Jeff Hawkins, founder of Palm Computing, just published a book: On Intelligence. Subtitled: How a New Understanding of the Brain will Lead to the Creation of Truly Intelligent Machines. Just completed this in my recent travels. This book really struck home to me because it paralleled my own exploration of the possibilities of Artificial Intelligence. From being inspired by the famous Crick article in Scientific American in 1979, where Crick basically admitted we knew very little about the brain, to building rule-based expert systems, to using neural nets to plumb hard pattern recognition problems. This always seemed like great progress to me ... from logic to networks that were at least inspired by brain structures. But all of this work, some very valuable for solving problems, never really got much closer to the goal of producing real 'intelligence'. I think he is right in thinking there is much to learn from the form and function of the brain.
Hawkins has the resources, and he has set up the Redwood Neuroscience Institute. Which is worth exploring, just scanning its publications gives me an impression of the difficulty of the problem, linking low level functionality to high level process. Its like being given the periodic table of elements and being asked to prepare the perfect omelette from sample chemicals. Basic chemistry is still missing and even the cookbook is missing. I look forward to following RNI's progress.
I particularly like the analogy to forecasting ... Hawkin's suggestion that the brain is a forecaster .. continually predicting future events ... aided by a huge amount of memory. Its not always right, but it learns and adapts based on the mistakes it makes and resulting feedback from its evironment.
Hawkins makes the case that with study of physiology and functionality of the neurocortex we are starting to make inroads into defining and ultimately leveraging machine intelligence. I think he is right but I am not as optimistic about how long it will take. I suggest we are mutliple decades rather than just a decade away from real progress. I am still battle-scarred with predictions in the 90s about how artificial neural networks would provide the ability to learn our way to intelligence. We learned a lot, but there was much hype as well. I hope I am wrong, then we can start working on the ethics of AI.
This book is very readable, largely non-technical though the latter parts of it contain some detail about the functioning of the neurocortex and may take some close thought. Also contains fascinating historical context. I strongly recommend this book if you have interest in this area.
Further, an insightful review from Corante. ... With resume bullets like inventor of the PalmPilot and CTO of PalmOne, a popular-science book about the future of computing certainly seems like an obvious choice for Hawkins. But as soon as you open up the sharp, electric-blue dust-cover, you’ll realize On Intelligence was probably the last thing you’d expect from a Silicon-Valley techie. Missing are detailed technology roadmaps and ethereal speculations about fantastical improbable futures. Instead, On Intelligence adeptly intertwines lay-English summaries of decades of research from neurophysiology, computer science, cognitive psychology, and even includes some well-placed philosophical sidebars that mesh into an approachable and well-written narrative addressing the plausible future of computing ...
I am a 68 year old graduate student of AI at the University of Minnesota (formerly a mathematician). I recently devoured Mr. Hawkins book "On intelligence" hoping to gain some new insight on my Ph.D. thesis topic (concerning the parsing and understanding of natural language). Instead, I came away with what I think are interesting ideas about how to make the entire internet "intelligent," by using concepts outlined in this great book.
If anyone can put me in contact with Mr. Hawkins, I would appreciate it.
Posted by: Willy Chaplin | November 15, 2004 at 11:15 AM