“We will never have true AI without first understanding the brain.” Jeff Hawkins, author of A Thousand Brains
Jeff Hawkins, one of the innovative progenitors of handheld computing (PalmPilot and Treo) and the founder of Numenta (a neuroscience research company), has devoted many years to the self-study of neurosciences as well as how it relates to artificial intelligence (specifically artificial general intelligence). He also founded the Redwood Center for Theoretical Neurosciences.
His previous book, On Intelligence, was an impressive scientific journey into intelligence as a memory-prediction system for experiences and therefore vastly different than a computer. He compared a human brain’s neuron to two machine neurons: a traditional neural network neuron and his version of a more sophisticated one in the form of hierarchical temporal memory (HTM) cells.
Almost 20 years later, he takes this earlier HTM concept to a higher plane and has truly established himself as both a formidable teacher and an ardent student of intelligence. His latest book, A Thousand Brains: A New Theory of Intelligence, is framed in the construct of how the neocortex represents object compositionally, object behaviors, and higher level concepts.
The main premise is that every part of the neocortex learns complete models of objects and concepts, and that there are many such models of each object distributed throughout the neocortex that have long range connections (hence the title of the book). The three sections of the book: A New Understanding of the Brain, Machine Intelligence, and Human Intelligence, are well organized and written.
Despite the details of the neuroscience aspects, the content is easily understandable for those without a neuroscience background. I particularly enjoyed chapters 7 (The Thousand Brains Theory of Intelligence), 8 (Why There Is No “I” in AI), 10 (The Future of Machine Intelligence), and 14 (Merging Brains and Machines). He argues that present day AI cannot continually learn new things without forgetting previous observations and that it cannot structure knowledge using reference frames (knowledge relative to our point of view).
Hawkins becomes appropriately philosophical towards the end of the book. It is overall a very enjoyable book to read with many insightful passages. Perhaps the only constructive criticism is that a few more key figures in the latter half of the book would have been very helpful as the concepts become more difficult to visualize (realizing that these figures would be difficult to design).
Being an entrepreneur and a neuroscientist does have drawbacks, and he has been criticized by the AI intelligentsia (even though most, if not all, of these scientists have intelligence and AI figured out themselves). Some AI experts do, however, agree with his combined machine learning and human neuroscience approach to intelligence to go beyond metrics like the Turing test or image interpretation.
Similar to how man eventually invented flight by not mimicking birds but rather understanding aerodynamics, Jeff Hawkins is probably correct in that we need to understand the brain and intelligence at the same time we improve on machine and deep learning. Perhaps a “thousand brains” of AI experts and neuroscientists are necessary to decipher the enigmas of intelligence, both human and machine, and ultimately, even reasons for preservation of our knowledge and our existence.