Shun’ichi Amari, “Brain, Mind, and AI”
| Author: Shun’ichi Amari |
This is a summary of the key points that stood out to me while reading Shun’ichi Amari’s “Brain, Mind, and AI.” Amari, a pioneer in AI research, reflects on the early days of deep learning and his contributions, which remain relatively unknown to many. With a background in mathematical engineering from the University of Tokyo, he has been at the forefront of AI research, utilizing advanced theories like statistical mechanics and information geometry. This book offers deep insights into the structure and function of the brain, as well as the progress of artificial intelligence.
Chapter 2: What is the Brain?
This chapter explores the neurophysiological structure and functions of the brain. The brain is a vast network of approximately 100 billion neurons, weighing about 1.4kg and occupying a volume of 1400cc.
1. Brainstem
The brainstem serves as the fundamental life-support system, controlling autonomic functions like breathing. It also regulates sleep and wakefulness, acting as a conduit for information traveling to and from the cerebrum.
2. Cerebellum
Although the cerebellum makes up only about 10% of the brain’s weight, it contains around 80 billion neurons, making it the most neuron-dense part of the brain. It primarily manages perception and motor control, working in concert with the cerebrum to coordinate precise movements. Over time, learned motor functions in the cerebrum are transferred to the cerebellum, where they become routine, allowing for faster and more efficient processing.
3. Cerebrum
The cerebrum is the most recently developed part of the brain. Comprising the cerebral cortex, basal ganglia, hippocampus, and amygdala, it consists of around 20 billion neurons. The cerebrum is responsible for higher mental functions such as cognition, thought, language, motor planning and control, and memory. The cerebral cortex, with a surface area of about 2000-2500 cm² (roughly the size of a newspaper page) and a thickness of 2-3 mm, is made up of six layers that form columnar structures.
How might Optimal Input in IT-MPC relate to the interaction between cerebrum and cerebellum in motor function?
Chapter 5: Interpreting the Brain through Mathematics
This chapter delves into the mechanisms of learning in the brain. Examination of neural circuits in the brain reveals that there is no observable ‘backpropagation of error’ in the brain as there is in artificial neural networks. So, is the learning mechanism of perceptrons completely different from that of the brain?
It’s not entirely different. The stochastic gradient descent method is an algorithmic expression of a principle in learning information processing. The brain might not have fully realized this principle, but that doesn’t mean the learning process in perceptrons is entirely unrelated to the brain’s structure.
Chapter 6: The History and Future of Artificial Intelligence
IBM has seriously entered the field of neuromorphic computing. The U.S. Department of Defense’s DARPA led the “SyNAPSE” program, aiming to create neural circuits on a chip. IBM successfully developed the “TrueNorth” chip as part of this initiative.
This chip contains 1 million neurons and 256 million synapses, yet consumes only 70mW of power. When these chips are interconnected, they form a system with 48 million neurons and 12 billion synapses, which is comparable to the brain of a mouse.
Reflections
As discussed in “Human vs. AI,” modern computing devices are fundamentally different from human cognitive processes. The most striking difference lies in their physical characteristics. It’s intriguing how such a small volume and surface area can address survival problems and sustain life experiences that no computer could solve. How can we scientifically and engineeringly approach this complexity and energy efficiency? The future of AI might hinge on hardware advancements. Just as mimicking a single function of the brain has profoundly impacted modern engineering and society, the full potential of science and imagination is something we can only truly grasp once we experience it.
Comments