AI and the paradox of trust by Yuval Noah Harari - Analysis

Blog PostJune 20, 20255 min readby Sergey Arkhipov

An interesting video "AI and the paradox of trust" by Yuval Noah Harari https://www.youtube.com/watch?v=8GaW36EfidI&ab_channel=YuvalNoahHarari

On the one hand, I really like the clear structured presentation and simplicity of the idea: the author worries people trust each other less than AI. But, I didn't like the loose interpretations of terms and incorrect conclusions.

1. 01The author defines AI as an agent who makes decisions on his own, can create new ideas and independently learn and change. It is very controversial and represents a narrow view of a specific AI implementation.

2. 02The author claims that "AI developers do not trust other AI developers in developing AI, but they trust their AI" and then incorrectly concludes that people trust each other less than AI. This compares different kinds of trust and misses the point: developers often trust the values and constraints they put into their own system more than the values put into another team's system.

3. 03The idea "but they trust their AI" does not specify what exactly they trust. It can sound like they trust AI in general, which is fundamentally wrong for most AI development teams I know from scientific articles on LLM development.

Interesting that the author specializes in Medieval History and increasingly talks and writes about AI. I remember Max Tegmark's ideas from Life 3.0. Our own attempt to measure what models get right, rather than argue about it, is zebra_simple: an open reasoning benchmark with the code published.