EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Bridging the human–AI knowledge gap through concept discovery and transfer in AlphaZero

Lisa Schut iD, Nenad Tomašev, Thomas McGrath iD, Demis Hassabis, Ulrich Paquet, Been Kim iD

Proceedings of the National Academy of Sciences · 2025-04 · 122 (13)

DOI: 10.1073/pnas.2406675122 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the National Academy of Sciences, online erschienen am 26.03.2025. Verfasst von Lisa Schut und 5 weiteren Autor:innen. Beteiligte Einrichtung: Oxford Applied and Theoretical Machine Learning Group, Department of Computer Science, University of Oxford, Oxford OX1 3QG, United Kingdom. Hauptthemen laut MEDLINE: Artificial Intelligence, Knowledge. Die Arbeit wurde bisher 18-mal zitiert (Crossref) und nennt 94 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

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.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Artificial IntelligenceKnowledgeHumans

Stichwörter der Autor:innen

AIconcept discoverymachine learningreinforcement learning

Autor:innen und Institutionen

  • Oxford Applied and Theoretical Machine Learning Group, Department of Computer Science, University of Oxford, Oxford OX1 3QG, United Kingdom.
  • Google DeepMind, London N1C 4AG, United Kingdom.
  • Goodfire AI, San Francisco, CA 94102.
  • Google DeepMind, Mountain View, CA 94043.

ORCID-Profile: Lisa Schut, Thomas McGrath, Been Kim

Zitieren

APA

Schut, L., Tomašev, N., McGrath, T., Hassabis, D., Paquet, U., & Kim, B. (2025). Bridging the human–AI knowledge gap through concept discovery and transfer in AlphaZero. Proceedings of the National Academy of Sciences, 122(13). https://doi.org/10.1073/pnas.2406675122

Vancouver

Schut L, Tomašev N, McGrath T, Hassabis D, Paquet U, Kim B. Bridging the human–AI knowledge gap through concept discovery and transfer in AlphaZero. Proceedings of the National Academy of Sciences. 2025;122(13). doi:10.1073/pnas.2406675122

BibTeX
@article{Schut2025,
  author = {Schut, Lisa and Tomašev, Nenad and McGrath, Thomas and Hassabis, Demis and Paquet, Ulrich and Kim, Been},
  title = {Bridging the human–AI knowledge gap through concept discovery and transfer in AlphaZero},
  journal = {Proceedings of the National Academy of Sciences},
  year = {2025},
  volume = {122},
  number = {13},
  doi = {10.1073/pnas.2406675122}
}
Kostenlos starten →