Inequality grows in AI research
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Aïda Amer/Axios
A new working paper finds that as high-level AI research began to require huge amounts of computing resources, large tech firms and elite universities increasingly dominated the field's most important conferences.
Why it matters: If only the most well-funded researchers are able to drive the direction of AI, the diversity of voices will narrow at the very moment when AI is poised to transform how we live and work.
By the numbers: Nur Ahmed and Muntasir Wahed, researchers at Western University and Virginia Tech, respectively, examined more than 170,000 papers presented at 57 prestigious computer science conferences since 2012.
- They found that representatives from elite universities — those ranked 1-50 in the QS World University Rankings — crowded out researchers from mid-tier and lower-tier universities.
- Researchers from large tech firms also increased their presence at the major computer science conferences, both by publishing their own papers and by partnering with academic researchers who themselves primarily came from elite universities.
How it works: Ahmed and Wahed chose 2012 as the starting point because that year a team from the University of Toronto achieved a major breakthrough using deep learning for visual recognition at the ImageNet challenge.
- Deep learning, however, requires massive amounts of computing power and vast data sets, paving the way to AI's unequal playing field.
What they're saying: "We say we need more diversity in the field of AI, but what myself and my colleague found is that we're seeing less diversity over time," says Ahmed.
What's next: Some large tech firms have taken steps to expand access to AI research by partnering with historically Black colleges and universities.
- A bolder idea comes from experts at Stanford University who have urged Washington to create a National Research Cloud that could provide a broader range of researchers access to high-end computing.
The bottom line: The ability to do high-end AI research should depend more on the quality of your ideas than the quantity of your CPUs.
