Sound Research WIKINDX
Sound Research WIKINDX |
|
 |
|
Dupoux, E., LeCun, Y., & Malik, J. (2026). Why AI systems don't learn and what to do about it: Lessons on autonomous learning from cognitive science. arXiv preprint arXiv:2603.15381. Added by: alexb44 (5/8/26, 6:41 AM) Last edited by: Mark Grimshaw-Aagaard (5/8/26, 8:46 AM) |
 |
| Resource type: Journal Article Language: en: English Published DOI: 10.48550/arXiv.2603.15381 BibTeX citation key: Dupoux2026 Email resource to friend View all bibliographic details  |
Categories: General Keywords: Artificial Intelligence, Cognition, Learning Creators: Dupoux, LeCun, Malik Collection: arXiv preprint arXiv:2603.15381 |
Views: 4/101
|
| Abstract |
|
We critically examine the limitations of current AI models in achieving autonomous learning and propose a learning architecture inspired by human and animal cognition. The proposed framework integrates learning from observation (System A) and learning from active behavior (System B) while flexibly switching between these learning modes as a function of internally generated meta-control signals (System M). We discuss how this could be built by taking inspiration on how organisms adapt to real-world, dynamic environments across evolutionary and developmental timescales.
Added by: alexb44 Last edited by: Mark Grimshaw-Aagaard |
| Notes |
|
Foreword is interesting: The dominant AI research paradigm today relies on hyperscaling of text-based LLMs with ever larger models, data and compute. But even prominent architects of this approach such as Ilya Sutskevera and Andrei Karpathy b suggest we may be hitting diminishing returns. Areas of concern include (1) confronting the "data wall" on quality text data (2) inability to learn new things beyond current human knowledge because of the absence of interaction with the environment (Silver and Sutton, 2025) (3) excessively languagecentrism as opposed to spatial, embodied and grounded reasoning in the physical world (4) lack of continual life-long learning (self-improvement after deployment). While these critiques echo long standing controversies within cognitive science on the non-verbal cognition (Johnson-Laird, 1983), and situated interactions (Piaget, 1952; Vygotsky and Cole, 1978) in intelligence, it behooves us as scientists to take stock of progress from both fields and look beyond the current paradigm. What could come next? Specifically mentions Silver (2025) and Sutton. Silver, D., & Sutton, R. S. (2025). Welcome to the era of experience. Google AI, 1.
|