What brains do that AI still does not: neuroscientists' list
October 6, 2026 · 23 min read
David Eagleman, Dileep George, Terry Sejnowski, Fei-Fei Li, Ted Gibson and Michael I. Jordan describe abilities such as rewiring from experience, feedback-driven perception, grounding in the physical world and thought without language that current models lack. The list shows where language models are a partial picture of intelligence and where research may go next.
The rules, numbers and steps worth keeping in front of you.
- Brains rewire themselves physically as they learn; trained model weights are fixed.
- Brains care about outcomes; relevance and drive decide what gets learned.
- Perception is explanation: the brain projects a model onto its senses, using feedback.
- Concepts are grounded in bodily experience; much world knowledge is never written down.
- Language models hold a model of the text world, not a world state.
- Thinking and language are separate brain systems; models master form more than meaning.
- Children learn from a few examples; models need enormous data.
- Humans pair fast episodic memory with slow semantic memory.
- A brain generates thoughts with no input; a model goes blank between prompts.
- Jordan: prediction is not decision; current systems lack error bars.
- Private, unrecorded thoughts are not in any training data.
- Disputed: Sejnowski says models build internal models of meaning; George says text lacks the information.
- Humans plus AI beat either alone where their strengths differ.
- Check meaning, ask what the system has seen, and take medical predictions to a doctor.
This is the short form. The full insight has every expert's reasoning, their own words and the sources.