Unlock Editor's Digest free of charge
Roula Khalaf, editor of the FT, picks her favorite stories on this weekly newsletter.
It may sound a bit like a science fiction plot, but in the longer term, artificial intelligence could reach the purpose of rapid self-improvement, escaping human control and wreaking havoc on humans through cyberattacks and even nuclear disasters. This is the priority of some scientists and developers and was the motivation for an AI safety law in California, where 32 of the world's top 50 AI firms are based. But on Sunday, the state's governor, Gavin Newsom, vetoed the bill. The decision is seen as an enormous win for Big Tech, a reckless decision for public safety, and a missed opportunity to set de facto AI security standards nationwide. It's not that easy.
Establishing rules to guard against the potential harms of a technology, especially a technology still in development, is a fragile balancing act. If it is simply too intrusive, it risks stifling innovation in the primary place, meaning society also misses out on its potential advantages. And although the California bill was weakened after intense lobbying from Silicon Valley, uncertainties remained about its impact on AI development and deployment.
A general goal of the California bill was to extend developers' liability for misuse of their models. As admirable as this will be, it may have negative effects. For example, it’s difficult for developers to know prematurely how their technology is likely to be used. They could reconcile this by retiring from research. AI experts also anxious that the bill's security protocols – which included a requirement for firms to construct a “kill switch” into models above a certain threshold – could hinder the event and deployment of open source models, where many Innovations happen.
Another concern was that the laws didn’t specifically goal AI systems utilized in high-risk environments, corresponding to critical infrastructure, or after they use sensitive data. Strict standards were even applied Basic functions.
Given these concerns, Newsom’s decision seems sensible. But that doesn't mean tech firms must have free rein. As AI competition accelerates, there may be serious concern that modelers may miss vulnerabilities. Therefore, it could now make sense for lawmakers to revise the proposed rules and make clear the vague language to raised balance concerns concerning the impact on innovation today. Newsom announced a promising partnership with experts to develop “workable guardrails.” It can be welcome that the Governor recently signed bills geared toward regulating clear and present AI risks – fairly than hypothetical ones – including those related to deepfakes and misinformation.
While California's leadership in AI regulation is commendable, it could even be higher if safety rules were set and enacted on the federal level. This would supply protection across America, prevent the creation of a patchwork of various state laws, and stop the Golden State – the epicenter of American and global AI innovation – from being put at a competitive drawback.
Although the appeal of Silicon Valley's investor and talent pool stays strong, there may be an actual risk that unilateral and overly strict AI regulation could spur model development elsewhere, weakening the state's AI technology ecosystem in the method. As it’s, California has high taxes and is the heaviest regulated state within the USA. Real estate can be expensive. Firms corresponding to U.S. data analytics firm Palantir and brokerage Charles Schwab have recently left the state, and a few tech firms have cut office space.
Dealing with security concerns related to AI development is an art that involves preserving the nice while protecting against the bad. Technological threats to our societies shouldn’t be taken flippantly, but neither should they hinder the emergence of innovation that might help diagnose disease, speed up scientific research, and increase productivity. It's definitely worth the effort to get it right.
