AI Experts Call for Targeted Regulation Amid Rising Capabilities

Posted on October 31, 2024

Recent advances in artificial intelligence have prompted experts to advocate for swift regulatory action within the next 18 months. AI systems have demonstrated significant improvements in mathematics, reasoning, and computer coding, with notable progress in cybersecurity and scientific understanding.

According to recent benchmarks, AI models have shown dramatic improvement in solving real-world coding problems, increasing from 1.96% success rate in October 2023 to 49% by October 2024. Similarly, on scientific knowledge tests, AI performance has approached human expert levels, rising from 38.8% to 77.3% (compared to human experts at 81.2%).

The proposed regulatory framework emphasizes three key elements: transparency in AI companies' safety policies, incentives for better security practices, and simplicity in implementation. Experts suggest using Responsible Scaling Policies (RSPs) as a model for regulation, which require safety measures to increase proportionally with AI capabilities.

While acknowledging that perfect regulations are difficult to design, proponents argue that careful, targeted oversight is essential to balance innovation with risk mitigation. The goal is to implement federal-level regulations in the US, though state-level action may be necessary if federal progress is too slow.