Scientists develop AI model mimicking human memory and prediction, blurring the line between imagination and reality
AI is increasingly taking on tasks traditionally performed by humans, raising concerns about potential job displacement. Furthermore, scientists are now exploring ways to imbue AI systems with capabilities to think like the human brain, incorporating memories and events into their training.
According to a new study by two pHD students at University College London’s Institute of Cognitive Neuroscience – Eleanor Spens & Neil Burgess utilised a generative neural network where this AI computational model mimics the learning and memory processes of neural networks in the human brain. The simulation involves the representation of a sequence of events, each depicted by a basic scene.
To explain it in a layman’s language, imagine using a computer program, specifically one called a generative neural network, to imitate how our brains learn and remember things. This program focuses on two important parts of the brain: the hippocampus and the neocortex. These brain regions are like teammates, cooperating when we remember things, imagine scenarios, or plan activities. The goal is to understand how these brain parts interact by studying their simulated counterparts in the computer model.
Basically, there is going to be a time soon when people are going to have robots as their audience in a football match who will be analysing the scores and probability of both the teams winning; also recording the weak links of Ronaldo and Messi to defeat.
Aim of the study
The research looked closely at two important parts of the brain, the hippocampus and neocortex, which team up to help us remember things, imagine scenarios, and plan activities. The idea is to understand how these brain components work together.
The researcher pointed out that when we remember or imagine things, our brain uses stored information and combines it with expectations about what might have happened. It’s like creating a mental picture by blending what we know with what we think could have happened.
Inspiration behind the model
This study published in Nature Human behaviour, funded by Wellcome states, “our proposed model takes inspiration from recent advances in machine learning to capture many of the intriguing phenomena associated with episodic memory, its (re)constructive nature, its relationship to schemas, and consolidation, as well as aspects of imagination, inference and semantic memory.”
Memory replay & prediction (How the new AI model works)
Think of it like this: humans are really good at predicting things, which is essential for our survival (like avoiding danger or finding food). The AI networks in this study suggest that when we replay memories while taking a break, it helps our brains learn from past experiences and pick up on patterns to make better predictions.
To teach the AI, researchers showed 10,000 pictures of simple scenes. One part of the AI, the hippocampal network, quickly stored each scene as it was shown. Then, it replayed these scenes multiple times to teach another part of the AI, the generative neural network in the neocortex.
The neocortical network, like the AI’s brain, learned to process the visual information from each scene. It did this by passing the signals from thousands of input neurons (those receiving visual information) through smaller layers of neurons (some layers having as few as 20 neurons). Eventually, it created patterns of activity in its thousands of output neurons, predicting what the visual information should look like. So, it’s like the AI is learning from pictures to predict and recreate scenes, similar to how our brains learn from memories to make predictions.
Conclusion of the study
To explain the study in a simplified manner, the model helps us understand how our brain’s neocortex gradually learns general ideas and concepts. This, combined with the hippocampus, enables us to “relive” past experiences by mentally reconstructing them.
The model also sheds light on how we can create new scenarios when we imagine or plan for the future. It explains why our memories sometimes have generalisations, where unique details get mixed up and remembered more like features from previous experiences. So, in a way, it helps us grasp how our brain recalls and fabricates memories.
Neil Burgess, senior author of the study explained, “The way that memories are reconstructed, rather than being veridical records of the past, shows us how the meaning or gist of an experience is recombined with unique details, and how this can result in biases in how we remember things.”
Also Read: The rise of AI in HR: How technology is transforming the HR function
