Weak AI Regulation: Worse Than No Regulation, Study Finds

Weak AI safety regulations could make AI products more dangerous than having no regulation at all. This is according to a new study in the Proceedings of the National Academy of Sciences. Researchers from Cornell and Carnegie Mellon University used theoretical economics to model effective AI regulation. They found that strict rules are needed, targeting companies that develop AI models like OpenAI and Google. Focusing only on companies that apply the technology, such as those using AI for medical diagnostics, could backfire. What's interesting is that when only downstream companies are regulated, general-purpose AI developers might cut corners on safety. They expect the downstream companies to handle the final product's safety. This creates a "free-riding behavior," where general providers offload the safety burden. The study suggests that stronger, well-placed regulation can benefit everyone by improving both product safety and utility. This matters because it highlights how government approaches to AI regulation could directly impact the safety of the AI tools we use every day.