PaperlayerView exact revision on arXiv

Source-linked AI summary

The Grey Hoodie Project: Big Tobacco, Big Tech, and the threat on academic integrity

Mohamed Abdalla, Moustafa Abdalla

arXiv:2009.13676v4cs.CY

TL;DR

The paper examines whether Big Tech funding of academic AI-ethics work raises conflicts of interest, using Big Tobacco’s history as a comparison. It identifies similar strategies and argues that universities should consider the tradeoffs and conditions of accepting such funding.

  • Problem

    The paper asks whether Big Tech funding of AI-related academic work creates conflicts of interest, a concern relevant to academics advising on AI policy.

    Source: §1 · p. 1

  • Method

    The authors compare Big Tech’s academic funding with Big Tobacco’s actions across four proposed goals, while not claiming to prove Big Tech’s intent.

    Source: §4 · p. 3

  • Results

    The authors find similarities between Big Tech and Big Tobacco’s interactions with academia and argue that funding can affect academic work through perverse incentives.

    Source: §4 · p. 3, §9 · p. 9

  • Takeaways & Limitations

    The authors encourage academia to consider stricter ethics and operating rules for AI-ethics research, including separation from traditional computer science departments.

    Source: §9 · p. 9

    The observed funding patterns cannot be generalized to all academics, and the Big Tobacco–Big Tech analogy is imperfect.

    Source: §9 · p. 9

Abstract

from arXiv · show

As governmental bodies rely on academics' expert advice to shape policy regarding Artificial Intelligence, it is important that these academics not have conflicts of interests that may cloud or bias their judgement. Our work explores how Big Tech can actively distort the academic landscape to suit its needs. By comparing the well-studied actions of another industry (Big Tobacco) to the current actions of Big Tech we see similar strategies employed by both industries. These strategies enable either industry to sway and influence academic and public discourse. We examine the funding of academic research as a tool used by Big Tech to put forward a socially responsible public image, influence events hosted by and decisions made by funded universities, influence the research questions and plans of individual scientists, and discover receptive academics who can be leveraged. We demonstrate how Big Tech can affect academia from the institutional level down to individual researchers. Thus, we believe that it is vital, particularly for universities and other institutions of higher learning, to discuss the appropriateness and the tradeoffs of accepting funding from Big Tech, and what limitations or conditions should be put in place.

1 INTRODUCTION

The paper frames Big Tech funding in machine learning as a conflict of interest analogous to Big Tobacco funding public-health research. It compares the industries’ actions to assess risks to research integrity and impartiality.

1 INTRODUCTION

  • The authors argue that Big Tech funding of machine-learning discussions about AI ethics presents a significant academic conflict of interest, though many in the field regard it as a non-issue.They compare it to the hypothetical case of tobacco-funded health-policy speakers and a tobacco-funded conference.Source: §1 · p. 1
  • The paper compares Big Tech’s actions with historical industry interference to examine its involvement in and leadership of academic discussions on AI ethics.The authors say the comparison raises concerns for current research integrity and impartiality.Source: §1 · p. 1

2 DEFINING BIG TECH

The authors define Big Tech for this paper by company size and involvement in AI ethics or fairness literature, or lobbying governments on those matters. They acknowledge that the boundary is unclear and use a deliberately conservative list.

2 DEFINING BIG TECH

  • The paper’s Big Tech criteria are large technology corporations involved in AI ethics or fairness literature, or lobbying governments on related matters.The authors say some cases are unclear, including whether Yahoo is large enough or Disney is a technology company.Source: §2 · p. 1
  • The authors’ deliberately conservative list includes Google, Amazon, Facebook, Microsoft, Apple, Nvidia, Intel, IBM, Huawei, Samsung, Uber, Alibaba, Element AI, and OpenAI.They say the list was reached through discussions and a vote with non-author parties.Source: §2 · pp. 1–2
  • The paper excludes companies including Disney, Adobe, Sony, Oracle, Cisco, Netflix, Yahoo, and VMWare from its Big Tech list.The authors state that adding companies to their conservative list would leave the paper’s conclusions intact or strengthen them.Source: §2 · pp. 1–2

3 MOTIVATING THE ANALOGY: BIG TOBACCO’S AND BIG TECH’S PLAYBOOK

The paper uses Big Tobacco’s history as a comparison for examining Big Tech’s response to public criticism and its funding of academic work. The authors caution that the analogy is imperfect.

3 MOTIVATING THE ANALOGY: BIG TOBACCO’S AND BIG TECH’S PLAYBOOK

  • The analogy offers a historical comparison for identifying industry actions and a prompt to examine Big Tech funding critically, but the authors say it is imperfect.They note that peer review was not treated as a solution to conflicts of interest in life science, biology, and public health.Source: §3 · p. 2

3.1 Big Tobacco

  • Big Tobacco’s 1954 public campaign presented the industry as concerned with health while seeking to sow doubt about research linking smoking and lung cancer.The campaign included the Tobacco Industry Research Committee, whose stated purpose was to fund independent research; litigation later exposed the funding’s underlying motivation.Source: §3.1 · p. 2

3.2 Big Tech

  • Big Tech’s response to a worsening public image included funding institutions and causes associated with ethical AI and responsible development.Examples include Amazon’s co-sponsorship of a $20 million fairness-in-AI program and companies’ support for centers studying technology’s societal impact.Source: §3.2 · p. 2
  • Figure 1 depicts various ways tobacco-industry money can enter academia.The figure is recreated from Cohen et al. [16].Source: Figure 1 · p. 3

4 METHODOLOGY

The authors compare Big Tobacco and Big Tech funding in academia across four proposed goals. They argue that funding can warp academia through perverse incentives without proving intentional influence by Big Tech.

4 METHODOLOGY

  • The paper describes Big Tobacco funding as advancing industry interests through covert means.The authors use the tobacco-industry literature as a basis for examining whether Big Tech funding can serve similar purposes.Source: §4 · p. 3
  • The paper compares the industries’ actions across four goals: improving public image, influencing university decisions, shaping researchers’ plans, and identifying receptive academics.These goals are drawn from the literature on Big Tobacco’s investment in academia.Source: §4 · p. 3
  • The authors make one-to-one comparisons between Big Tobacco and Big Tech actions, but do not claim that Big Tech intends to influence academia.They argue instead that industry funding can warp academia through perverse incentives regardless of intent.Source: §4 · p. 3
  • The paper notes that Google questioned anti-Google work by the Campaign for Accountability because Oracle funded the nonprofit.The authors use this example to illustrate a challenge to research based on the funder’s external interests.Source: §4 · p. 3

5 REINVENT ITSELF IN THE PUBLIC IMAGE AS SOCIALLY RESPONSIBLE

Big Tobacco used research funding to project social responsibility while steering funded work away from questions that threatened its interests. The paper describes analogous Big Tech initiatives and argues their benefits remain constrained by corporate profit motives.

5.1 Big Tobacco

  • CTR funding supported quality research, but most grants went to work unrelated to tobacco’s health effects, and funders rejected proposals acknowledging smoking’s dangers.The paper says funding also supported research shifting blame for lung cancer to other sources.Source: §5.1 · p. 3
  • CTR funding was presented as evidence of social responsibility, including to jurors in a cigarette product liability trial.Source: §5.1 · p. 3

5.2 Big Tech

  • Leaked Google documents show PR directing scientists to use a positive tone in research, including work with external academic collaborators.Source: §5.2 · pp. 3–4
  • Google, Microsoft, Facebook, IBM, and Amazon founded the Partnership on AI, which non-industry members said served a legitimating function without giving them power.Access Now later said the partnership had not systematically changed member companies’ attitudes or engagement with civil society.Source: §5.2 · p. 4
  • Companies also created advisory bodies, committees, and grants framed around ethical AI or social issues, including Facebook’s $300 million journalism grants.Source: §5.2 · p. 4
  • Big Tech’s initiatives may produce tangible good, but the authors say corporate profit motives limit them and companies use research to argue against scrutiny.The paper cites Zuckerberg’s invocation of independent academics and Google’s claims about responsible-AI publications.Source: §5.2 · p. 4

6 INFLUENCE THE EVENTS AND DECISIONS MADE BY FUNDED UNIVERSITIES

The paper describes how industry funding can give sponsors influence over university decisions and events. It argues that Big Tech funding and affiliations shape conference agendas and institutional responses, while disclosure policies may not prevent bias.

6.1 Big Tobacco

  • Big Tobacco funding could influence university decisions, including Imperial Tobacco withholding support after UofT law students helped prompt charges against a retailer.Source: §6.1 · p. 4
  • Tobacco-linked university actors were associated with institutional inaction, and tobacco-sponsored events were described as skewed and lower quality than unsponsored events.Source: §6.1 · p. 4

6.2 Big Tech

  • Big Tech funding gives companies a strong voice in conferences and academia, and ethics- or fairness-related NeurIPS workshops almost always had a Big Tech-affiliated organizer from 2015 to 2019.NeurIPS had at least two highest-tier Big Tech sponsors yearly since 2015; in recent years, more than five sponsored at that tier.Source: §6.2 · p. 4
  • FAccT had Big Tech funding in every year examined, while organizers’ sponsorship statements left unclear how effectively disclosure prevents unconscious bias.The authors note research in other fields links industry funding to negative effects and subconscious bias.Source: §6.2 · pp. 4–5
  • Google-sponsored workshops and conferences could shape which policy questions were discussed and which privacy framing received emphasis.A Google-sponsored policy conference included a “token Google critic”; a federated-learning workshop emphasized data confidentiality over privacy as a human right.Source: §6.2 · p. 5
  • The Vector Institute had industry-linked leaders and about one third of its funding from industry, while the authors caution that funding loss could affect its public stance.They report that the institute remained quiet about funders’ ethical issues, without claiming this was necessarily caused by funders.Source: §6.2 · p. 5

7 INFLUENCE THE RESEARCH QUESTIONS AND PLANS OF INDIVIDUAL SCIENTISTS

The paper describes tobacco’s use of grant choices and pressure to redirect research toward less threatening questions. It presents evidence that Big Tech can influence research tone and plans, while noting that many funded projects also produce scientific benefits.

7.1 Big Tobacco

  • Big Tobacco funded research shifting blame from tobacco or redirected funding toward basic cancer science, which the paper says could sow confusion.Source: §7.1 · p. 5
  • Tobacco companies threatened to publicly attack researchers and worked with politicians to block funding for scientists opposing their views.Source: §7.1 · p. 5
  • The paper argues researchers seeking grants may change their questions because Big Tobacco would not fund some lines of inquiry.Source: §7.1 · p. 5

7.2 Big Tech

  • Big Tech funding supports many beneficial projects, but the paper says the questions, projects, and answers are also influenced by the funder.Source: §7.2 · p. 5
  • Google’s leaked internal documents instructed scientists to “strike a positive tone” in research.Researchers studying some topics, including AI effects, had to consult company policy and public-relations teams.Source: §7.2 · p. 5

Faculty Funding

Across four universities, the authors find extensive Big Tech financial ties among computer science faculty, especially among those publishing on AI ethics or fairness. Their analysis uses public faculty information and applies several definitions of funding and affiliation.

Faculty Funding

  • Across four R1 universities, 52% (77/149) of faculty with known funding sources had direct Big Tech funding; including graduate funding and past employment, 84% (125/148) had received compensation.Source: Faculty Funding · p. 6
  • 58% (19/33) of faculty with known funding sources who published on AI ethics or fairness had direct Big Tech funding; with graduate funding and past employment included, 97% (32/33) had received Big Tech compensation.The publication criterion was at least one ethics or fairness paper from January 2015 through April 2020.Source: Faculty Funding · p. 6

7.2.1 Methodology.

  • The study covered research-stream tenure-track computer science faculty at MIT, Toronto, Stanford, and Berkeley, comparing overall, AI, and ethics-of-AI groups.Source: §7.2.1 · p. 6, Faculty Funding – Analysis · p. 6
  • Faculty were classified by AI and AI-ethics work, direct grants or awards, graduate funding, and any Big Tech employment, with unknowns where public information was insufficient.The authors treated published CVs as comprehensive, while acknowledging that faculty classified as unfunded might simply omit awards from personal websites.Source: §7.2.1 · p. 6

Faculty Funding – Analysis.

  • Figure 2 presents career-long direct-funding percentages by computer-science specialization, while the analysis also expands financial ties to include graduate funding and employment.The figure’s axis intentionally does not start at zero.Source: Figure 2 · p. 6, Faculty Association – Analysis · p. 6

Faculty Association – Analysis.

Faculty Funding – Analysis.

  • The authors say past financial ties do not necessarily imply conscious research bias, but repeated positive exposure to industry views may influence scholars subconsciously.Source: Faculty Association – Analysis · p. 7

Funder – Analysis.

  • After finding widespread direct funding, the authors examine which corporations provided it in a separate analysis.Source: Funder – Analysis · p. 7

Journal Author Funding and Association

A review of AI-society position papers in Nature and Science found frequent author ties to Big Tech. The authors interpret these statistics as evidence of a conflict between industry and academic research agendas.

7.2.2 Methodology.

  • The review examined commentary and prospective pieces in Nature and Science, identifying 17 AI societal-impact papers among 51 articles remaining after screening.The search covered machine-learning terms related to fairness, bias, ethics, racism, and disparity.Source: Journal Author Funding and Association · p. 7, §7.2.2 · p. 8

Journal Author Funding and Association

  • Figure 3 and Tables 4–7 frame financial associations across specializations, schools, AI faculty, AI-ethics authors, and leading company contributors.The supplied captions describe these analyses but provide no underlying values.Source: Figure 3 · p. 7, Table 4 · p. 7, Table 5 · p. 7
  • 59% (10/17) of position papers on AI’s ethical or societal impact had at least one author financially involved with a Big Tech company at some point.The count includes faculty awards and former consulting gigs.Source: Journal Author Funding and Association · p. 7

7.2.2 Methodology.

  • The authors say these statistics demonstrate a conflict between Big Tech and academic research agendas, and suggest fairness work often reflects the view that social problems can be addressed through technical solutions.They note that opposing viewpoints are present but likely a minority in workshops, conferences, and symposiums.Source: §7.2.2 · p. 8

8 DISCOVER RECEPTIVE ACADEMICS WHO CAN BE LEVERAGED

The paper compares Big Tech’s recruitment and use of receptive academics with Big Tobacco’s earlier tactics. It cites examples of academic influence and notes that the relevant Big Tech connections are difficult to analyze because they are not fully exposed.

8.1 Big Tobacco

  • Big Tobacco funded sympathetic academics to sow doubt, amplify skeptical messages, and identify researchers whose work could cast tobacco evidence as inconclusive.Its lawyers also sought academics to testify against regulation, including scientists with no prior tobacco-company ties.Source: §8.1 · p. 8

8.2 Big Tech

  • The authors say Big Tech uses funding and organizational power similarly to identify receptive academics who can be used to oppose legislation and litigation.This is the paper’s stated parallel with Big Tobacco’s recruitment strategy.Source: §8.2 · p. 8, §8.1 · p. 8
  • Examples include Google’s effort to groom academic standard-bearers, executive input into whom to fund, and a Google-funded researcher later criticizing antitrust scrutiny before joining the FTC.The paper also describes a Google-funded paper cited to Congress without disclosure of its funding.Source: §8.2 · p. 8
  • A former Google policy and public-relations employee described intentionally leveraging a professor to influence policy, disputing Google’s claim that its academic collaborations were not policy-driven.Source: §8.2 · p. 8
  • Investigative reporting described academics receiving industry funding after pro-industry policy views and later publishing aligned position papers without disclosing the financial ties.The authors say these connections are difficult to analyze because, unlike Big Tobacco, Big Tech has not faced litigation that exposed relevant documents.Source: §8.2 · pp. 8–9

9 DISCUSSION AND CONCLUSION

The authors argue that academia–Big Tech interactions resemble earlier academia–Big Tobacco interactions and raise concerns about research integrity. They frame industry-funding influence as systemic and call for structural responses, while noting limits to their evidence and analogy.

9 DISCUSSION AND CONCLUSION

  • Industry-funding conflicts are systemic, and influence can arise through repeated interactions without an explicit intent to manipulate.The authors say this does not call individual researchers’ integrity into question and that most researchers are well-intentioned and skilled.Source: §9 · p. 9
  • The funding patterns observed cannot be generalized to all academics because the study considered four elite institutions.The authors note that computing power and funding are concentrated at elite institutions [3].Source: §9 · p. 9
  • The Big Tobacco analogy is imperfect: Big Tech is generally viewed as having a net positive impact, and AI’s societal effects can be harder to demonstrate.The authors also note that critical research may depend on access to systems protected as trade secrets.Source: §9 · p. 9
  • The authors see parallels between academia’s interactions with Big Tech and Big Tobacco as grounds for concern and for the academy to prevent history from repeating.They note that decisive evidence about Big Tobacco’s behavior emerged only after years of litigation.Source: §9 · p. 9
  • The authors recommend a stricter AI-ethics code and structural separation from traditional computer science departments to protect ethics research from Big Tech influence.They distinguish this from technical research, where industry relationships may be more acceptable.Source: §9 · p. 9
  • Researchers should disclose complete funding and historic affiliations, while universities should state their policies on direct Big Tech funding.The authors identify missing funding information as a major obstacle to analyzing funding sources and possible industry effects.Source: §9 · p. 9
  • The authors call for discussion of industry relationships in AI ethics and for computer science to engage with antagonistic thinkers.They urge researchers to meet opposition to their ideas with an open and receptive mind.Source: §9 · p. 10
Loading 2009.13676v4…