California Forces a Rethink of A.I. Regulation

ImageHow should A.I.be regulated?The most sweeping effort yet to regulate artificial intelligence, a California bill that could have informed laws around the world, is going back to the drawing board.Gov.

Gavin Newsom vetoed the legislation, known as S.B.1047 — under strong pressure from Silicon Valley giants.

Now, governments must again try to figure out the best way to rein in the fast-growing technology’s excesses, while letting innovation flourish.“I do not believe this is the best approach to protecting the public,” Newsom said of his veto.It was a rebuke that underscored the divide over S.B.

1047, which mandated safety testing of A.I.models that required a certain level of computing power and cost at least $100 million to train.Proponents, including Geoffrey Hinton, an A.I.

pioneer, and Elon Musk, said that S.B.1047 provided necessary guardrails, and they urged California policymakers to reject intense pressure from software giants against the bill.

Hollywood actors and writers also supported the legislation.Opponents, including prominent venture capitalists and tech executives, called S.B.1047 a blunt instrument that threatened to choke off innovation.

Smaller tech companies also pushed back, worried that A.I.giants might not make their models publicly available if the legislation passed.

(Representative Nancy Pelosi also urged state legislators to reject the bill and applauded Newsom’s decision.)The Wall Street Journal notes that there are nuances in how A.I.models work: Some smaller models handle decision-making for critical situations such as power grids, while larger models are sometimes deployed for relatively safe matters including customer service.Regulating A.I.

has proved tricky to do.While governments around the world (and A.I.

leaders including Sam Altman of OpenAI and Demis Hassabis at Google) broadly agree that guardrails are needed, none has passed anything as sweeping as S.B.1047.

The broadest law so far is the European Unio...