Research about the influence of computing technologies, such as artificial intelligence (AI), on society relies heavily upon the financial support of the very companies that produce those technologies. Corporations like Google, Microsoft, and IBM spend millions of dollars each year to sponsor labs, professorships, PhD programs, and conferences in fields like computer science (CS) and AI ethics at some of the world’s top institutions. Industry is the main consumer of academic CS research, and 84% of CS professors receive at least some industry funding. All of these factors contribute to the significant influence tech firms wield over the kinds of questions that are and aren’t asked about their products, and which information is and isn’t made available about their social impact.
As consciousness about these conflicts of interest builds, we are seeing growing calls from scholars in and around CS to disentangle the discipline from Big Tech’s corporate agenda. However, given the extent to which much of CS academia relies on funding from major tech corporations, this is much easier said than done. As I argue below, a more achievable yet valuable goal might be to introduce better safeguards in spaces like conferences to mitigate undue corporate influence over essential research.
I will make my case in two parts. First, in today’s post, I will:
- Provide a quick overview of discourse regarding Big Tech’s dominance in CS research, and
- Use a dataset I’ve compiled to illustrate the extent to which conferences—an essential arena for knowledge sharing in the field of computer science—are financially reliant upon some of the world’s most powerful technology companies.
In my second post, I will follow up with my recommendations for steps that can be taken to minimize the potential chilling or agenda-setting effects brought on by corporate funding on CS research.
A short survey of concerns about Big Tech’s influence
Relying on large companies and the resources they control can create significant limitations for the kinds of CS research that are proposed, funded and published. The tech industry plays a large hand in deciding what is and isn’t worthy of examination, or how issues are framed. For instance, a tech company might have a very different definition of privacy from that which is used by consumer rights advocates. But if the company is determining the parameters for the kinds of research it wishes to sponsor, it can choose to fund proposals that align with or uphold its own interpretation.
The scope of what is reasonable to study is therefore shaped by what is of value to tech companies. There is little incentive for these corporations to fund academic research about issues that they consider more marginal or which don’t relate to their priorities.
A 2020 study on artificial intelligence research found that “with respect to AI, firms have increased corporate research significantly,” in the form of both company-level publications as well as collaborations with elite universities. This trend was illustrated in an analysis by Birhane et al. of top-cited papers published at premier machine learning conferences, which revealed “substantive and increasing corporate presence” in that research. In 2018-19, nearly 80% of the annotated papers had some sort of corporate ties, by either author affiliation or funding. Moreover, the analysis found that corporate presence is more pronounced in the conference papers that end up receiving the most citations.
Birhane et al. write, “the top stated values of ML… such as performance, generalization, and efficiency may not only enable and facilitate the realization of Big Tech’s objectives, but also suppress values such as beneficence, justice, and inclusion.”
One of the most vocal critics of Big Tech’s “capture” of CS academia is Meredith Whittaker, a former Google employee-turned Senior Advisor on AI at the Federal Trade Commission. She argues that tech companies, hoping to muffle critics and fend off mounting regulatory pressure, are eager to shape the narrative around their technologies’ social impact by funding favorable research. This has led to widespread corporate sponsorship of labs, faculty positions, graduate programs, and conferences—all of which are reliant on these companies for not only funding, but often also access to data and computing resources. This industry capture of tech research—wherein corporations are strategically funding research or public campaigns in a way that serves their own agenda—has been described by scholars like Thao Phan et al. as “philanthrocapitalism.”
Furthermore, as Whittaker argues, the tech industry’s dominance in CS research “threatens to deprive frontline communities, policymakers, and the public of vital knowledge about the costs and consequences of AI and the industry responsible for it—right at the time that this work is most needed.” Recognizing this threat, other ex-Googlers like Timnit Gebru and Alex Hanna have taken the initiative to launch the Distributed AI Research Institute, in an effort to create space for “independent, community-rooted AI research free from Big Tech’s pervasive influence.”
I do wish to make clear that receiving funding from an organization that doesn’t completely align with one’s values does not necessarily mean one’s research is compromised. Corporate funding of AI research is not inherently bad, and academics who do not accept Big Tech money can still produce ethically questionable research. Furthermore, individuals who accept Big Tech funding can still be critical of the corporations’ products and their influence on society.
However, I agree with academics like Moshe Y. Vardi who argue that we must grapple with the contradictions inherent in accepting funding for research such as AI ethics from companies whose interests may run counter to the public good. In a recent article, Vardi, who is the senior editor of Communications of the ACM(1), urged his colleagues to think more critically about their field’s relationship to “surveillance-capitalism corporations”, writing: “The biggest problem that computing faces today is not that AI technology is unethical—though machine bias is a serious issue—but that AI technology is used by large and powerful corporations to support a business model that is, arguably, unethical.”
Analysis: FAAMG companies dominate conference sponsorship
One way to begin to address these conflicts of interest is by reflecting on the conditions of knowledge creation and exchange—in spaces such as academic conferences—and thinking critically and openly about the compromises and tradeoffs inherent in accepting funding from the industry that controls the subject of one’s study. In the field of computer science, conferences are the primary venue for sharing one’s research with others in the discipline. Therefore, sponsoring these gatherings gives firms valuable influence over and insight into what’s happening at the cutting edge of topics like machine learning and human-computer interaction.
In an effort to get a better understanding of who the major players are in this realm, I reviewed the websites for the top 25 CS conferences (based on H-5 index and impact score) to compile information about all of the organizations that have financially supported them between 2019 and 2021. I found that a majority of the most frequent and most generous sponsors, often donating tens of thousands of dollars per conference, were powerful technology companies.
This spreadsheet contains sponsorship data for the top 25 most frequent sponsors (2). Of the 10 sponsors who supported the largest numbers of different conferences in the past three years, five are “FAAMG” companies (Facebook, Apple, Amazon, Microsoft, Google)—six if you count DeepMind, a subsidiary of Google’s parent company Alphabet. No non-profit organizations, government science funding agencies, or sponsors from outside the U.S. or China appeared among the top 10.
Overall, among the most frequent and most generous supporters of the top 25 CS conferences, the only non-tech/non-corporate donor was the National Science Foundation, which sponsored five different conferences (11 total gatherings) with donations typically ranging between $15,000 and $25,000.
In addition to having their company name and logo listed on conference promotional materials, top sponsors (who often give upwards of $50,000) receive perks such as opportunities to sponsor prizes or students grants, complimentary registrations and private meeting rooms, access to databases of conference registrants interested in recruitment opportunities, virtual booths or priority exhibition spaces, advertising opportunities and press support, and access to attendee metrics on “exhibitor dashboards”. A “Hero Sponsor” who gave $50,000 or more to the 2021 Conference on Human Factors in Computing Systems (CHI), for example, would have received 34 different benefits – which cumulatively create opportunities for continuous access to and influence on attendees throughout the event.
It is difficult to get an accurate estimate of exactly how much money each company donates to these conferences, as these numbers are not consistently reported to the public. Some conferences only publish a list of supporters with no details about how much each one gave. Others assign sponsorship levels such as “Platinum” or “Diamond”, but the monetary value associated with each level varies by conference and year. When dollar amounts are provided, they often represent a potential range of several thousand dollars—for instance, a Platinum Sponsor of the 2021 SIGMOD/PODS conference might have given anywhere between $16,000 and $31,999. Furthermore, it is difficult gain insight into how exactly these funds are used.
Given the extent of financial entanglement between Big Tech and academia, it might be unrealistic to expect CS scholars to completely resist accepting any industry funding—instead, it may be more practicable to make a concerted effort to establish higher standards for and greater transparency regarding sponsorship.
In Part 2 of this article, I will recommend steps that can be taken to minimize the potential chilling or agenda-setting effects brought on by corporate funding on CS research.
(1) Six of the top 25 CS conferences in the world are organized by ACM, the Association for Computing Machinery. Between 2019 and 2021, many of those conferences were largely funded by American tech companies like Apple, Amazon, Facebook, Google, IBM, and Microsoft, and Chinese ones like Alibaba, Baidu, ByteDance, and Huawei.
(2) I have compiled a conference sponsorship database that includes extensive data that is not included in this spreadsheet. If you are interested in reviewing it, or in collaborating on further data collection, I would be happy to share it privately.
Many, many thanks to Prof. Arvind Narayanan and Karen Rouse for their thoughtful guidance on and support with this piece.
When you say “many of those conferences were largely funded by American tech companies” that suggests that these sponsorship covered a large fraction of the cost of the conference. I suspect this is demonstrably false (registration fees for these conferences are pretty large) and perhaps what you mean to say is that “for many of these conferences, a large fraction the sponsor funding came from American tech companies”.
Also, the question implicit in “Furthermore, it is difficult gain insight into how exactly these funds are used.” can likely be answered asking the conference organizers. Especially in case of conferences run by academics through ACM and such, I believe the conference organizers would be happy to provide said insight.
Defining some of these things is tricky of course. The cost of the conference includes the cost that the conference pays (e.g. hotel conference rooms) and the cost that attendees pay (e.g. travel, hotel). Some sponsor money goes to travel awards that party cover the latter for some subset of the attendees. This would otherwise have come from grants from NSF and/or companies. Money being fungible, there is a lot of leeway in defining what the sponsor money is going towards.
A less cynical observer may say that companies provide sponsorship to these conferences because they want to recruit from the group of people attending the conference. Putting their name out there is valuable as a way to communicate to the attendees that they are looking to hire researchers in these areas. Students also find the job opportunities valuable. Since most PhD graduates in CS do not get academic positions, connecting students to potential employers is a need that the conferences serve.
While I appreciate that there is a broader question of industry-funded research becoming dominant in many areas of CS, I think the narrow focus on conference sponsorship may be misplaced as a demonstration of that. In aggregate the amount of money spent by tech companies on conference sponsorship is negligible, both in numbers and in impact, compared to the money spent on hiring researchers who work on these areas and publish papers in these conferences.
physician, heal thyself:
https://www.princeton.edu/news/2022/03/10/venture-forward-gift-launches-initiative-blockchain-and-decentralization-power
Money has to come from somewhere, to fund research. An alternative to corporate sponsorship could be funding from taxation. That would mean that funding is controlled by politicians. Is that an improvement? Politicians these days are probably as self-serving as any private corporation.
In the US, the National Science Foundation is fairly well insulated from political influences. Peer review is the primary factor in deciding what gets funded; research community input through the CCC plays a big role in setting 10-20-year priorities. The National Institutes of Health are similarly insulated. Tax funding is the largest source of funding for university-based research. Several companies also accept substantial amounts of taxpayer money (e.g., IBM, Xerox, and many tech startups).
This posting reveals many fears and concerns, but presents no evidence that there is actual “control” of the research agenda. Have you considered the alternative hypothesis that by hiring so many faculty into their research labs, the companies find themselves in a situation where the research agenda is being set by academically-minded current and former professors rather than by the C-suite?
More importantly, there many research topics where there is wide consensus across academia, industry, and government about research priorities. We all want ML to generalize better, to model causal relationships, to use fewer computational resources, to be more explainable/interpretable/debuggable, to be more robust to domain shift, measurement error, labeling error, and so on. We all want systems with deeper understanding than current large language and vision models evince. We all are interested in making ML more private and secure. We want to figure out how to create federated learning systems where parties can learn collaboratively while preserving privacy. We all want to strengthen ML operations, continuous deployment, continuous improvement, and so on. When there is such a consensus about the important questions, there is no issue of “control”.
The most contentious topics are those that extend beyond pure technical questions to questions of socio-technical systems, power dynamics, and so on. These range from human-centered AI in the small (e.g., building AI to empower individual people) to the study of feedback loops between recommender systems and human behavior to structural questions about who is empowered vs disempowered by the deployment of various AI technologies at massive scale. These are topics where academic research must act as a counterweight to the profit-seeking orientation of industry. A few corporations have the courage to fund such research, and universities accepting such funding must make it absolutely clear that the corporations have no role in determining the detailed direction of the research. Researchers accepting such funds must be extra-careful to ensure the integrity of the research. One way to do that is to create an advisory committee that can provide critical feedback about the research agenda and periodic evaluation of the results. Researchers must also be prepared for industry funding to disappear. It is important for government and foundations to provide steady funding for these research directions.
I have participated in many discussions among the conference organizations about the potential risks of accepting industry funding. As a general rule, industry funding is devoted to things, such as student travel grants, that can be easily scaled and, if necessary, lost entirely. All of the conferences that I have been involved with cover all of the core costs through registration fees. Hence, if corporate donations were to vanish, the conferences would still be able to function.
Minor point: The fact that many of the top-cited papers are authored by folks in corporate labs reflects the fact that these labs have many outstanding researchers who do excellent work. It also reflects the fact that these are very large labs that produce many papers, only some of which are outstanding. Places like Google Brain, Google Deep Mind, Microsoft Research, and Facebook AI Research are today’s version of Bell Labs. Of course they have effective PR departments as well. But this is not so different from MIT, Stanford, and Harvard. I have written elsewhere about the need to promote outstanding work from smaller and less-well-known institutions.
Right on the mark! See also https://cacm.acm.org/magazines/2022/3/258894-acm-ethics-and-corporate-behavior/fulltext