Why AI systems don't learn and what to do about it: Lessons on autonomous learning from cognitive science

2026/03/16 by Emmanuel Dupoux, Yann LeCun, Jitendra Malik · 21 voices
#cs.AI

paper · pdf

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.

Citations

• Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model
• Quiet-STaR: Language Models Can Teach Themselves to Think Before Speaking
• Tree of Thoughts: Deliberate Problem Solving with Large Language Models
• DINOv2: Learning Robust Visual Features without Supervision
• Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
• Mastering Diverse Domains through World Models
• Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videos
• A Generalist Agent
• Training language models to follow instructions with human feedback
• Gradients are Not All You Need
• Grounded Language Learning Fast and Slow
• Mastering Atari, Go, chess and shogi by planning with a learned model
• Infant Statistical Learning
• Neural Architecture Search with Reinforcement Learning
• Concrete Problems in AI Safety
• Learning to learn by gradient descent by gradient descent
• Humans use directed and random exploration to solve the explore–exploit dilemma.
• The Neuroscience of Learning: Beyond the Hebbian Synapse
• Information landscapes and sensory ecology of the Cambrian Radiation
• Learning and development in neural networks: the importance of starting small
• Newborns' preferential tracking of face-like stimuli and its subsequent decline
• THE ROLE OF TUTORING IN PROBLEM SOLVING ^*
• I.—COMPUTING MACHINERY AND INTELLIGENCE
• BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

Cited by

• A solution to generalized learning from small training sets found in infant repeated visual experiences of individual objects

Discussions

• Why AI systems don't learn – On autonomous learning from cognitive science [hn, 205 points, 116 comments]
• Very good new paper from Dupoux, Lecun and Malik. Why current AI systems don't learn as well as toddlers and what to do about it. arxiv.org/abs/2603.15381 [bsky, 29 points, 2 comments]
• New position paper from Emmanuel Dupoux, Yann LeCun, Jitendra Malik: arxiv.org/pdf/2603.15381 CogSci insights, especially from evolution and early development, can be critical for a new AI paradigm th [bsky, 10 points, 0 comments]
• Why AI systems don't learn and what to do about it: Lessons on autonomous learning from cognitive science Emmanuel Dupoux, Yann LeCun, Jitendra Malik arxiv.org/abs/2603.15381 [bsky, 5 points, 1 comments]
• I disagree (although you’re right to point out the financial traps), the biggest bear case is that LLMs have fundamental limitations that will prevent them from ever achieving AGI. A better model will [bsky, 2 points, 2 comments]
• LeCun, Dupoux, and Malik argue current AI cannot actually learn autonomously and propose a dual-system architecture inspired by how biological organisms adapt. position paper worth a read https://arxi [bsky, 1 points, 0 comments]
• LLMs won’t take us to AGI and this paper explains why [lemmy, 1 points, 0 comments]
• https://bsky.app/profile/buzzing.cc.web.brid.gy/post/3mhd6pd2mrt62 [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science https://arxiv.org/abs/2603.15381 https://news.ycombinator.com/item?id=47418722 [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science https://arxiv.org/abs/2603.15381 [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science https://arxiv.org/abs/2603.15381 [comments] [7 points] [bsky, 0 points, 0 comments]
• LLMs sind eine Einbahnstraße und können weder zu einer AGI weiterentwickelt werden, noch können sie selber irgendetwas, ganz zu schweigen von "Bewusstsein erlangen". AGI is not imminent, and LLMs are [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science https://arxiv.org/abs/2603.15381 [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science https://arxiv.org/abs/2603.15381 (http://news.ycombinator.com/item?id=47418722) [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science https://arxiv.org/abs/2603.15381 (https://news.ycombinator.com/item?id=47418722) [bsky, 0 points, 0 comments]
• 今日読んだ論文 Why AI systems don't learn and what to do about it: Lessons on autonomous learning from cognitive science arxiv.org/abs/2603.15381 観察による学習（自己教師あり学習）と行動による学習（強化学習など）を繋ぐメタコントローラとしてSystem Mなるものを考 [bsky, 0 points, 0 comments]
• Why AI systems don't learn and what to do about it arxiv.org/abs/2603.15381 [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science https://arxiv.org/abs/2603.15381 (https://news.ycombinator.com/item?id=47418722) [bsky, 0 points, 0 comments]
• Why AI systems don't learn – On autonomous learning from cognitive science https:// arxiv.org/abs/2603.15381 # ai # arxiv # science [mastodon, 0 points, 0 comments]
• [2603.15381] Why AI systems don't learn and what to do about it: Lessons on autonomous learning from cognitive science — A clear, cognitive-science-informed take on why today’s AI struggles with auton [bsky, 0 points, 0 comments]

Related

• What Neuroscience Can Teach AI About Learning in Continuously Changing Environments
• Discovering Latent States for Model Learning: Applying Sensorimotor Contingencies Theory and Predictive Processing to Model Context
• Bridging the Gap: Toward Cognitive Autonomy in Artificial Intelligence
• Leveraging insights from neuroscience to build adaptive artificial intelligence
• World Models in Artificial Intelligence: Sensing, Learning, and Reasoning Like a Child