The Myth of Artificial Intelligence: Why Computers Can’t Think the Way We Do

“Humm I don’t know that one…”We have all heard it from Alexa. Given the level of investment that has gone into such projects the result is more than disappointing. Having been a keen (amateur) follower of AI, I avidly soaked up Nick Bostrom’s book “Superintelligence” recommending it to friends. But lately I have grown suspicious of the hype. What is AI and machine learning after all? As a Go player for many decades, I was astonished at the power of AlphaGo. The “divine move” in game four against Lee Sedol astonished the Go playing community. A sign of true creativity?Of course not. And Erick J. Larson spells it all out beautifully. It is a brave book. It meticulously dismantles all the hype and, in easy to understand words. tells us the true story of AI and its evolution from Turing to today. Clear, well-paced, insightful and fun to read. One of the best books on the subject I have ever read. And that includes Bostrom’s work. Larson assures us that AI holds nothing to be scared of – and I think he’s right.

…but the main argument is based on the claim that no-one knows how to implement abductive inference which is more a statement of opinion about currently published research than an eternal truth. So the book is best viewed as an argument about why general AI is not an inevitable consequence of current mainstream approaches and a call to arms for theorists to come up with some testable hypotheses to spur progress rather than an argument that general AI is somehow impossible (as implied by the title).

Amidst all the claims and excitement about AI, this is a well-written, well-argued case for caution. At first I thought it might be a little lightweight, but once it gets going, it's both interesting and convincing. The author describes the narrowness of much of AI and how escaping that narrowness - being brilliant at chess say - is extremely difficult. Much of this is via a discussion of logic, and the problems we face trying to program the necessary types of logic and then via a discussion what human intelligence is - and essentially how we have no idea. I highly recommend it for anyone interested in the subject and in how AI might develop in the future.

(First, I won't be a verified purchaser because I got it on my wife's kindle.)This is an important book, insofar as it highlights that current progress in Deep learning and other commercially driven AI is not going to result in AGI, simply by applying more data and more computing power. It explains (fairly) well why deductive and inductive reasoning won't crack it. It introduces the notion of abductive reasoning to explain what's missing. But I found the book unsatisfactory, because:1. It's quite repetitive (a lot of the same points are made multiple times)2. It cries out for a more considered review of what IS going on outside the field of deep learning etc. (for which, one must engage with philosophy as this is where a lot of the ground work is going on).3. The analysis of abductive reasoning is a little "wooh, this is black-box magic". Yes it's hard (impossible) for us to introspect to see ourselves thinking, and therefore also hard (but not impossible) to see how we do it. (We've more or less mastered deductive logic, without anyone ever peering inside or feeling themselves being deductive.)4. While it's plainly true that 'more of the same' won't get us there, that doesn't mean we won't get there - and in a sense, this is what we should expect. There will need to be more sparks of insight, both to understand how intelligent beings (e.g. humans) think, and then also how to replicate that. This was always so, and will remain so.For one example of this work, in the world of philosophy rather than the technosphere of commercial AI, see "Fodor’s riddle of abduction" by Matthew J. Rellihan. (Philos Stud (2009) 144:313–338 DOI 10.1007/s11098-008-9212-6). This isn't easy stuff, but the key point is that theorists aren't being as wooden as EL makes out. There will be break throughs.To be fair to EL, he sets a limited objective and achieves it. I just wish he'd set a more ambitious goal for himself.