Opinion: Experts often sound less certain than non-experts on complex AI topics

The Nuance of Expertise

The author observed a conversation with AI researchers John Schulman, Beren Millidge, and Charlie O’Neill, noting their use of hedging language and filler words when discussing difficult questions like AI automating research or choosing objectives. This behavior is attributed to their deep, first-hand knowledge and attempts to reason at the edge of current understanding.

Contrast with Public Discourse

This contrasts with public conversations about AI, often led by politicians, executives, and commentators who deliver confident, simplified statements on topics they may not fully understand. The article suggests that those with less direct engagement with reality tend to speak with more certainty.

The Advantage of Simplicity

Researchers often qualify claims because the honest answers to technical questions are complex and depend on multiple factors. However, speakers who remove conditions, compress probabilities into predictions, and simplify complex systems into slogans gain an advantage in public discourse. This is because public discourse tends to reward confidence more than calibration or accuracy.

Implications for AI

While this phenomenon is not exclusive to AI, the article posits that AI amplifies it. The complex and rapidly evolving nature of AI means that truly informed perspectives often involve significant caveats and uncertainties, which are less palatable in a public sphere that prefers definitive answers.