AI Regulation: State vs. Federal Roles in a Rapidly Evolving Field

The Pace of AI and Regulation

• 💡 AI technology is evolving rapidly, making prescriptive regulations potentially outdated before they are written.
• 🧠 The speaker, drawing from experience as a former insurance regulator, favors concepts-based guidance over rigid rules to adapt to this speed.
• ⚠️ A key concern from former colleague Eric Schmidt was the need to be ready to unplug AI due to its unpredictable nature.

Bias, Causation, and Data

• 📊 The challenge of machine bias is acknowledged, but the focus shifts to the quality and breadth of training data as a primary driver of AI output.
• 📈 The speaker argues that high statistical probability of results (correlation) can be a valid basis for regulatory action, not just strict causation, especially when outcomes affect protected classes.
• ⚖️ Addressing undesirable outcomes is framed as a public policy decision, not an inherent flaw of the machine itself.

The 'Black Box' Problem and Oversight

• 🔍 The concept of looking into the 'black box' of AI is discussed, noting that with complex neural networks (N=infinity), human understanding is impossible.
• 🤖 The tool to understand these complex AI systems may itself need to be generated through AI, raising the question of 'who watches the watchers'.

State vs. Federal AI Regulation

• 🏛️ The speaker expresses a strong preference for states to lead on AI regulation where feasible, emphasizing the importance of states' rights.
• 🌐 However, there's also a recognition that the technology needs legal and regulatory flexibility to innovate in the U.S., suggesting a need for interoperable federal standards.
• 🧩 The expert witness suggests exploring whether existing federal and state laws can address AI concerns, noting that a kaleidoscope of state regulations can be burdensome for consumers and businesses.
• 🤝 There is value in interoperable federal standards, with states potentially addressing specific elements as a complement, not a substitute.