One year ago Google artificial intelligence researcher Timnit Gebru tweeted, “I was fired” and caused a dispute over the right of employees to question the technical consequences of their companies. On Thursday, he set up a new research body to ask questions about the efficient use of technology that Gebru says Google and other technology companies would not do.
“Instead of fighting inwardly, I want to set an example for an independent organization with a variety of advocacy groups,” says Gebru, founder and director of Distributed Artificial Intelligence Research (DAIR). The first part of the name refers to its goal of being more integrated than most AI labs – which skew white, Western, and male-and recruiting people from other parts of the world who are rarely represented in professional companies.
Gebru was removed from Google after a dispute with management over a research paper on caution and new word processing technology was enthusiastically received by Google and other professional companies. Google said it resigned and was not fired, but later approved fired Margaret Mitchell, another researcher named Gebru led a team of ethical AI researchers. The company put in new checks on topics that his researchers could research. Google spokesman Jason Freidenfelds declined to comment but directed WIRED to a a recent report on the company’s role in AI leadership, which said Google has published more than 500 papers on “reliable innovations” since 2018.
Google’s visibility was highlighted innate conflicts in technical companies that assist or employ researchers to learn the meaning of the technology they want to benefit from. Earlier this year, organizers of a high-level conference on technology and people banned Google’s support for the event. Gebru says DAIR will be free to question the seriousness of AI and will not be distracted by academic policy and pressure to publish what it says could disrupt university research.
DAIR will also work to showcase AI applications that may not be available elsewhere, says Gebru, with the aim of inspiring others to adopt new technologies. One such initiative is engaging the South African public awareness team to see how the legacy of apartheid is still entrenched in land use. An initial analysis of these images found that in a densely populated area, where non-white people live in poverty, most of the uninhabited areas created between 2011 and 2017 were converted into affluent habitats.
A paper on that work will be DAIR’s flagship at AI research later this month at NeurIPS, the world’s most popular AI conference. DAIR’s first researcher, Raesetje Sefala, who lives in South Africa, is the lead author of the paper, which includes foreign researchers.
Safiya Noble, a professor at UCLA who studies how technology platforms are designed for people, works on a team of DAIR consultants. He says the Gebru project is an example of new and integrated organizations that need to move forward in understanding and responding to technical outcomes in the community.
“Black women have been instrumental in helping us understand the high level of professional injury and the various types of technologies that are harmful to people, but we know the limits of American companies and the education that black women face,” Noble said. “Timnit recognized Google’s harm and tried to intervene but did not get much help – for a company that needs such recognition.”
Noble recently started its own nonprofit business, Equity Engine, to support the aspirations of black women. He joined the DAIR faculty team with Ciira wa Maina, a lecturer at Dedan Kimathi University of Technology in Nyeri, Kenya.
DAIR is currently a non-profit service Code for Science and Society but later it will also include as a non-profit in itself, says Gebru. His project has raised $ 3 million from Ford, MacArthur, Rockefeller, and the Open Society Foundation, as well as the Kapor Center. In time, he hopes to disrupt DAIR’s financial aid by taking advice related to its research.