789 F.3d at 516.
[80] Id.
[81] Id. at 514 n.7.
[82] See generally id. at 515-16.
[83] Id. at 515.
[84] Id.
[85] Id.
[86] See Brief of Law Professors as Amici Curiae in Support of Plaintiffs-Appellants’ Petition for Rehearing and for Rehearing En Banc, and in the Alternative, for Modification of Opinion and Judgment, Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505 (4th Cir. 2015) (No. 14-1724) ECF No.45.
[87] Id. at 1.
[88] Professor Rebecca Tushnet, whose advertising textbook is extensively cited in this Note, appears to have been the lead academic on the Amicus Brief. It is co-signed with Brian Wolfman, then a Visiting Professor at Stanford Law School. Other Amici include Mark Lemley of Stanford Law School, Jessica Litman of the University of Michigan Law School, and Barton Beebe of New York University School of Law, who supervised this Note’s completion. Notably, Professor Eric Goldman of Santa Clara University School of Law is not a signatory to the Amicus Brief, although he co-authored the seminal textbook on advertising law, Advertising & Marketing Law, with Professor Tushnet.
[89] Brief of Law Professors as Amici Curiae in Support of Plaintiffs-Appellants’ Petition for Rehearing and for Rehearing En Banc, and in the Alternative, for Modification of Opinion and Judgment at 1-2, Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505 (4th Cir. 2015) (No. 14-1724) ECF No.45. The brief is inexplicably wrought with references to consumer messaging in literal falsity cases. As Amici later state, “To determine if an ad makes a false claim, a court must determine what message consumers will perceive . . . .”
[90] Perhaps there is some suggestion that Plaintiffs’ should have argued that the joint health claims were misleading, rather than literally false. The argument would appear to be that glucosamine-chondroitin is particularly attractive to osteoarthritis patients, who seek to alleviate the pain of their condition. Plaintiffs offered numerous studies that allegedly prove that glucosamine-chondroitin is not beneficial to osteoarthritis sufferers. On that basis, and with an allegation of consumer confusion, the claims could have been deemed misleading to a segment of consumers. However, a claim of misleading advertising would, too, seem to fail, as the District Court noted that “the TriFlex labels expressly disclaim any ability to ‘diagnose, treat, cure, or prevent any disease.” (citations omitted). In re GNC Corp. TriFlex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 at *3-4 (D. Md. June 20, 2014).
[91] Brief of Law Professors as Amici Curiae in Support of Plaintiffs-Appellants’ Petition for Rehearing and for Rehearing En Banc, and in the Alternative, for Modification of Opinion and Judgment at 5, Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505 (4th Cir. 2015) (No. 14-1724) ECF No.45.
[92] Id at 2.
[93] In the discussion that follows that quote, the C.B. Fleet Court states: “Fleet challenges the district court’s determinations of no literal falsity on the grounds that the court imposed upon Fleet a wrong– overly stringent– ‘burden of proof’ on the issue, and, relatedly, that both of the ultimate fact findings were clearly erroneous[;]” “[W]e think that whether an advertising claim implicitly, though not expressly, asserts that it is test-validated must be considered a question of fact whose resolution is subject to clearly erroneous review[;]” “Our standard of review of district court fact findings is greatly deferential under controlling authority[]” (emphasis added). C.B. Fleet Co. v. SmithKline Beecham Consumer Healthcare, L.P., 131 F.3d 430, 434-436 (4th Cir. 1997).
[94] L&F Prods. v. Procter & Gamble Co., 45 F.3d 709, 712 (2d Cir. 1995).
[95] Brief of Law Professors as Amici Curiae in Support of Plaintiffs-Appellants’ Petition for Rehearing and for Rehearing En Banc, and in the Alternative, for Modification of Opinion and Judgment at 3, Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505 (4th Cir. 2015) (No. 14-1724) ECF No.45.
[96] Plaintiffs’ Reply Memorandum of Law in Further Support of Motion to Correct Mistake of Law Pursuant to F.R.C.P. 60 at 4, In re GNC Corp. TriFlex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014) ECF No.50.
[97] Defendant’s Memorandum in Support of its Motion to Dismiss Class Action Complaint at n.2, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014) ECF No.38.
[98] See generally Brief of Law Professors as Amici Curiae in Support of Plaintiffs-Appellants’ Petition for Rehearing and for Rehearing En Banc, and in the Alternative, for Modification of Opinion and Judgment, Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505 (4th Cir. 2015) (No. 14-1724) ECF No.45; Olamide Orebamjo, Comment, Brown v. GNC Corp.: The Fourth Circuit’s New Standard for Literal Falsity, 12 J. Bus. & Tech. L. Proxy 1 (2017).
[99] Pfizer, 81 F.T.C. at 29 (1972).
[100] John P. Swann, The history of efforts to regulate dietary supplements in the USA, 8 Drug Testing & Analysis 271, 275 (2016).
[101] Brown v. GNC Corp. (In re GNC Corp.), 789 F.3d 505, 514 (4th Cir. 2015) (“In construing the diverse state statutes at issue here, we apply this broadly shared understanding of the difference between false and misleading representations.”).
[102] See Memorandum at 4 n.4, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 51.
[103] Eckler v. Wal-Mart Stores, Inc., No. 12-727, 2012 WL 5382218, at *8 n.2 (S.D. Cal. 2012.
[104] Memorandum at 4 n.4, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 51.
[105] Plaintiffs’ Memorandum of Law in Support of Motion to Correct Mistake of Law Pursuant to F.R.C.P. 60, In re GNC Corp. Triflex Prods. Mktg. & Sales Practices Litig. (No. II), No. 14-2491, 2014 U.S. Dist. LEXIS 84184 (D. Md. June 20, 2014), ECF No. 44.
Think Big! The Need for Patent Rights in the Era of Big Data and Machine Learning
By: Hyunjong Ryan Jin*
Download a PDF version of this article here
Introduction
With AlphaGo’s triumph over the 9-dan Go professional Lee Sedol in March 2016, Google’s DeepMind team conquered the last remaining milestone in board game artificial intelligence.[1] Just nineteen years after IBM Deep Blue’s victory over the Russian chess grandmaster Garry Kasparov,[2] Google’s success exceeded expert predictions by decades.[3]
AlphaGo demonstrated how machine learning algorithms could enable processing of vast amounts of data. Played out on a 19 by 19 grid, the number of possible configurations on a Go board is astronomical.[4] With near-infinite number of potential moves, conventional brute-force comparison of all possible outcomes is not feasible.[5] To compete with professional level human Go players, the gaming artificial intelligence requires a more sophisticated approach than the algorithms employed for chess — machine learning. The underlying science and implementation of machine learning was described in a Nature article two months prior to AlphaGo’s match with Lee. In the article, the Google team described how a method called “deep neural networks” decides between the insurmountable number of possible moves in Go.[6] The AlphaGo model was built by reinforcement learning from a database consisting of over thirty million moves of world-class Go players.[7] This allowed the algorithm to optimize the search space of potential moves, therefore reducing the required calculations to determine the next move.[8] In other words, the algorithm mimics human intuition based on the “experience” it gained from the database “fed” into the algorithm, which drastically increases computational efficiency by eliminating moves not worth subsequent consideration. This allows the algorithm to devote computational resources towards the outcomes of “worthwhile” moves.
The advent of such powerful analytical tools, capable of mimicking human intuition alongside massive computation power, opens endless possibilities—early stage cancer detection[9], accurate weather forecasting,[10] prediction of corporate bankruptcies,[11] natural event detection,[12] and even prediction of elections.[13] For information technology (“IT”) corporations, investment in such technology is no longer an option, but a necessity. The question that this Note addresses is whether the current state of intellectual property law is adequate to harness the societal benefits that we hope to enjoy through the advances in machine learning. In particular, are patents necessary in the age of big data? And if they are, how should we apply patent protection in the field of big data and machine learning?
Part I of this Note examines the need for intellectual property rights in machine learning and identifies the methods by which such protection may be achieved. The differences between trade secret, copyright and patent protection in software are discussed, followed by the scope of protection offered by each means. This background provides the basis to discuss the effectiveness of each method in the context of machine learning and big data innovations.
Part II discusses the basics of the underlying engineering principle of machine learning and demonstrates how the different types of intellectual property protection may apply. Innovators may protect their contributions in machine learning by defending three areas—(1) the vast amount of data required to train the machine learning algorithm, (2) innovations in the algorithms itself including advanced mathematical models and faster computational methods, and (3) the resulting machine learning model and the output data sets. Likewise, there are three distinct methods of protecting these intellectual properties: patents, copyright, and secrecy.[14] This Note discusses the effectiveness of each method of intellectual property protection with three principles of machine learning innovation in mind: facilitating data sharing, avoiding barriers to entry from data network effects, and providing incentives to address the key technological challenges of machine learning. This Notes proposes that patents on computational methods adequately balance the concern of patent monopoly and promoting innovation, hence should be the primary means of intellectual property protection in machine learning.
Part III then visits the legal doctrine of patentable subject matter starting with the United States Supreme Court’s Alice decision. While Alice imposed a high bar for software patents, the post-Alice Federal Circuit decisions such as Enfish, Bascom, and McRO suggest that certain types of software inventions are still patentable. Specifically, this section will discuss the modern framework pertinent to subject matter analysis: (1) inventions that are directed to improvements of computer functionality rather than an abstract idea, (2) inventions that contain an inventive concept, and (3) inventions that do not improperly preempt other solutions. The Note will apply this framework to innovations in machine learning.
The Note proposes that patents for computational methods balance the need for intellectual property protection while permitting data sharing, paving the pathway for promoting innovation in machine learning. The Note further argues that machine learning algorithms are within patentable subject matter under 35 U.S.C. §101.
I. Need for Intellectual Property Rights in Machine Learning
“He who receives an idea from me, receives instruction himself without lessening mine; as he who lights his taper at mine, receives light without darkening me.”
– Thomas Jefferson
“I’m going to destroy Android, because it’s a stolen product. I’m willing to go thermonuclear war on this. They are scared to death, because they know they are guilty.”
– Steve Jobs
The two quotes above demonstrate the conflicting views on protecting intangible ideas with intellectual property law. Thomas Jefferson implied that the free circulation of inventive ideas and thoughts would not dampen the progress of innovation nor disadvantage innovators. On the other hand, Steve Jobs exhibited fury over the similarity between the iOS and the Android OS. Why? Was it because his company was worse off due to the similarity between the two products? Would Apple have refrained from inventing the iPhone had it known others would enter the smartphone market?
This section discusses the motives behind the grant of intellectual property rights and whether such protection should be extended to machine learning innovations. Basics of patent law, copyright law, and trade secret are introduced to provide the analytical tools for subsequent discussion on which type of intellectual property protection best promotes the socially-beneficial effects of machine learning.
A. Do We Need Intellectual Property Rights for Machine Learning?
The primary objectives of intellectual property rights are to encourage innovation and to provide the public with the benefits of those innovations.[15] In the context of machine learning, it is not clear whether we need any additional incentives to promote participation in this field. Machine learning is already a “hot field,” with countless actors in industry and academia in active pursuit to keep pace.[16] Hence investment incentivizing may not be a valid justification for granting intellectual property rights in machine learning. Rather, such protection is crucial to promote competition and enhance public benefits.
The quality of inferences that may be drawn from a given data set increases exponentially as the aggregation diversifies, which is why cross-industry data aggregation will greatly enhance the societal impact of machine learning.[17] Companies will need to identify new data access points outside of their own fields to gain access to other data sets to further diversity their data. Yet the incentive structures of behemoth corporations may not be well-suited to identify and grow niche markets.[18] It would be up to the smaller, specialized entities to find the gaps that the larger corporations overlooked and provide specialized services addressing the needs of that market. Protective measures that assist newcomers to compete against resource-rich corporations may provide the essential tools for startups to enter such markets. Sufficient intellectual property protection may serve as leverage that startups may use to gain access to data sets in the hands of the Googles and Apples of the world, thus broadening the range of social benefits from machine learning.
B. The Basics of Patent Law
“To promote the progress of science and useful arts, by securing for limited times to authors and inventors the exclusive right to their respective writings and discoveries” – United States Constitution, Article I, § 8
The United States Constitution explicitly authorizes Congress to promote useful arts by granting inventors the exclusive rights of their discoveries. Such constitutional rights stems from two distinct bases — (1) a quid pro quo where the government issues a grant of monopoly in exchange for disclosure to society, and (2) property rights of the inventor. The purpose for such rights is explicitly stated in the Constitution—to promote new inventions. The goal is to prevent second arrivers who have not invested in the creation of the initial invention from producing competing products and services at a lower price, undercutting the innovator whose costs are higher for having invested to create the invention. As an incentive for innovators willing to invest in new, useful arts, the patent system provides the innovator rights to exclude others from practicing the invention. Another purpose of such rights is the concept of “ “mining rights.” Akin to the grant of mining rights to the owner in efforts to suppress aggressive mining, the inventor should have the right to define and develop a given field by excluding other people from the frontiers of that knowledge. Considering the importance of industry standards in modern electronics, such a purpose acknowledges the importance of early stage decisions that may define the trajectory of new technological advances.
C. The Thin Protection on Software Under Copyright Law
The Copyright Act defines a “computer program” as “a set of statements or instructions to be used directly or indirectly in a computer to bring about a certain result.”[19] Though it may be counterintuitive to grant copyright protection for “useful arts” covered by patents, Congress has explicitly mandated copyright protection for software.[20] However, as will be discussed below, copyright protection of software has been significantly limited due to case law.
Copyright protects against literal infringement of the text of the program. Source code, code lines that the programmers “author” via computer languages such as C++ and Python, is protected under copyright as literary work.[21] In Apple v. Franklin Corp., the Third Circuit Court of Appeals held that object code, which is the product of compiling the source code, is also considered a literary work.[22] Given that compiled code is a “translation” of the source code, this ruling seems to be an obvious extension of copyright protection. Removing the copyright distinction between source code and object code better reflects the nature of computer languages such as Perl, where the source code is not translated into object code but rather is directly fed into the computer for execution. However, the scope of protection on either type of code is very narrow. The copyright system protects the author against literal copying of code lines. This leaves open the opportunity for competitors to avoid infringement by implementing the same algorithm using different text.
Fortunately, in addition to protection against literal copying of code, copyright law may provide some protection of the structure and logical flow of a program. Equivalent to protecting the “plot” of a novel, the Second Circuit Court of Appeals ruled that certain elements of programming structure are considered an expression (copyrightable) rather than idea (not copyrightable), extending copyright protection to non-literal copying.[23] The Computer Associates International v. Altai court applied a three-step test to determine whether a computer program infringes other programs—(1) map levels of abstraction of the program; (2) filter out protectable expression from non-protectable ideas; and (3) compare which parts of the protected expression are also in the infringing program.[24]
The merger doctrine is applied to step two of the Altai test to limit what may be protected under copyright law. Under the merger doctrine, code implemented for efficiency reasons is considered as merged with the underlying idea, hence not copyrightable.[25] Since most algorithms are developed and implemented for efficiency concerns, the Altai framework may prevent significant aspects of software algorithms from receiving copyright protection. This means that for algorithms related to computational efficiency, patents may provide significantly more meaningful protection than copyright. The Federal Circuit, in the 2016 case McRO Inc. v. Bandai Namco Games America Inc., ruled that patent claims with “focus on a specific means or method that improves the relevant technology” may still be patentable.[26] Although preemption concerns may impede patentability, exemption of patent right by preemption is narrow compared to that of copyright by the merger doctrine.
Scène à faire doctrine establishes yet another limitation on copyright for computer programs. Aspects of the programs that have been dictated by external concerns such as memory limits, industry standards and other requirements are deemed as non-protectable elements.[27] For mobile application software, it is difficult to imagine programs that are not restricted by form factors such as mobile AP computation power, battery concerns, screen size, and RAM limitations. As for machine learning software, the algorithms determine the “worthiness” of computation paths based on conserving computational resources. The external factors that define the very nature and purpose of such machine learning algorithms may exempt them from copyright protection.
D. Comparing Trade Secret and Non-disclosures with Patents
The crucial distinction between trade secret and patent law is secrecy. While patent applicants are required to disclose novel ideas to the public in exchange for a government granted monopoly, trade secret requires owners to keep the information secret. Though trade secret protection prevents outsiders from acquiring the information by improper means, it does not protect the trade secret against independent development or even reverse engineering of the protected information. In trade secret doctrine, the existence of prior disclosed art is only relevant for discerning whether the know-how is generally known, a different and simpler analysis than the issue of novelty in patent law.[28]
The United States Supreme Court has specified in Kewanee Oil that all matters may be protected under trade secret law, regardless whether it may or may not be patented.[29] The Kewanee Oil court predicted that inventors would not resort to trade secret when offered a presumptively stronger protection by patent law:
"The possibility that an inventor who believes his invention meets the standards of patentability will sit back, rely on trade secret law, and after one year of use forfeit any right to patent protection, 35 U.S.C. § 102(b), is remote indeed."[30]
Trade secret is an adequate form of protection for innovators that are concerned with the limits of what may be patentable. The secrecy requirement of trade secret inherently provides protection that may potentially outlive any patent rights, provided a third party does not independently acquire the secret. This coincides with an interesting aspect of machine learning and big data—the need for massive amounts of data. Developers need data to “train” the algorithm, and increase the accuracy of the machine learning models. Companies that have already acquired massive amounts of data may opt to keep their data secret, treating the aggregated data as a trade secret.
In addition to the amount of amassed data, companies have all the more reason to keep their data secret if they have access to meaningful, normalized data. Even if a company amasses an enormous amount of data, the data sets may not be compatible with each other. Data gathered from one source may have different reference points or methodologies that are not immediately compatible with data from another source. This raises the concern of “cleaning” massive amounts of data.[31] Such concerns of data compatibility mean that parties with access to a single, homogenous source of high quality data enjoy a significant advantage over parties that need to pull data from multiple sources.
However, data secrecy may not be a suitable strategy for companies that are aiming for cross-industry data aggregation. Institutions such as Global Alliance for Genomics and Health are promoting data sharing between research participants. The Chinese e-commerce giant Alibaba announced a data sharing alliance with companies such as Louis Vuitton and Samsung to fight off counterfeit goods.[32] To facilitate the development of technology and to mitigate risks, various companies and research institutions across diverse fields are engaging in joint development efforts and alliances. Seeking protection under trade secret runs against this trend of engaging in effective cross-industry collaboration. Yet there are countervailing arguments that trade secret promotes disclosure by providing legal remedies that can replace the protection of secrets.[33] Parties can sidestep the limitations of trade secrets by sharing proprietary information under the protection of contract law. While data sharing practices may void trade secret protection, the nature of continued accumulation of data and carefully drafted contractual provisions may provide sufficient protection for the data owners.
II. Placing Machine Learning within Intellectual Property Law
“Learning is any process by which a system improves performance from experience.”
– Herbert Simon, Nobel Prize in Economics 1978.
The concept of machine learning relates to computer programs that have the capability to improve performance based on experience, with limited intervention of the programmer.[34] Machine learning models have the capability to automatically adapt and customize for individual users, discover new patterns and correlations from large databases, and automate tasks that require some intelligence by mimicking human intuition.[35] This section dissects the mechanics of machine learning to identify the aspects of machine learning innovations that are at issue as intellectual property.
A. Machine Learning Basics
Machine learning methods are divided into two different approaches—supervised machine learning and unsupervised machine learning. For supervised machine learning, models are typically established by applying “labeled” sets of data to a learning algorithm. Labeled data refers to data sets that have both relevant features and the target results that the programmer is interested in. For example, we may be interested in developing a machine learning model that classifies images with dogs in them. The data sets for supervised machine learning would indicate whether a given images has dogs or not. The learning process begins with the algorithm fitting trends found in the training data set into different types of models. The algorithm compares the prediction errors of the models by inputting the validation set data into each model, measuring their accuracy. This allows the algorithm to decide which of the various models is best suited as the resulting machine learning model. Finally, the machine learning model is then evaluated by assessing the accuracy of the predictive power of the model. The developed model is then applied to data without a correct answer to test the validity of the model. In unsupervised machine learning, the data sets are “unlabeled” data, which may not contain the result that the programmer is interested in. Returning to our dog image classification example, data sets for unsupervised machine learning will have pictures of various animals that are not labeled—the computer does not know which pictures are associated with dogs. The unsupervised machine learning algorithm develops a model that extracts common elements from the picture, teaching itself the set of features that makes the subject of the picture a dog. In essence, unsupervised machine learning uses data sets that do not have specific labels fed into the algorithm for the purpose of identifying common trends embedded in that data set.
The objective of developing such machine learning models varies. Sometimes the goal is to develop a prediction model that can forecast a variable from a data set. Classification, which assigns records to a predefined group, is also a key application of the algorithm. Clustering refers to splitting records into distinct groups based on the similarity within such group. Association learning identifies the relationship between features.
Figure 1. Overview of Machine
Learning Model Development Figure
1 illustrates the overall process of machine learning model development. The
learning process of machine learning algorithms begins with aggregation of
data. The data originates from an array of diverse sources ranging from user
input, sensor measurement, or monitoring of user behavior.[36] The data sets are then
preprocessed. The quality of data presents a challenge in improving machine
learning models—any data that has been manually entered contains the
possibility of error and bias.[37] Even if the data is collected
through automatic means, such as health monitoring systems or direct tracking
of user actions, the data sets require preprocessing to account for systematic
errors associated with the recording device or method.[38] This includes data skews due to
difference between individual sensors, errors in the recording or transmission
of data, and incorrect metadata about the sensor.[39] Simply put, the data sets may
have differing reference points, embedded biases, or differing formats. The
“cleaning” process accommodates for the data skews. The
objective of machine learning models is to identify and quantify “features”
from a given data set. The term “feature” refers to individually measurable
property of an observed variable.[40] From the outset, there may be an extensive list of features
that are present in a set of data. It would be computationally expensive to
define and quantify each feature, and then to identify the inter-feature
relationships, from massive amounts of data. Due to the high demand for the computational
power required for processing massive amounts of data, dedication of
computational resources to features that are outside the scope of the
designer’s interest would be a waste of such limited computational capacity.[41] The machine learning algorithm
reduces waste of computational resources by applying dimensionality reduction
to the pre-processed data sets.[42] The algorithm can identify an
optimal subset of features by reducing the dimension and the noise of the data
sets.[43] Dimensionality reduction allows
the machine learning model to achieve higher level of predictive accuracy,
increased speed of learning, and improves the simplicity and comprehensibility
of the results.[44] However, the reduction process
has limitations—reducing dimensionality inevitably imposes a limit on the
amount of insights and information that may be extracted from the data sets. If
the machine learning algorithm discerns a certain feature, the model would not
be able to draw inferences related to said feature. Following
dimensionality reduction, the machine learning algorithm attempts to fit the
data sets into preset models. Typically, three different types of data are fed
into the machine learning model—training set, validation set, and test set.[45] The machine learning algorithm
“trains” the model by fitting the training set data into various models to
evaluate the accuracy of each selection. Then the validation set is used to
estimate error rates of each model when applied to data outside the training
set that was used to develop each model. Through this process, the machine
learning algorithm selects the model that best describes the characteristics
and trends of the target features from the test and validation sets.[46]
The test set is then used to calculate the generalized prediction error,
which is reported to the end user for proper assessment of the predictive power
of the model.[47] Simply put, the training test
and validation set is used to develop and select a model that reflects the
trends of the given data set, and the test set is used to generate a report on
the accuracy of the selected model. The
crucial elements in developing a machine learning model are (1) training data,
(2) inventions related to the machine learning algorithm such as the method of
preprocessing the training data, the method of dimensional reduction, feature
extraction, and the method of model learning/testing, and (3) the machine
learning model and output data.[48] An ancillary element associated
with the three elements above is the human talent that is required to implement
such innovation.[49] Innovators in the field of machine learning may protect
their investments by protecting one or more of the elements listed above. The
difference between training data and output data, as well as the difference
between the machine learning algorithm and the machine learning model, are best
illustrated with an example. Let us assume a credit card company wants to use
machine learning to determine whether the company should grant a premium credit
card to a customer. Let us further assume that the company would prefer to
grant this card to customers that would be profitable to the company while
filtering out applicants that are likely to file for bankruptcy. Data sets
about prior applicant information would correspond to training data. The company would apply a mathematical method of
extracting insight about the correlation between features and the criteria that
the company wants to evaluate (e.g., profitable for the firm or likely to file
bankruptcy). The mathematical methods are referred as machine learning algorithms. The resulting mechanism, such as a
scoring system, that determines the eligibility of card membership is the machine learning model. The credit card
applicant’s personal data would be the input
data for the machine learning model, and the output data would include information such as expected
profitability of this applicant and likelihood of bankruptcy for this
applicant. Discussing
incentive structures and trends behind the machine learning industry is
essential in identifying adequate methods of intellectual property rights. The
current trends in the world of machine learning will predict what intellectual
property regime is most useful to companies to protect their work. The United
States has chronically struggled to maintain adequate supply of talent in the
high-tech industry, a deficit of talent that continues in the field of machine
learning.[50] From a report by the McKinsey
Global Institute, the United States’ demand for talent in deep learning “could
be 50 to 60 percent greater than its projected supply by 2018.”[51] Coupled with the dearth of
machine learning specialists, the short employment tenure of software companies
further complicates the search for talent. Software engineers from companies
such as Amazon and Google have reported an average employment tenure of one
year.[52] While some parts of the high
attrition rate may be attributed to cultural aspects of the so-called “Gen Y”
employees, the “hot” demand for programming talent has significant impact on
the short employee tenure.[53] Job mobility within the
software industry is likely to increase as the “talent war” for data scientist
intensifies. Employee mobility and California’s prohibition against “covenants
not to compete” have been accredited as a key factor behind the success of
Silicon Valley.[54] Another trend in the field is
the rapid advances in machine learning methods. Due to the fast-paced
development of the field, data scientists and practitioners have every reason
to work with companies that would allow them to work at the cutting edge of
machine learning, using the best data sets. This may influence the attrition
rates and recruiting practices of the software industry mentioned above.[55] Eagerness of employees to
publish scientific articles and contribute to the general machine learning
committee may be another factor of concern. To
accelerate innovation by repurposing big data for uses different from the
original purpose, and to form common standards for machine learning, more
industries are joining alliances and collaborations.[56] Cross-industry collaborations
may enable endless possibilities. Imagine the inferences that may be drawn by
applying machine learning methods to dietary data from home appliances,
biometric data, and data on the weather patterns around the user. Putting
privacy nightmares aside, machine learning with diverse data sets may unlock
applications that were not previously possible. More companies are attempting
to capitalize on commercial possibilities that data sharing may unlock.[57]
Though
it may seem intuitive that patent protection may be the best option,
innovations in machine learning may not need
patent protection. Trade secret protection on the data sets may be sufficient
to protect the interests of practicing entities while avoiding disclosure of
their inventions during the patent prosecution process. Furthermore, numerous
software patents have been challenged as unpatentable abstract subject matter
under 35 U.S.C. §101 since the Alice
decision in 2014.[58] Though subsequent decisions
provided guidelines for types of software patents that would survive the Alice decision, it is not clear how the
judiciary will view future machine learning patents. Such issues raise the
question about the patentability of machine learning – should we, and can we,
resort to patents to protect machine learning inventions? Following
the discussion on the building blocks of machine learning and recent emerging
trends in the field, this section discusses the mode and scope of protection
that current legal system provides for each element pertinent to innovation in
machine learning. The possible options for protecting innovations are (1)
non-disclosure agreements and trade secret law, (2) patent law, and (3)
copyright. The three options for protection may be applied to the three primary
areas of innovation—(1) training data, (2) inventions related to computation,
data processing, and machine learning algorithms, and (3) machine learning
models and output data. This discussion will provide context about the methods
of protection for innovations in machine learning by examining the costs and benefits
of the various approaches. Access
to massive amounts of training data is a prime asset for companies in the realm
of machine learning. The big data phenomenon, which triggered the surge of
interest in machine learning, is predicated on the need for practices to analyze
large data resources and the potential advantages from such analysis.[59] Lack of access to a critical
mass of training data prevents innovators from making effective use of machine
learning algorithms. Previous
studies suggest that companies resent sharing data with each other.[60] Michael Mattioli discusses the hurdles against sharing data
and considerations involved with reuse of data in his article Disclosing Big Data.[61] Indeed, there may be practical issues that prevent recipients of data from engaging in data
sharing. Technical challenges in comparing data from different sources, or
inherent biases embedded in data sets may be reasons that complicate receiving
outside data.[62] Mattioli also questions the
adequacy of the current patent and copyright system to promote data sharing and
data reuse—information providers may
prefer not to disclose any parts of their data due to the rather thin legal
protection for databases.[63] Perhaps
this is why secrecy seems to be the primary method of protecting data.[64] The difficulty of reverse
engineering to uncover the underlying data sets promotes the reliance on
non-disclosure.[65] Compared to the affirmative
steps required to maintain trade secret protection if the data is disclosed,
complete non-disclosure may be a cost effective method of protecting data.[66] Companies that must share data
with external entities may exhibit higher reliance on contract law rather than
trade secret law. In absence of contract provisions, it would be a challenge to
prove that the trade secret has been acquired by misappropriation of the
recipient party. The
“talent war” for data scientists may also motivate companies to keep the
training data sets secret. With a shortage of talent to implement machine
learning practices and rapid developments in the field, retaining talent is
another motivation for protecting against unrestricted access to massive
amounts of data. Companies may prefer exclusivity to the data sets that
programmers can work with — top talents in machine learning are lured to companies
with promises of exclusive opportunities to work with massive amounts of data.[67] The rapid pace of development
in this field encourages practitioners to seek opportunities that provide the
best resources to develop their skill sets. This approach is effective since a
key limitation against exploring new techniques in this field is the lack of
access to high quality big data. Overall, secrecy over training data fits well
with corporate recruiting strategies to retain the best talents in machine
learning. Non-disclosure
and trade secret protection seems to be the best mode of protection. First,
despite the additional legal requirements necessary to qualify as trade
secrets, trade secret protection fits very well with non-disclosure strategy.
On the other hand, patent law is at odds with the principle of non-disclosure.
While trade secret law provides companies protection without disclosing
information, patent law requires disclosure in exchange for monopolistic
rights. Furthermore, neither patent nor copyright provide adequate protection
for underlying data. Patent law rewards creative concepts and inventions, not
compiled facts themselves. Copyright may protect labeling or distinct ways of
compiling information, but does not protect underlying facts. Also, as a
practical matter, the difficulty of reverse engineering of machine learning
models does not lend well to detecting infringement. Analysis of whether two
parties used identical training data would not only be time consuming and
costly, but may be fundamentally impossible. If
companies were to seek protection of training data, it would be best to opt for
secrecy by non-disclosure. This would mean companies would opt out of the
cross-industry collaborations that were illustrated above. This may be less of
a concern for innovation, as companies may still exchange output data as means
of facilitating cross-industry collaboration. Adequate
protection over inventive approaches in processing data is becoming
increasingly important as various industries begin to adopt a collaborative
alliance approach in machine learning. Cross-industry collaboration requires
implementation of methods such as preprocessing diverse data sets for
compatibility. As the sheer amount of data increases, more processing power is
required. The machine learning algorithm needs to maintain a high degree of
dimensionality to accurately identify the correlations between a high number of
relevant features. The need for more innovative ideas to address such
technological roadblocks will only intensify as we seek more complex
applications for machine learning. The
three primary areas where novel ideas would facilitate innovations in machine
learning are pre-training data processing, dimensional reduction, and the
machine learning algorithm. Access
to massive amounts of data alone is not sufficient to sustain innovation in
machine learning. The raw data sets may not be compatible with each other,
requiring additional “cleaning” of data prior to machine learning training.[68] The data provided to the
machine learning algorithm dictates the result of the machine learning model,
hence innovations in methods to merge data with diverse formats is essential to
enhancing the accuracy of the models. As cross-industry data analysis becomes
more prominent, methods of merging data will have more significant impact on
advancing the field of machine learning than mere collection of large data
sets. Cross-industry data sharing would be useless unless such data sets are
merged in a comparable manner.[69] Companies can opt to protect their inventive methods by
resorting to trade secret law. The difficulty of reverse engineering machine
learning inventions, coupled with the difficulty of patenting software methods
provides incentives for innovators to keep such inventions secret from the
public. However, two factors would render reliance on non-disclosure and trade
secret ineffective—frequent turnover of software engineers and rapid speed of
development in the field. Rapid
dissemination of information from employment mobility may endanger intellectual
property protection based on secrecy. Furthermore, while the law will not
protect former employees that reveal trade secrets to their new employers, the
aforementioned fluid job market coupled with general dissemination of
information make it difficult to distinguish between trade secrets from former
employment and general knowledge learned through practice. The difficulties of
reverse engineering machine learning models work against the trade secret owner
as well in identifying trade secret misappropriation—how do you know others are
using your secret invention? The desire for software communities to discuss and
share recent developments in the field does not align well with the use of
secrecy against innovations in machine learning. Secrecy practices
disincentivize young data scientists from joining due to the limits against
gaining recognition.[70] The
rapid development of machine learning technology also presents challenges
against reliance on trade secret law.
Secret methods may be independently developed by other parties. Neither
trade secret law nor non-disclosure agreements protect against independent
development of the same underlying invention.[71] Unlike training data, machine
learning models, or the output data, there are no practical limitations that
impedes competitors from independently inventing new computational methods of
machine learning algorithms. With
such a fluid employment market, high degree of dissemination of expertise, and
rapid pace of development, patent protection may provide the assurance of
intellectual property protection for companies developing inventive methods in
machine learning. Discussions on overcoming the barriers of patenting software
will be presented in later sections.[72] The two
primary products from applying the machine learning algorithms to the training
data are the machine learning model and the accumulation of results produced by
inputting data into the machine learning model. The “input data” in this
context may refer to individual data that is analyzed by the insights gained
from the machine learning model. In a
recent article, Brenda Simon and Ted Sichelman discuss the concerns of granting
patent protection for “data-generating patents,” which refers to inventions
that generate valuable information in their operation or use.[73] Exclusivity based on patent protection may be extended
further by trade secret protection over the data that has been generated by the
patented invention.[74] Simon and Sichelman argue that
the extended monopoly over data may potentially overcompensate inventors since
the “additional protection was not contemplated by the patent system[.]”[75] Such expansive rights will
cause excessive negative impact on downstream innovation and impose exorbitant
deadweight losses.[76] The added protection over the
resulting data derails the policy rationale behind the quid pro quo exchange
between the patent holder and the public by excluding the patented information
from public domain beyond the patent expiration date.[77]
The
concerns addressed in data-generating patents also apply to machine learning
models and output data. Corporations may obtain patent protection over the
machine learning models. Akin to a preference for secrecy for training data,
non-disclosure would be the preferred mode of protection for the output data.
The combined effect of the two may lead to data network effects where users
have strong incentives to continue the use of a given service.[78] The companies that have
exclusive rights over the machine learning model and output data gather more
training data, increasing the accuracy of their machine learning products. The
reinforcement by monopoly over the means of generating data allows few companies
to have disproportionately strong dominance over their competitors.[79] Market
dominance by data-generating patents becomes particularly disturbing when the
patent on a machine learning model preempts other methods in the application of
interest. Trade secret law does not provide protection against independent
development. However, if there is only one specific method to obtain the best
output data, no other party would be able to create the output data
independently. The exclusive rights over the only methods of producing data
provides means for the patent holder to monopolize both the patent and the
output data.[80] From a policy perspective, the
excessive protection does seem troubling. Yet such draconian combinations are
less feasible after the recent rulings on patentable subject matter of
software, which will be discussed below.[81] Mathematical equations or
concepts are likely directed to an “abstract concept,” thus will be deemed
directed to a patent ineligible subject matter.[82] Furthermore, though recent
cases in the Federal Circuit have found precedents where software patents
passed the patentable subject matter requirement, those cases expressed
limitations against granting patents that would improperly preempt all
solutions to a particular problem.[83] The rapid pace of innovation in
the field of machine learning compared to the rather lengthy period required to
obtain patents may also dissuade companies from seeking patents. Overall,
companies have compelling incentives to rely on non-disclosure and trade
secrets to protect their machine learning models instead of seeking patents. The
secrecy concerns regarding training data applies to machine learning models and
the output data as well. Non-disclosure would be the preferred route of
obtaining protection over the two categories. However, use of non-disclosure or
trade secrets to protect machine learning models and output data presents
challenges that are not present in the protection of training data. The use of
secrecy to protect machine learning models or output data conflicts with
recruiting strategies to hire and retain top talent in the machine learning
field. The non-disclosure agreements limit the employee’s opportunity to gain
recognition in the greater machine learning community. In a rapidly developing
field where companies are having difficulty hiring talent, potential employees
would not look fondly on corporate practices that limit avenues of building a
reputation within the industry.[84] Companies have additional incentives to employ a rather lenient
secrecy policy for machine learning models and the output data. They have
incentives to try to build coalitions with other companies to monetize on the
results. Such cross-industry collaboration may be additional source of income
for those companies. The data and know-how that Twitter has about fraudulent
accounts within their network may aid financial institutions such as Chase with
novel means of preventing wire fraud. The reuse of insights harvested from the
large amount of raw training data can become a core product the companies would
want to commercialize. Data reuse may have an incredible impact even for
applications ancillary to the primary business of the company. Interesting
aspects of disclosing machine learning models and output data are the
difficulty of reverse engineering and consistent updates. If the company
already has sufficient protection over the training data and/or the
computational innovations, competitors will not be able to reverse engineer the
machine learning model from the output data. Even with the machine learning
model, competitors will not be able to provide updates or refinements to the
model without the computational techniques and the sufficient data for training
the machine learning algorithm. In certain cases, the result data becomes
training data for different applications, which raises concerns of competitors
using the result data to compete with the innovator. Yet the output data would
contain less features and insights compared to the raw training data that the innovator
possesses, and therefore would inherently be at a disadvantage when competing
in fields that the innovator has already amassed sufficient training data. Grant
of patents on machine learning models may incentivize companies to build an
excessive data network while preempting competitors from entering competition.
This may not be feasible in the future, as technological preemption is becoming
a factor of consideration in the patentable subject matter doctrine. Companies
may use secrecy as an alternative, yet may have less incentives to keep secrecy
compared to the protection of training data. The
current system, on its surface, does not provide adequate encouragement for
data sharing. If anything, companies have strong incentives to avoid disclosure
of their training data, machine learning model, and output data. Despite
these concerns, data reuse may enable social impacts and advances that would
not be otherwise possible. Previous studies have pointed out that one of the
major barriers preventing advances in machine learning is the lack of data
sharing between institutions and industries.[85] Data scientists have
demonstrated that they were able to predict flu trends with data extracted from
Twitter.[86] Foursquare’s location database
provides Uber with the requisite data to pinpoint the location of users based
on venue names instead of addresses.[87] Information about fraudulent
Twitter accounts may enable early detection of financial frauds.[88] The possibilities that
cross-industry data sharing may bring are endless. To
encourage free sharing of data, companies should have a reliable method of
protecting their investments in machine learning. At the same time, protection
based on non-disclosure of data would defeat of purpose of promoting data
sharing. Hence protection over computation methods involved with machine
learning maintains the delicate balance between promoting data sharing and
protecting innovation. Protection
over inventions in the machine learning algorithm provides one additional merit
other than allowing data sharing and avoiding the sort of excessive protection
that leads to a competitor-free road and data network effects. It incentivizes
innovators to focus on the core technological blocks to the advancement of
technology, and encourages disclosure of such know-how to the machine learning
community. Then what are the key obstacles in obtaining patents in
machine learning inventions? While there are arguments that the definiteness
requirement of patent law is the primary hurdle against patent protection of
machine learning models due to reliance on subjective judgment, there is no
evidence that the underlying inventions
driving big data faces the same challenge.[89] Definiteness may be overcome by
providing reasonable certainty for those skilled in the art of defining what
the scope of the invention is at the time of filing.[90] There is no inherent reason why
specific solutions for data cleaning, enhancement of computation efficiency,
and similar inventions would be deemed indefinite by nature. Since
the United States Supreme Court invalidated a patent on computer implemented financial
transaction methods in the 2014 Alice
decision, the validity of numerous software and business method patents were
challenged under 35 U.S.C. §101.[91] As of June 8th,
2016, federal district courts invalidated 163 of the 247 patents that were
considered under patentable subject matter—striking down 66% of challenged
patents.[92] The U.S. Court of Appeals for
the Federal Circuit invalidated 38 of the 40 cases it heard.[93] Arguably,
the public benefits more from such high rates of post-issuance invalidity. The
public still has access to the disclosures from the patents and patent
applications. In reliance on granted patents, companies may have already
invested in growing related businesses, catering to the need of consumers. At
the same time, the patent holder’s monopolistic rights have been shortened as
the result of litigation. Effectively, the price that the public pays to
inventors in exchange for the benefits of disclosure is reduced. Yet the
high degree of invalidity raises several concerns for the software industry.
Smaller entities, lacking market influence and capital, have difficulty
competing against established corporations without the monopolistic rights
granted through the patent system. Investors become hesitant to infuse capital
into startups for fear that invalidity decreases the worth of patents. Reliance
on trade secret has its own limitations due to the disclosure dilemma—the
inventor needs to disclose the secret to lure inventors, but risks losing
secrecy in the process. Copyright law does not provide appropriate protection.
The restrictions imposed by the merger doctrine and scène à
faire doctrine
constrain copyright protection of software. Though copyright provides an
alternative method of protecting literal copying of code, it does little to
protect the underlying software algorithms and innovation. Ultimately,
the increase of alliances and collaboration provides incentives for parties to
obtain patent rights. Reliance on trade secret or copyright are not suitable
methods of protecting their intellectual property. Furthermore, market power or
network effects alone cannot sufficiently mitigate the risks involved with
operating a business. Patents become even more important for startups since
patents provide investors with assurance that in the worst case, the patents
may still serve as potential collateral. Patentable subject matter continues to be a barrier for
patenting innovations in software. Additional doctrines such as enablement,
written description, and obviousness are also serious obstacles against
obtaining patents, yet such requirements are specific to each claimed invention
and the draftsmanship of claims. Subject matter is considered a broader, categorical
exclusion of patent rights. This section explores the current landscape of the
patentable subject matter doctrine in the software context. The
complexity involved with software, coupled with the relatively broad scope of
software patents, has presented challenges in identifying the boundaries of the
claims.[94] Many members of the software community detest imposing
restrictions on open source material and attest that many key innovations in
algorithms are rather abstract.[95] Such hostility against
patenting software has raised the question of whether patent rights should be
the proper method of protecting innovations in software. Alice was a case that embodied such
opposition to the grant of software patents. The case involved patents on
computerized methods for financial trading systems that reduce “settlement
risk” when only one party to financial exchange agreement satisfies its
obligation.[96] The method proposed the use of
a computer system as a third-party intermediary to facilitate the financial
obligations between parties.[97] The United States Supreme Court
ruled that the two-step test established from Mayo governed all patentable subject matter questions.[98] In particular, for the abstract
idea context, the Supreme Court established the following two-step framework
for patentable subject matter of software inventions: The Alice Court found that the patent on
financial transaction was “directed to a patent-ineligible concept: the
abstract idea of intermediated settlement,” and therefore failed step one.[101] Furthermore, the Court ruled
that the claims did “no more than simply instruct the practitioner to implement
the abstract idea of intermediated settlement on a generic computer” and did
not provide an inventive concept that was sufficient to pass step two.[102] The Alice framework was considered as a huge
setback for the application of patentable subject matter doctrine to software.
It was a broad, categorical exclusion of certain inventions that were deemed
“directed to” an abstract idea, natural phenomenon, or law of nature. The
biggest misfortune was the lack of guidance in the Alice decision on the threshold for such categorical exclusion—we
were left without any suggestions on the type of software patents that would be
deemed as patentable subject matter. The
recent line of cases in the Federal Circuit provides the software industry with
the much-needed clarification on the standards that govern patentability of
software inventions. Enfish v. Microsoft,
decided on March 2016, involved a “model of data for a computer database
explaining how the various elements of information are related to one another”
for computer databases.[103] In June 2016, the Federal
Circuit decided another case on the abstract idea category for patentable
subject matter. Bascom Global v. AT&T
Mobility is on a patent disclosing an internet content filtering system located
on a remote internet service provider (ISP) server.[104] Shortly after Bascom, the Federal Circuit decided McRO v. Bandai Namco Games in September
2016.[105] The case ruled that an
automated 3D animation algorithm that renders graphics in between two target facial
expressions is patentable subject matter.[106] The rulings from the Federal Circuit on the aforementioned
three cases provide guidelines along the two-step Alice test of patentable subject matter. The software patents in Enfish and McRO were deemed “directed to” a patent eligible subject matter,
informing the public of what may pass the first set of the Alice test. Bascom failed
the first step.[107] Yet the court ruled that those
patents had inventive concepts sufficient to transform a patent ineligible
subject matter into a patent eligible application. Combined together, the three
cases give more certainty in what may pass the 35 U.S.C. §101 patentable
subject matter inquiry. Reiterating
the Alice test, whether an invention
is a patentable subject matter is determined by a two-step process—(1) is the
invention directed to, rather
than an application of, an abstract idea, natural phenomenon, or law of nature,
and even if so, (2) do the elements of the claim, both individually and
combined, contain an inventive concept
that transforms this invention into a patent-eligible application? The Federal
Circuit fills in the gaps that were left unexplained from the Alice ruling. The Enfish court discussed what constitutes
an abstract idea at the first step of the Alice
inquiry. Judge Hughes instructs us to look at whether the claims are directed
to a specific improvement rather than an abstract idea. In this case, the
patent provides the public with a solution to an existing problem by a
specific, non-generic improvement to computer functionality. The Enfish court ruled that such invention
is patent eligible subject matter.[108] McRO also ruled that the facial
graphic rendering for 3D animation was not an abstract concept. Here, the
Federal Circuit again emphasized that a patent may pass step one of the Alice test if the claims of the patent
“focus on a specific means or method that improves the relevant technology.”[109] The McRO court also noted that preemption concerns may be an important
factor for the 35 U.S.C. §101 subject matter inquiry—that improper
monopolization of “the basic tools of scientific and technological work” is a
reason why such categorical carve outs against granting patents on abstract
ideas exist.[110] Bascom provides the standards on what
would fail step one of the Alice
patentable subject matter inquiry. If the patent covers a conventional, well-known
method in the field of interest, then the invention would be considered
abstract. This is akin to the inventive concept considerations conducted at the
second phase of the 35 U.S.C. §101 subject matter inquiry. The
main takeaway from Enfish and McRO is that in the first step of the Alice test, a patent application is not
directed to an abstract idea if (1) the invention addresses an existing problem
by specific improvements rather than by conventional, well-known methods and
(2) the claims do not raise preemption concerns. This encourages practitioners
to define the problem as broadly as possible, while defining the scope of
improvement in definite terms. The
second step of the Alice test is an
inquiry of whether the patent application, which is directed to a patent
ineligible subject, still contains a patent-worthy inventive concept. Bascom ruled in favor of granting the
patent following the second step of the Alice
test.[111] While the patent at hand was
considered directed to patent ineligible subject matter, the Bascom court found that the content
filter system invention still had an inventive concept worthy of a patent.[112] Even if elements of a claim are
separately known in prior art, an inventive concept can be found in the
non-conventional and non-generic arrangement of known, conventional pieces.
This inquiry seems like a lenient standard compared to the 35 U.S.C. §103
obviousness inquiry; hence, it is not clear if this step has an independent
utility for invalidating or rejecting a patent. Nonetheless, the court found
that merely showing that all elements of a claim were already disclosed in
prior art was not sufficient reason to make an invention patent ineligible.[113] While
it is possible to infer sufficient reasons of ruling out inventive concepts
from the Bascom case, it is still
unclear what would warrant an invention to pass the second step of the Alice test. Cases such as DDR Holdings v. Hotels.com have
suggested that the second step of Alice
is satisfied since it involved a solution to a specific technological problem
that “is necessarily rooted in computer technology in order to overcome a
problem specifically arising in the realm of computer networks.”[114] This
interpretation of inventive concept
becomes perplexing when comparing the two steps of Alice—both steps look to whether the proposed solution addresses
problems that are specific to a given field of interest. While we would need
additional cases to gain insight on whether the two steps have truly distinct
functions, at the very least the Federal Circuit provided essential guidelines
on what may be deemed as patentable software. As the Bascom court has taught, the first step
in the Alice inquiry is to ask
whether an invention (1) provides a solution to an existing problem by (2) a
specific, non-generic improvement that (3) does not preempt other methods of
solving the existing problem. Applying this test to inventions in machine
learning, mathematical improvements
and computational improvements would
be treated differently. As
mentioned before, a key aspect of machine learning is the “noise” associated
with the data sets.[115] Another concern is the fitting
of a given algorithm to a certain model. Methods that facilitate the
computations of the training process may be deemed as a specific improvement.
However, machine learning algorithms themselves, including the base models that
the algorithm fits the training into would not be pertinent to just a specific
improvement. Hence, generic mathematical methods applicable to various problems
are directed to an abstract idea. For example, an invention that addresses the
issue of normalizing data from different sources would be a computational issue
and hence would pass the Alice test
given that it did not preempt other solutions to the problem of data
normalization. On the other hand, a specific mathematical equation that serves
as a starting model for the machine learning algorithm would be mathematical
and hence directed to an abstract idea. Even if the mathematical starting model
is only good for a specific application, the model is not a specific
improvement pertinent to that application. Although the model may not
necessarily be a good starting model for other applications, it is nonetheless
a generic solution that applies to other applications as well. While
highly restrictive, the guidelines from the Federal Circuit still allow the
grant of patent rights for the computational aspects of machine learning
algorithms. The guidelines also would prevent highly preemptive mathematical
innovations, including data-generating patents such as machine learning models.
The
narrow range of patentability makes a patent regime appealing for computational
methods. The recent emphasis on preemption concerns acts in favor of preventing
data network effects based on data-generating patents. While not discussed in
this paper, other patentability requirements such as obviousness or
definiteness would further constraint the grant of overly broad data-generating
patents. Such an
approach strikes the appropriate balance between promoting innovation and
encouraging data reuse for societal benefits. Compared to other approaches of
providing protection over innovations in machine learning, the narrowly
tailored approach for patent rights for computational inventions fits best with
the policy goal of promoting innovation through data reuse. The industry trends
in collaboration and recruiting also matches the proposed focus on patent law
protection.
B. Industry Trends in Machine Learning
C. Machine Learning Innovators – Protect the Data or Inventions?
1. Protecting the Training Data—Secrecy Works Best
2. Protecting the Inventions—Patent Rights Prevail
3. Protecting the Machine Learning Models and Results—Secrecy Again
D. Need of Patent Rights for Machine Learning Inventions in the Era of Big Data
III. Patentability of Machine Learning Innovations in the Era of Big Data
A. Alice: The Legal Framework of Patentable Subject Matter in Software
1. Step
one: “[D]etermine whether the claims at issue are directed to a patent-ineligible concept. If so, the Court then asks
whether the claim’s [additional] elements, considered both individually and ‘as
an ordered combination,’ ‘transform the nature of the claim’ into a
patent-eligible application.”[99]
2. Step
two: “[E]xamine the elements of the claim to determine whether it contains an ‘inventive concept’ sufficient to
‘transform’ the claimed abstract idea into a patent-eligible application. A
claim that recites an abstract idea must include ‘additional features’ to
ensure that the [claim] is more than a drafting effort designed to monopolized
the [abstract idea]” which requires “more than simply stat[ing] the [abstract
idea] while adding the words ‘apply it.’”[100]
B. The post-Alice cases from the Federal Circuit
1. The Federal Circuit’s Standard for Alice Step One
2. The Federal Circuit’s Standard for Alice Step Two, and the Overlap with Step One
C. Applying Patentable Subject Matter to Machine Learning Inventions
Conclusion
[1] Sang-Hun Choe & John Markoff, Master of Go Board Game Is Walloped by Google Computer Program, N.Y. Times (March 9, 2016), https://www.nytimes.com/2016/03/10/world/asia/google-alphago-lee-se-dol.html (reporting the shocking defeat of Go Master Lee Se-dol to Google DeepMind’s AlphaGo).
[2] Laurence Zuckerman, Chess Triumph Gives IBM a Shot in the Arm, N.Y. Times (May 12, 1997), http://politics.nytimes.com/library/cyber/week/051297ibm.html (detailing IBM’s highly publicized win through Deep Blue’s victory over world chess champion Garry Kasparov).
[4] David Silver et al., Mastering the game of Go with deep neural networks and tree search, 529 Nature 484, 484 (2016).
[5] Id.
[6] Id.
[7] Id. at 485.
[8] Id.
[9] See Andre Esteva et al., Dermatologist-level classification of skin cancer with deep neural networks, 542 Nature 115 (2017).
[10] See Sue Ellen Haupt & Branko Kosovic, Big Data and Machine Learning for Applied Weather Forecasts, (2015).
[11] See Wei-Yang Lin et al., Machine Learning in Financial Crisis Prediction: A Survey, 42 IEEE Transactions on Systems, Man, and Cybernetics 421 (2012).
[12] See Farzindar Atefeh & Wael Khreich, A Survey of Techniques for Event Detection in Twitter, 31 Computational Intelligence 132 (February 2015).
[13] See Corey Blumenthal, ECE Illinois Students Accurately Predicted Trump’s Victory, ECE Illinois (Nov. 18, 2016), https://www.ece.illinois.edu/newsroom/article/19754.
[14] For the purpose of this Note, secrecy refers to the use of trade secret and contract based non-disclosure agreements.
[15] Mark A. Lemley, The Surprising Virtues of Treating Trade Secrets As IP Rights, 61 Stan. L. Rev. 311, 332 (2008) ("Patent and copyright law do not exist solely to encourage invention, however. A second purpose — some argue the main one — is to ensure that the public receives the benefit of those inventions.”).
[16] Andrew Ng et al., How Artificial Intelligence Will Change Everything, Wall Street Journal (March 7, 2017), https://www.wsj.com/articles/how-artificial-intelligence-will-change-everything-1488856320.
[17] Limor Peer, Mind the Gap in Data Reuse: Sharing Data Is Necessary But Not Sufficient for Future Reuse, London Sch. Econ. & Poli. Sci. (Mar. 28, 2014) http://blogs.lse.ac.uk/impactofsocialsciences/2014/03/28/mind-the-gap-in-data-reuse (“The idea that the data will be used by unspecified people, in unspecified ways, at unspecified times . . . is thought to have broad benefits”).
[18] See Saeed Ahmadiani & Shekoufeh Nikfar, Challenges of Access to Medicine and The Responsibility of Pharmaceutical Companies: A Legal Perspective, 24 DARU Journal of Pharmaceutical Sciences 13 (2016) (discussing how “pharmaceutical companies find no incentive to invest on research and development of new medicine specified for a limited population . . .”).
[19] 17 U.S.C. §101 (2012).
[20] Id.
[21] 17 U.S.C. §102(a) (Copyright exists “in original works of authorship fixed in any tangible medium of expression . . .”).
[22] Apple Comput., Inc. v. Franklin Comput. Corp., 714 F.2d 1240 (3d Cir. 1983).
[23] Comput. Assocs. Int’l v. Altai, 982 F.2d 693 (2d Cir. 1992).
[24] Id.
[25] See id. at 707-09.
[26] 837 F.3d 1299, 1314 (Fed. Cir. 2016).
[27] Altai, 982 F.2d at 698.
[28] See Dionne v. Se. Foam Converting & Packaging, Inc., 240 Va. 297 (1990).
[29] Kewanee Oil v. Bicron Corp., 416 U.S. 470 (1974).
[30] Id. at 490.
[31] Nikolay Golova & Lars Rönnbäck, Big Data Normalization For Massively Parallel Processing Databases, 54 Computer Standards & Interfaces 86, 87 (2017).
[32] Jon Russell, Alibaba Teams Up with Samsung, Louis Vuitton and Other Brands to Fight Counterfeit Goods, TechCrunch (Jan. 16, 2017) https://techcrunch.com/2017/01/16/alibaba-big-data-anti-counterfeiting-alliance.
[34] See Lior Rokach, Introduction to Machine Learning, Slideshare 3 (July 30, 2012), https://www.slideshare.net/liorrokach/introduction-to-machine-learning-13809045.
[35] Id. at 4.
[36] Id. at 10.
[37] See Lars Marius Garshol, Introduction to Machine Learning, Slideshare 26 (May 15, 2012) https://www.slideshare.net/larsga/introduction-to-big-datamachine-learning.
[38] Id.
[39] Id.
[40] See Lei Yu et al., Dimensionality Reduction for Data Mining – Techniques, Applications and Trends, Binghamton University Computer Science 11, http://www.cs.binghamton.edu/~lyu/SDM07/DR-SDM07.pdf (last visited Feb. 23, 2018).
[41] Id.
[44] Laurens van der Maaten et al., Dimensionality Reduction: A Comparative Review, Tilburg Centre for Creative Computing, TiCC TR 2009-005, Oct. 26, 2009, at 1 (“In order to handle such real-world data adequately, its dimensionality needs to be reduced. Dimensionality reduction is the transformation of high-dimensional data into a meaningful representation of reduced dimensionality. Ideally, the reduced representation should have a dimensionality that corresponds to the intrinsic dimensionality of the data. The intrinsic dimensionality of data is the minimum number of parameters needed to account for the observed properties of the data”).
[45] Andrew Ng, Nuts and Bolts of Applying Deep Learning (Andrew Ng), YouTube (Sept. 27, 2016), https://www.youtube.com/watch?v=F1ka6a13S9I.
[46] Andrew Ng, Model Selection and Train/Validation/Test Sets, Machine Learning, https://www.coursera.org/learn/machine-learning/lecture/QGKbr/model-selection-and-train-validation-test-sets (last visited Feb. 23, 2018).
[47] Id.
[49] Alex Rampell & Vijay Pande, a16z Podcast: Data Network Effects, Andreesen Horowitz (Mar. 8, 2016), http://a16z.com/2016/03/08/data-network-effects/.
[50] James Manyika et. al., Big Data: The Next Frontier for Innovation, Competition, and Productivity, McKinsey Global Inst., May 2011, at 11, available at https://www.mckinsey.com/~/media/McKinsey/Business%20Functions/McKinsey%20Digital/Our%20Insights/Big%20data%20The%20next%20frontier%20for%20innovation/MGI_big_data_exec_summary.ashx.
[51] Id.
[52] Leonid Bershidsky, Why Are Google Employees So Disloyal?, Bloomberg (July 13, 2013, 11:41 AM), https://www.bloomberg.com/view/articles/2013-07-29/why-are-google-employees-so-disloyal-.
[53] Id.
[54] Rob Valletta, On the Move: California Employment Law and High-Tech Development, Federal Reserve Bank of S.F. (Aug. 16, 2002), http://www.frbsf.org/economic-research/publications/economic-letter/2002/august/on-the-move-california-employment-law-and-high-tech-development/#subhead1.
[55] Id.
[56] See Quentin Hardy, IBM, G.E. and Others Create Big Data Alliance, N.Y. Times (Feb. 15, 2015), https://bits.blogs.nytimes.com/2015/02/17/ibm-g-e-and-others-create-big-data-alliance.
[57] See, e.g., Finicity and Wells Fargo Ink Data Exchange Deal, Wells Fargo (Apr. 4, 2017), https://newsroom.wf.com/press-release/innovation-and-technology/finicity-and-wells-fargo-ink-data-exchange-deal.
[58] Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 134 S. Ct. 2347 (2014).
[59] Karen E.C. Levy, Relational Big Data, 66 Stan. L. Rev. Online 73, 73 n.3 (2013), https://review.law.stanford.edu/wp-content/uploads/sites/3/2013/09/66_StanLRevOnline_73_Levy.pdf (explaining that the big data phenomenon is due to the need of practices to analyze data resources).
[60] Christine L. Borgman, The Conundrum of Sharing Research Data, 63 J. Am. Soc’y for Info. Sci. & Tech. 1059, 1059-60 (2012) (discussing the lack of data sharing across various industries).
[61] See Michael Mattioli, Disclosing Big Data, 99 Minn. L. Rev. 535 (2014).
[62] See id. at 545-46 (discussing the technical challenges in merging data from different sources, and issue of subjective judgments that may be infused in the data sets).
[63] See id. at 552 (discussing how institutions with industrial secrets may rely on secrecy to protect the big data they have accumulated).
[64] See id. at 570 (“[T]he fact that these practices are not self-disclosing (i.e., they cannot be easily reverse-engineered) lends them well to trade secret status, or to mere nondisclosure”).
[65] Id.
[66] Id. at 552.
[67] Patrick Clark, The World’s Top Economists Want to Work for Amazon and Facebook, Bloomberg (June 13, 2016, 10:47 AM), https://www.bloombergquint.com/technology/2016/06/09/the-world-s-top-economists-want-to-work-for-amazon-and-facebook (“If you want to be aware of what interesting questions are out there, you almost have to go and work for one of these companies”).
[68] Bill Franks, Taming the Big Data Tidal Wave 20 (2012) (discussing that the biggest challenge in big data may not be developing tools for data analysis, but rather the processes involved with preparing the data for the analysis).
[69] See Borgman, supra note 60, at 1070 ("Indeed, the greatest advantages of data sharing may be in the combination of data from multiple sources, compared or "mashed up’ in innovative ways." (citing Declan Butler, Mashups Mix Data Into Global Service, 439 Nature 6 (2006))).
[70] Jack Clark, Apple’s Deep Learning Curve, Bʟᴏᴏᴍʙᴇʀɢ Bᴜsɪɴᴇssᴡᴇᴇᴋ, (Oct 29, 2015) https://www.bloomberg.com/news/articles/2015-10-29/apple-s-secrecy-hurts-its-ai-software-development.
[71] Kewanee Oil v. Bicron Corp., 416 U.S. 470, 490 (1974).
[72] See infra Section III-B.
[73] Brenda Simon & Ted Sichelman, Data-Generating Patents, 111 Nw. U.L. Rev. 377 (2017).
[74] Id. at 379.
[75] Id. at 414.
[76] Id. at 415 (“[B]roader rights have substantial downsides, including hindering potential downstream invention and consumer deadweight losses . . .”).
[77] Id. at 417.
[79] Lina Kahn, Amazon’s Antitrust Paradox, 126 Yale L.J. 710, 785 (2017) (“Amazon’s user reviews, for example, serve as a form of network effect: the more users that have purchased and reviewed items on the platform, the more useful information other users can glean from the site”).
[81] See infra Section III-A.
[82] Id.
[83] See infra Section III-B.
[84] Jack Clark, Apple’s Deep Learning Curve, Bʟᴏᴏᴍʙᴇʀɢ Bᴜsɪɴᴇssᴡᴇᴇᴋ (Oct 29, 2015), https://www.bloomberg.com/news/articles/2015-10-29/apple-s-secrecy-hurts-its-ai-software-development.
[85] Peer, supra note 17 (“The idea that the data will be used by unspecified people, in unspecified ways, at unspecified times . . . is thought to have broad benefits”).
[86] See Harshavardhan Achrekar et al., Predicting Flu Trends using Twitter data, IEEE Conference on Comput. Commc’ns. Workshops 713 (2011), http://cse.unl.edu/~byrav/INFOCOM2011/workshops/papers/p713-achrekar.pdf.
[87] Jordan Crook, Uber Taps Foursquare’s Places Data So You Never Have to Type an Address Again, TechCrunch, (May 25, 2016) https://techcrunch.com/2016/05/25/uber-taps-foursquares-places-data-so-you-never-have-to-type-an-address-again/.
[89] See Mattioli, supra note 61, at 554 (“A final limitation on patentability possibly relevant to big data is patent law’s requirement of definiteness”).
[90] See Nautilus, Inc. v. Biosig Instruments, Inc., 134 S. Ct. 2120 (2014).
[91] See Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 134 S. Ct. 2347 (2014).
[92] Robert R. Sachs, Two Years After Alice: A Survey of the Impact of a "Minor Case" (Part 1), Bilski Blog (June 16, 2016), http://www.bilskiblog.com/blog/2016/06/two-years-after-alice-a-survey-of-the-impact-of-a-minor-case.html.
[93] Id.
[94] Stephanie E. Toyos, Alice in Wonderland: Are Patent Trolls Mortally Wounded by Section 101 Uncertainty, 17 Loy. J. Pub. Int. L. 97,100 (2015).
[95] Id.
[96] Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 134 S. Ct. 2347, 2349 (2014).
[97] Id.
[98] Id.
[99] Id. at 2350 (emphasis added) (citation omitted).
[100] Id. at 2357 (emphasis added) (alteration in original) (citation omitted).
[101] Id. at 2350.
[102] Id. at 2351.
[103] Enfish, LLC v. Microsoft Corp., 822 F.3d 1327, 1330 (Fed. Cir. 2016).
[104] Bascom Glob. Internet Servs. v. AT&T Mobility LLC, 827 F.3d 1341, 1349 (Fed. Cir. 2016).
[105] McRO, Inc. v. Bandai Namco Games Am. Inc., 837 F.3d 1299, 1308 (Fed. Cir. 2016).
[106] Id.
[107] Bascom, 827 F.3d at 1349.
[108] Enfish, 822 F.3d at 1330.
[109] McRO, Inc., 837 F.3d at 1314.
[110] Id.
[111] Bascom, 827 F.3d at 1349.
[112] Id.
[113] Id.
[114] See Toyos, supra note 94, at 121; DDR Holdings, LLC v. Hotels.com, 773 F.3d 1245, 1257 (Fed. Cir. 2014).
[115] See supra Section II-A.
