Human-level learning of complex novel tasks as theory-based modelling, exploration and planning
Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences · 2026
DOI: 10.1098/rsta.2024.0529 ↗Kurz gefasst
Fachartikel in der Zeitschrift Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2026. Verfasst von Pedro Tsividis und 10 weiteren Autor:innen. Beteiligte Einrichtung: MIT , Cambridge, MA, USA. Hauptthemen laut MEDLINE: Learning. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 99 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Honda Research Institute, Office of Naval Research, Center for Brains, Minds, and Machines, Massachusetts Institute of Technology.
Abstract (Originalsprache)
Humans are remarkable in their ability to quickly learn to perform complex tasks. Reinforcement learning (RL) has long been proposed as a model of human learning, and while leading machine RL models have surpassed human expertise at many classic board games and video games, they require vast experience to learn successfully—none of today’s algorithms accounts for humans’ ability to learn so many different tasks so quickly. We study human learning on 90 simple yet challenging video games, showing how people learn most games within a few minutes. To explain this behaviour, we propose a strong form of model-based RL, which we call theory-based RL, because it uses cognitively grounded intuitive theories—rich, abstract, causal representations of objects, agents and their interactions—to explore and model an environment and plan effectively to achieve goals. We instantiate the approach in an agent called EMPA (the exploring, modelling and planning agent). EMPA matches human learning efficiency, generalizing robustly to new game situations and levels, as humans do, and exhibiting similar exploration and learning dynamics. Our work points the way for future efforts to build more detailed behavioural models as well as more human-like learning of complex tasks in artificial intelligence systems. This article is part of the theme issue ‘World models in natural and artificial intelligence’.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- MIT , Cambridge, MA, USA.
- Deep Dive AI , Mexico City, Mexico.
- NYU , New York, NY, USA.
- Department of Psychology, Harvard University, Cambridge, MA, USA.
ORCID-Profile: Samuel J Gershman, Joshua B Tenenbaum
Förderung
- Honda Research Institute
- Office of Naval Research
- Center for Brains, Minds, and Machines, Massachusetts Institute of Technology
- Microsoft Research
Zitieren
APA
Tsividis, P., Loula, J., Burga, J., Rodriguez, J. P., Arnaud, S., Foss, N., Campero, A., Subramanian, A., Pouncy, T., Gershman, S. J., & Tenenbaum, J. B. (2026). Human-level learning of complex novel tasks as theory-based modelling, exploration and planning. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences. https://doi.org/10.1098/rsta.2024.0529
Vancouver
Tsividis P, Loula J, Burga J, Rodriguez JP, Arnaud S, Foss N, et al. Human-level learning of complex novel tasks as theory-based modelling, exploration and planning. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences. 2026. doi:10.1098/rsta.2024.0529
BibTeX
@article{Tsividis2026,
author = {Tsividis, Pedro and Loula, Joao and Burga, Jake and Rodriguez, Juan Pablo and Arnaud, Sergio and Foss, Nate and Campero, Andres and Subramanian, Ajay and Pouncy, Thomas and Gershman, Samuel J and Tenenbaum, Joshua B},
title = {Human-level learning of complex novel tasks as theory-based modelling, exploration and planning},
journal = {Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences},
year = {2026},
doi = {10.1098/rsta.2024.0529}
}
EUVIMED