At the intersection of neuroscience and AI
January 7th, 2025
Artificial intelligence algorithms have long drawn inspiration from our latest understanding of how the human brain works. Deep neural networks, a component of some of the most prominent AI applications these days spanning GenAI, Reinforcement Learning and PredictiveAI, were inspired by how billions of interconnected neurons fire and transmit electrical activity in the human brain. Ironically, neural networks are now a technique broadly utilised in neuroscience to help us decode how the human brain works!
Acquiring knowledge on how the human brain works, in parallel with knowledge about how different AI algorithms work, can help us achieve two interesting objectives:
(1) develop a richer perspective of human experience; and
(2) gain inspiration on how to further improve AI algorithms
A rarely discussed aspect of neuroscience caught my attention a few years ago when Iain McGilchrist appeared on Sam Harris’ Making Sense podcast. The discussion focussed on Iain’s book, The Master and His Emissary, exploring how the right and left hemispheres of the brain are divided and profoundly differ. Here is a link to the episode which I highly recommend listening to. In a nutshell, it discusses how the left brain tends to have a narrow focus and excel at well-defined tasks, whilst the right brain is required to retain a holistic perspective. More specifically, the left brain is more analytical, focused on details, categorization, language, and abstract thinking. The right brain on the other hand is concerned with the big picture, the connections between things, and the direct experience of the world. Whilst both brains fulfil important functions, McGilchrist suggests that a person would be more severely impaired if they lost access to the right hemisphere, which he refers to as the "master" in his analogy.
This idea is especially intriguing when considering the current state of AI, particularly the quest for Artificial General Intelligence (AGI). Most AI algorithms today are designed to function like the left brain: narrow, analytical, and task-oriented. However, when Google released its landmark paper Attention is All You Need, which introduced the transformer architecture, significant advancements were made in GenAI. The key innovation in transformers is the attention mechanism, which allows the model to consider all relevant context simultaneously, rather than focusing on a smaller subset of information.
This new architecture can be seen as somewhat analogous to the division of labor between the right and left hemispheres of the brain. In the transformer model, the attention step (which gathers and integrates context) resembles the role of the right hemisphere, while the feed-forward step (which processes and makes predictions based on the context) is more like the left hemisphere, focused on specifics and predictions.
I believe that deepening our understanding of how the brain's hemispheres work in tandem could provide valuable insights for improving AI. If AI algorithms could emulate the holistic judgment and context-awareness of the right hemisphere, they might one day exhibit greater common sense and decision-making abilities.
If this topic has piqued your interest in neuroscience, here are some accessible books to explore:
The Brain: The Story of You by David Eagleman
Lifewired: The Inside Story of the Ever-Changing Brain by David Eagleman
The Master and His Emissary: The Divided Brain and the Making of the Western World by Iain McGilchrist
How Emotions Are Made: The Secret Life of the Brain by Lisa Feldman Barrett