Grokking the Brain: What AI Reveals About How You Learn • Barbara Oakley  • GOTO Copenhagen 2026

Grokking the Brain: What AI Reveals About How You Learn • Barbara Oakley • GOTO Copenhagen 2026

Title history
  • Grokking the Brain: What AI Reveals About How You Learn • Barbara Oakley • GOTO Copenhagen 2026 (PSI archive)
  • Grokking the Brain: What AI Reveals About How You Learn • Barbara Oakley • GOTO Copenhagen 2026 — observed on 20261109-1422 UTC (YouTube)
🎙 Barbara Oakley 👥 1.1M 📅 November 9, 2026 ⏱ 49 min 👁 83 📄 expert opinion 🧭 2026-11-09
Available in: English (current) Français

Keywords

grokkingworking memorybasal gangliahippocampusdopamine

Summary

Barbara Oakley presents a talk at GOTO Copenhagen 2026 exploring how artificial intelligence, particularly large language models, provides insights into human brain function and learning. She begins by citing an Ernst Strüngmann Forum on cognitive offloading, where neuroscientists, psychologists, and edtech experts gathered to discuss the impact of AI on the brain. She then explains the basic neuron structure and its analogy to transformers, emphasizing the flow of information. The talk distinguishes between two learning systems: the conscious declarative pathway (hippocampus) and the automatic habitual pathway (basal ganglia), linking them to the concepts of slow and fast thinking from Kahneman. She discusses the phenomenon of ‘grokking’ in over-trained AI models, which parallels the human ability to form deep patterns and intuition. She also addresses working memory limitations shared by humans and AI, referencing Anthropic’s research on a hypothetical ‘J-space’ and ‘workspace’ in LLMs. The talk highlights the decline in PISA scores and the Flynn effect, attributing it to the offloading of memory to external tools. Finally, she proposes that curiosity and dopamine are key to maintaining learning and cognitive health, offering practical advice for using AI while preserving one’s own cognitive abilities.

192 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value of the information lies in its synthesis of neuroscience, AI, and education, providing actionable insights for developers and learners. Oakley presents a compelling argument that storing knowledge in long-term memory is essential for critical thinking and intuition, and that overreliance on AI can weaken these abilities. She supports her claims with references to research (e.g., Anthropic’s Gemini paper, Nature articles on video games, PISA scores) and personal anecdotes, making the argument engaging and accessible. The argumentation is solid, though some analogies (e.g., LLM’s workspace to human working memory) are presented as genuine parallels while being more metaphorical.

Scientific Rigor, Source Quality, Title Accuracy

Oakley demonstrates scientific rigor by citing specific research, such as the ‘Attention is All You Need’ paper, Neuroscientific studies on the basal ganglia and hippocampus, and the Ernst Strüngmann Forum. She uses empirical evidence where available, and acknowledges when ideas are speculative. The title is apt and captures the central theme. The sources are largely credible, including personal website, conference materials, and referenced papers. The talk is well-structured and the content aligns with the title.

189 words

Title / Content Match

The title accurately reflects the content: the talk bridges AI concepts (grokking, working memory in LLMs) and human brain learning, with a focus on how learning works.

Quality & Reliability

8/10

The talk is delivered by a well-known professor of engineering and learning expert, based on established neuroscience research and recent AI findings. However, some analogies between AI and brain are speculative and not fully empirically validated. Overall, the information is reliable and well-sourced, but should be interpreted with caution.

Chapters

Cited Sources

Concurring Sources

Dissenting Sources

  • Nelson Cowan's working memory research — Oakley had a disagreement with Cowan about the reason for the 4-6 chunk limit in working memory; she speculates physical constraints, while Cowan's model attributes it to attentional focus.

External References

Contribution & Novelties

The talk offers a novel cross-disciplinary perspective by comparing the ‘grokking’ phenomenon in AI with human learning, and by linking AI model limitations (working memory constraints) to human cognitive architecture. It provides a clear synthesis of existing research and implies practical recommendations for lifelong learning in the age of AI.

Pour aller plus loin :

89 words

Radar Profile

The radar profile shows high scores in quality and reliability, with moderate technical depth. This indicates a content that is credible and well-sourced, but accessible to a general technical audience rather than requiring specialized expertise.

Reliability 8/10