Recent research from UCL highlights that advancements in generative AI provide insights into the role of memories in learning, recalling past experiences, and creating new imaginative scenarios for planning.

AI can simulate cognitive processes in the brain

In a study published in Nature Human Behaviour, researchers utilized a generative neural network, an AI computational model, to simulate the learning and memory processes of neural networks in the brain. The model included representations of the hippocampus and neocortex regions, exploring their interaction in memory-related functions such as memory, imagination, and planning. The findings shed light on the collaborative nature of these brain regions during cognitive processes.

The lead author, Eleanor Spens, a Ph.D. student at UCL Institute of Cognitive Neuroscience, discusses recent advancements in AI generative networks that extract information from experience. This enables the recollection of specific experiences and the flexible imagination of new ones. The process is described as remembering through imagination of the past based on concepts, combining stored details with expectations about what might have occurred.

Humans rely on predictions for survival, such as avoiding danger or finding food. AI networks propose that when we replay memories during rest, our brains discern patterns from past experiences, aiding in making accurate predictions.

Hippocampus encodes scenes without recalling old scenes

Researchers presented 10,000 simple scenes to a model, with the hippocampal network encoding and replaying scenes. This trained the generative neural network in the neocortex, which learned to pass visual information through intermediate layers, recreating scenes as patterns of activity in thousands of output neurons.

The neocortical network efficiently learns conceptual representations of scenes, capturing their meaning and enabling the recreation of old scenes or generation of new ones. This process allows the hippocampus to encode scene meaning without focusing on every detail, emphasizing unique features. The model elucidates neocortex’s gradual acquisition of conceptual knowledge and its collaboration with the hippocampus for mental event reconstruction.

Professor Neil Burgess, a senior author from UCL, explains that memories aren’t accurate records but reconstructed with meaning and unique details. This process leads to biases in memory recall, indicating the influence of experience on recollection.