Generative AI helps explain human memory and imagination

According to a new study led by UCL researchers, recent advances in generative AI help explain how memories allow us to know the world, relive old experiences and construct completely new experiences for the world. imagination and planning.

The study, published in Human behavioruses an AI computer model, known as a generative neural network, to simulate how the brain’s neural networks learn and remember a series of events (each represented by a simple scene).

The model included networks representing the hippocampus and neocortex, to study how they interact. The two parts of the brain are known to work together during memory, imagination and planning.

Main author, Ph.D. Eleanor Spens (UCL Institute of Cognitive Neuroscience), said: “Recent advances in generative networks used in AI show how information can be extracted from experience so that we can both remember a specific experience and flexibly imagine what new experiences might look like. …We think that memory involves imagining the past on the basis of concepts, combining certain stored details with our expectations of what might have happened.

Humans need to make predictions to survive (such as to avoid danger or find food), and AI networks suggest how, when we replay memories at rest, it helps our brains pick up patterns of past experiences that can be used to create these events. predictions.

The researchers presented the model with 10,000 images of simple scenes. The hippocampal network rapidly encoded each scene as he experienced it. It then replayed the scenes over and over again to train the neocortex’s generative neural network.

The neocortical network has learned to transmit the activity of thousands of input neurons (neurons that receive visual information) representing each scene through smaller intermediate layers of neurons (the smallest containing only 20 neurons), to recreate scenes as activity patterns in the thousands. of output neurons (neurons that predict visual information).

This caused the neocortical network to learn highly efficient “conceptual” representations of scenes that capture their meaning (e.g. the arrangement of walls and objects), allowing both the recreation of old scenes and the generation of completely new scenes. .

Therefore, the hippocampus was able to encode the meaning of new scenes presented to it, rather than having to encode every detail, allowing it to focus its resources on encoding unique features that the neocortex could not. reproduce, like new types of objects.

The model explains how the neocortex slowly acquires conceptual knowledge and how, together with the hippocampus, this allows us to “relive” events by reconstructing them in our minds.

The model also explains how new events can be generated during imagination and planning for the future, and why existing memories often contain “gist-like” distortions, in which unique features are generalized and remembered. as more closely resembling those of previous events.

Lead author Professor Neil Burgess (UCL Institute of Cognitive Neuroscience and UCL Queen Square Institute of Neurology) explained: “The way memories are reconstructed, rather than being truthful records of the past, shows us how the meaning or gist of an experience is recombined with unique details, and how this can lead to biases in the way we remember things. »