Fei-Fei Li Warns Against California's AI Legislation

In a recent op-ed for Fortune, renowned AI expert Fei‑Fei Li, often dubbed ‘The Godmother of AI,’ voiced her concerns over California’s proposed AI legislation, SB‑1047, known as the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act. Li argues that while the bill is well‑meaning, it would result in significant unintended consequences for both the state and the nation’s AI ecosystems.

Li emphasized that AI policy must walk a fine line between encouraging innovation and imposing necessary restrictions to mitigate potential risks. Policies that do not strike this balance, she warned, tend to either fall short of their goals or cause unintended damage. In the case of SB‑1047, Li believes the bill leans too heavily towards regulation, which could stymie innovation and place undue burdens on smaller players in the AI field.

One of the main concerns Li highlights is the potential harm SB‑1047 could cause to budding developers and smaller tech companies. The bill holds developers liable for any misuse of their AI models, an expectation that Li argues is unrealistic. She posits that forcing developers, especially new and small‑scale ones, to predict every possible misuse of their models would compel them to act defensively, limiting their creative and innovative capabilities.

Li also pointed out that SB‑1047 mandates a ‘kill switch’ for AI models over a certain size, allowing for shutdowns at any moment. She argues that this requirement would hinder the open‑source community, which has been a bedrock for innovation in technology sectors far beyond just AI. The knowledge that their projects could be abruptly terminated would make developers hesitant to experiment and collaborate, effectively stifling technological progress.

Additionally, Li warned that the bill would severely impact public sector and academic AI research. Since these institutions rely heavily on open‑source models and collaborative efforts to advance, introducing restrictive regulations like a kill switch could be detrimental. It would exacerbate the existing resource and computational disadvantages that public and academic researchers face compared to large tech companies, thereby stalling advancements and innovation in these crucial areas.

Perhaps most strikingly, Li criticized SB‑1047 for not addressing more pressing AI‑related issues such as bias and the creation of deepfakes. The bill focuses instead on models that surpass an arbitrary computational threshold or cost $100 million to train. By doing so, it could restrict advancements across all sectors without effectively tackling the real harms associated with AI technology, Li argued.

Li advocates for an alternative approach to AI legislation—one that embodies a ‘moonshot mentality’ to spur comprehensive AI education, research, and development. She has conveyed these ideas in conversations with prominent leaders, including President Biden, underscoring the need for policies that promote broad‑based innovation while ensuring safety and ethical standards.

In her closing remarks, Li made it clear that although she supports AI governance, it must be designed to foster open‑source development, implement uniform and well‑considered rules, and bolster consumer confidence. She extended an offer to collaborate with the bill’s author, Senator Scott Wiener, to develop legislation that truly supports the growth of a technology‑enabled, human‑centered society. Li’s perspective underscores the importance of balanced AI policy that can keep pace with technological advancements without stifling innovation.