A recent study conducted by researchers from the Institute for Basic Science (IBS) has uncovered a remarkable similarity between the memory processing of artificial intelligence (AI) models and the human brain’s hippocampus. The findings shed light on the process of memory consolidation in AI systems, offering valuable insights into the pursuit of Artificial General Intelligence (AGI).
The research team, consisting of interdisciplinary experts from the Center for Cognition and Sociality and the Data Science Group within the IBS, focused on understanding how AI models learn and remember information. They specifically examined the role of memory consolidation, where short-term memories are transformed into long-term ones.
To explore this phenomenon, the researchers drew inspiration from the NMDA receptor in the hippocampus of the human brain. This receptor acts as a smart door, facilitating learning and memory formation. They discovered that the Transformer model, a key component of powerful AI systems, appeared to employ a gatekeeping process similar to the NMDA receptor.
Further investigation revealed that mimicking the NMDA receptor’s gating process in the Transformer model could enhance its memory consolidation. By tweaking the model’s parameters to imitate the magnesium ion’s role as a gatekeeper, the researchers observed improved long-term memory in the AI system. This breakthrough discovery suggests that established knowledge in neuroscience can explain how AI models learn.
C. Justin LEE, a neuroscientist director at the institute, emphasized the significance of this research, stating, This research makes a crucial step in advancing AI and neuroscience. It allows us to delve deeper into the brain’s operating principles and develop more advanced AI systems based on these insights.
One important aspect that sets this study apart is its incorporation of brain-inspired nonlinearity into AI design. By simulating human-like memory consolidation, researchers have achieved significant progress in replicating cognitive mechanisms in AI models. This convergence between human cognition and AI development holds promise not only for creating cost-effective and high-performance AI systems but also for providing invaluable insights into the workings of the human brain.
Cha Meeyoung, a data scientist from KAIST, praised the study’s approach, saying, What sets this study apart is its initiative to incorporate brain-inspired nonlinearity into an AI construct, signifying a significant advancement in simulating human-like memory consolidation. The convergence of human cognitive mechanisms and AI design not only holds promise for creating low-cost, high-performance AI systems but also provides valuable insights into the workings of the brain through AI models.
The implications of this research extend beyond advancements in AI. The human brain is known for its energy efficiency, while large AI models often require substantial resources. The study’s findings open up new possibilities for developing low-cost, high-performance AI systems that can learn and remember information similar to humans. This has the potential to revolutionize the field of AI and pave the way for more efficient and intelligent technologies.
As researchers continue to bridge the gap between neuroscience and AI, further breakthroughs await. By unraveling the mysteries of the brain’s memory processing and applying these principles to AI models, scientists are pushing the boundaries of human-like intelligence in machines. The study’s findings not only contribute to the pursuit of AGI but also deepen our understanding of the brain’s intricate mechanisms.
