UCL scientists utilized generative AI to model brain functions, uncovering how memories are formed, replayed, and utilized for creativity. The study highlights the reconstructive and predictive nature of memory, offering brand-new perspectives on human cognition. Credit: SciTechDaily.comA UCL research study utilizing AI models advances our understanding of memory, demonstrating how the brain rebuilds previous occasions and pictures new scenarios.Recent advances in generative AI aid to explain how memories allow us to discover about the world, re-live old experiences and construct absolutely new experiences for imagination and planning, according to a new research study by UCL researchers.AI Models Mimicking Brain FunctionsThe research study, released in Nature Human Behaviour and funded by Wellcome, utilizes an AI computational model– understood as a generative neural network– to replicate how neural networks in the brain gain from and keep in mind a series of events (each one represented by an easy scene). The model included networks representing the hippocampus and neocortex, to investigate how they connect. Both parts of the brain are known to interact during memory, creativity, and planning.Lead author, PhD student Eleanor Spens (UCL Institute of Cognitive Neuroscience), stated: “Recent advances in the generative networks utilized in AI reveal how information can be extracted from experience so that we can both recollect a particular experience and also flexibly imagine what brand-new experiences might be like.” We consider keeping in mind as imagining the previous based upon principles, combining some kept details with our expectations about what might have occurred.” Memory Replay and PredictionHumans require to make predictions to make it through (e.g. to prevent danger or to discover food), and the AI networks suggest how, when we replay memories while resting, it assists our brains detect patterns from previous experiences that can be used to make these predictions.Researchers played 10,000 images of basic scenes to the design. The hippocampal network rapidly encoded each scene as it was experienced. It then replayed the scenes over and over again to train the generative neural network in the neocortex.The neocortical network learned to pass the activity of the countless input neurons (neurons that get visual information) representing each scene through smaller sized intermediate layers of nerve cells (the tiniest consisting of just 20 nerve cells), to recreate the scenes as patterns of activity in its thousands of output nerve cells (neurons that anticipate the visual info). Implications of the StudyThis triggered the neocortical network to discover highly efficient “conceptual” representations of the scenes that record their meaning (e.g. the arrangements of objects and walls)– allowing both the recreation of old scenes and the generation of totally brand-new ones.Consequently, the hippocampus had the ability to encode the significance of brand-new scenes presented to it, instead of having to encode every single detail, allowing it to focus resources on encoding unique functions that the neocortex couldnt reproduce– such as new kinds of objects.The design describes how the neocortex slowly gets conceptual understanding and how, together with the hippocampus, this allows us to “re-experience” events by reconstructing them in our minds.The model likewise discusses how brand-new events can be generated throughout imagination and preparation for the future, and why existing memories often contain “gist-like” distortions– in which distinct features are generalized and kept in mind as more like the features in previous events. Senior author, Professor Neil Burgess (UCL Institute of Cognitive Neuroscience and UCL Queen Square Institute of Neurology), described: “The manner in which memories are re-constructed, instead of being veridical records of the past, reveals us how the significance or gist of an experience is recombined with distinct details, and how this can lead to biases in how we remember things.” Reference: “A Generative Model of Memory Construction and Consolidation” 19 January 2024, Nature Human Behaviour.DOI: 10.1038/ s41562-023-01799-z.
