NEW YORK, Jan 20: Recent strides in generative artificial intelligence (AI) shed light on the intricacies of how memories facilitate learning, recollection of past experiences, and the creation of novel scenarios for imagination and planning, as outlined in a new study by researchers at University College London (UCL).

The study, featured in Nature Human Behaviour, employs a generative neural network, a type of AI computational model, to replicate the learning and remembering processes of neural networks in the human brain concerning a sequence of events, each represented by a simple scene.

The model incorporates networks simulating the hippocampus and neocortex, exploring their interactions. Both components collaborate during memory functions, imagination, and planning. Lead author Eleanor Spens, a Ph.D. student at the UCL Institute of Cognitive Neuroscience, emphasizes the recent advancements in AI generative networks and their role in extracting information from experiences.

The research suggests that replaying memories during resting periods helps the brain identify patterns from past experiences, contributing to survival predictions crucial for avoiding danger or finding food.

The model underwent training by exposing it to 10,000 images of simple scenes. The hippocampal network swiftly encoded each scene during exposure and subsequently replayed them to train the generative neural network in the neocortex.

The neocortical network, through multiple layers of neurons, learned to reproduce scenes based on the activity of input neurons representing visual information. This led to the creation of efficient "conceptual" representations, capturing the meaning of scenes and facilitating both the recreation of old scenes and the generation of entirely new ones.

The study explains the gradual acquisition of conceptual knowledge by the neocortex and how, in conjunction with the hippocampus, it enables the mental reconstruction of events. Furthermore, the model illuminates the process of generating new events during imagination and future planning, shedding light on why memories often exhibit distortions with generalized features, known as "gist-like" distortions.

Senior author Professor Neil Burgess, affiliated with the UCL Institute of Cognitive Neuroscience and UCL Queen Square Institute of Neurology, clarifies that memories are not veridical records of the past but rather reconstructed experiences that combine meaning or gist with unique details, resulting in biases in recollection.