Big Tech control over AI research: an explanation and an alternative
The data and computing resources necessary for research into artificial intelligence are dominated by a handful of companies, creating a fundamental dependence on them, and associated business practices, a prominent AI ethicist warns.
AI Now Institute Co-founder and Minderoo Research Professor at New York University Meredith Whittaker explains the situation in a paper titled ‘The Steep Cost of Capture,’ printed by ACM Interactions, comparing the relationship between big technology companies and AI academia as analogous to the influence the U.S. military had over scientific research during the Cold War.
Advances in AI over the past decade have not, Whittaker argues, resulted from any scientific breakthroughs in AI technique, but rather in the concentration of data and computing resources in the hands of a few large corporations. From the success of the AlexNet object recognition algorithm in 2012 to the call for the “democratizing” access to AI research by the U.S. National Security Commission on Artificial Intelligence (NSCAI), “populated by tech executives” in 2020, she sets out a pattern of public-sector initiatives driven by concentrated private-sector power.
Whittaker also takes aim at Stanford University’s launch of the Center for Research on Foundation Models (CRFM), arguing that what the school calls “foundation models” are actually rebranded “large language models,” which have been criticized by various AI researchers, including Timnit Gebru, over alleged bias, resource usage, and concentration of power.
The ties also include industry-sponsored Ph.D. programs, tech company offices on campus, and a partnership between the National Science Foundation and Amazon to define “fairness.”
Whittaker argues that academic freedom is being stymied, with reference to researchers forced into the private sector and Cold War-era politics. When raising the specter of AI bias, the article also refers to “Republican apparatchiks,” potentially inviting accusations of politicization and the introduction of noise into the signal, as has so often happened in technology policy discussions, particularly in the U.S.
A different approach to funding AI research
The Distributed Artificial Intelligence Research Institute (DAIR) has been launched in response to this restrictive AI research environment by well-known AI and facial recognition researcher Timnit Gebru.
DAIR is described as “a space for independent, community-rooted AI research free from Big Tech’s pervasive influence.”
The MacArthur Foundation, Ford Foundation, Kapor Center, Open Society Foundation and the Rockefeller Foundation have ponied up $3.7 million to support DAIR, The Washington Post reports. DAIR has also joined the Code for Science & Society (CS&S) initiative as part of its Sponsored Projects Program.
Gebru was fired by Google a year ago after criticizing the company’s work on large language models.
“I’ve been frustrated for a long time about the incentive structures that we have in place and how none of them seem to be appropriate for the kind of work I want to do,” Gebru told the Post. The researcher further suggested that even while with Google, she found it easier to influence the company’s policy through outside pressure.
In the event that a DAIR funder is antagonized by the group’s research, Gebru intends to tap other sources of philanthropic support.