Apr 1, 2020 - Technology

Stanford experts call for national resource for AI research

An illustration of a building made out of code.
Illustration: Rebecca Zisser/Axios

Artificial intelligence experts at Stanford University are calling for the creation of a task force to establish a National Research Cloud to aid American AI research.

Why it matters: Government support for basic science helped create the post-war American technological colossus. But the unique resource needs of advanced AI research demand a new approach to ensure the field isn't dominated by a few large, rich companies.

The research used to train advanced AI requires a great deal of two things: computational power and data.

  • Google needed nearly $1.5 million in computational cycles to train its Meena chatbot announced earlier this year, while Facebook is able to tap its enormous user-generated dataset for its own AI research. It's impossible for most academic AI researchers to access anywhere near that much computational power or raw data.
  • To open up AI research to a wider group of players, experts at the Stanford Institute for Human-Centered Artificial Intelligence want to create a public-private task force that will build what they call a National Research Cloud.

How it works: The National Research Cloud would seek to even the playing field by providing academic researchers affordable access to high-powered computational resources as well as access to datasets held in government agencies like Medicare and the VA.

  • "If you're at a college in a state like Kansas, there's no way for you to do this research now," says John Etchemendy, the co-director of the Stanford center. "But if we have a real national push to provide that compute and data, in a privacy protective way, it would benefit society at large."

The bottom line: The only way to ensure the boons of AI research are spread widely is to eliminate the barriers to doing that work.

Editor's note: John Etchemendy was mistakenly quoted as referring to Kansas State. The quote has been corrected.

Go deeper