What California AI Bill Could Mean
As we described in this piece on Thursday night, the bill requires AI developers to anticipate “what their customers do with their AI and to pay steep fines if the software causes harm.”
Whoa! You can just imagine what that could mean. As our story pointed out, some people worry this would make AI developers liable for what their customers do with their software. That’s a nightmare scenario.
The devil is in the details, as they say. On a conference call today for subscribers to The Information, proponents of the bill, such as California state Sen. Scott Wiener, debated it with critics, such as Ian Stoica, co-founder of AI startups Anyscale and Databricks and a computer science professor at the University of California, Berkeley. Wiener argued that amendments to the bill deal with the worst-case scenarios outlined by opponents—particularly as related to how people use open-source models.
Stoica, though, argued that the bill won’t do anything to promote AI safety and that it would hurt development of open-source AI software made by local companies. If the bill is signed into law, Stoica predicted the open-source software we use will come from “foreign nations,” not California. “Now you ask yourself…whether that is [a] safer, safer world.”
There’s no guarantee Newsom will sign. Aside from the intense industry opposition, the bill is opposed by Democrats on the national stage, such as Nancy Pelosi. Of course, one of Wiener’s arguments is that California has to act, because Congress won’t. He’s got a point there.
Still, given that Wiener also claims the bill is meant to be forward looking, applying to “models that do not exist yet,” you have to wonder what the rush is. Better to get this right, no?