Tuesday, April 01, 2025

An example of AI representing concepts outside the current sphere of human knowledge that are teachable to human experts.

An open source article from the latest PNAS from Schut et al.:

Significance

 
As AI systems become more capable, they may internally represent concepts outside the sphere of human knowledge. This work gives an end-to-end example of unearthing machine-unique knowledge in the domain of chess. We obtain machine-unique knowledge from an AI system (AlphaZero) by a method that finds novel yet teachable concepts and show that it can be transferred to human experts (grandmasters). In particular, the new knowledge is learned from internal mathematical representations without a priori knowing what it is or where to start. The produced knowledge from AlphaZero (new chess concepts) is then taught to four grandmasters in a setting where we can quantify their learning, showing that machine-guided discovery and teaching is possible at the highest human level.
 

Abstract

 
AI systems have attained superhuman performance across various domains. If the hidden knowledge encoded in these highly capable systems can be leveraged, human knowledge and performance can be advanced. Yet, this internal knowledge is difficult to extract. Due to the vast space of possible internal representations, searching for meaningful new conceptual knowledge can be like finding a needle in a haystack. Here, we introduce a method that extracts new chess concepts from AlphaZero, an AI system that mastered chess via self-play without human supervision. Our method excavates vectors that represent concepts from AlphaZero’s internal representations using convex optimization, and filters the concepts based on teachability (whether the concept is transferable to another AI agent) and novelty (whether the concept contains information not present in human chess games). These steps ensure that the discovered concepts are useful and meaningful. For the resulting set of concepts, prototypes (chess puzzle–solution pairs) are presented to experts for final validation. In a preliminary human study, four top chess grandmasters (all former or current world chess champions) were evaluated on their ability to solve concept prototype positions. All grandmasters showed improvement after the learning phase, suggesting that the concepts are at the frontier of human understanding. Despite the small scale, our result is a proof of concept demonstrating the possibility of leveraging knowledge from a highly capable AI system to advance the frontier of human knowledge; a development that could bear profound implications and shape how we interact with AI systems across many applications.

 

No comments:

Post a Comment