What brains do that AI still does not: neuroscientists' list
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.
What does David Eagleman mean by livewired? Show answer
The brain constantly reconfigures itself physically as it learns. Activity passing through the system changes the system.
What does Eagleman think AI most lacks? Show answer
Relevance and drive. Humans encode what matters to them, grounded in survival and personal goals.
What surprises machine learning people about the visual cortex? Show answer
It has more feedback connections than feed-forward ones. The brain projects expectations onto what it senses.
Why does Dileep George say more text will not close the gap? Show answer
Most of what we know about how the world works is never written down, so the information is not there.
What does the nail on the ceiling example show? Show answer
People answer by picturing the scene. Language steers a simulation built from sensory and motor experience.
What does Ted Gibson's evidence show about language and thought? Show answer
They are separate. The language network is silent during math and music, and patients without it can still reason.
How does Gibson describe what language models do well? Show answer
They do the form of language very well, but can follow a familiar form and miss the meaning.
What is the data gap Fei-Fei Li points to? Show answer
A child recognizes a cat after seeing perhaps three to ten. A model needs enormous amounts of data.
What two kinds of memory does Charan Ranganath describe? Show answer
Episodic memory records single events quickly. Semantic memory builds knowledge slowly. Together they allow fast change without erasing.
What does Terry Sejnowski consider the clearest difference? Show answer
A brain keeps thinking with no input. A language model goes blank when you stop talking to it.
What is Michael I. Jordan's objection to prediction systems? Show answer
They predict without evaluating risk or uncertainty. Real decisions need what-if questions and attention to causes.
Where do the experts disagree? Show answer
On the size of the gap. Sejnowski says models build internal models of meaning, and George says text cannot supply it.