Veto against Californian AI law – Silicon Valley gets its way
An AI law should be introduced in California. However, the governor responsible sees too many burdens for AI companies.
Here Gavin Newsom signs a bill, but not SB-1047.
(Image: gov.ca.gov)
It was created with good intentions, but does not lead to the goal – writes the responsible Californian governor, Gavin Newsom, about the planned AI law SB-1047. The abbreviation stands for Senate Bill 1047, the long title is "Safe and Secure Innovation for Fontier Artificial Intelligence Models Act". Newsom has made use of his veto right. AI experts and companies had already voiced harsh criticism of the law – but of a slightly different kind than Newsom's opinion.
In a public statement, the governor explains his reasons. The law takes too little account of the environment in which AI is used –, whether in areas with high risks for society or in completely mundane functions. Newsom even fears that SB-1047 could give people a false sense of security. This is because only large AI models are regulated. Small, specialized models could be at least as dangerous, he warns.
Governor nevertheless calls for regulation
Nevertheless, he believes that AI must be regulated, writes Newsom. There should be protocols for monitoring, guard rails and consequences for misuse of AI. A law must also be adaptable, because AI is a technology that is changing and developing. "To be clear – I agree with the author – we can't afford to wait for a major disaster before taking action to protect the public."
Videos by heise
SB-1047 had also provided for a state board to be created to oversee AI models and providers. Researchers and companies criticized, among other things, the required kill switch for models above a certain size. AI expert Fei-Fei Li, for example, says that this will discourage the open source community in particular from developing. According to Newsom, California is home to 32 of the world's 50 leading AI companies. Newsom also believes it is right that the US state is moving ahead with regulation, while Congress is dragging its feet.
In contrast to the Californian proposal, the European AI regulation regulates AI according to a risk-based approach. Numerous AI applications can be used without further ado, such as spam filters for emails. Other applications, such as social scoring, are prohibited. Still others must fulfill certain transparency and documentation obligations.
(emw)
