Pocket Einstein: 10 Short Lessons in Artificial Intelligence and Robotics - Peter Bentley ****

Another handsome little hardback in the '10 short lessons' format that manages to pack in a surprising amount of information. AI is a subject where it's easy to get carried away with enthusiasm for the wonders of the subject, so we end up with much marvelling about too little substance - a trait that has dogged the AI profession leading a couple of 'winters' where it over-promised and under-delivered. Thankfully, Peter Bentley largely avoids this trap. Although he is certainly broadly positive about the topic, he does make some of the shortcomings clear.

The book is genuinely interesting and carries the reader along with a light touch that never betrays the author's academic background - it is a heartfelt compliment that this is a book by a professor that feels like it was written by a science writer. We get a good mix of the history with a brief explanation of how the technology works and a broad exploration of applications, both what has already been achieved and what may be possible.

Bentley manages to squeeze in robotics as well as AI, which was interesting as there have been a lot of books on AI of late, but not so many on robots - both what they do well and what they do badly. Here, as elsewhere, Bentley gives us a really good insight into the history and background. On the AI side, there is good mix of the different approaches, and a reasonable suppression of the inclination of some to over-worry about impending AI doom as the things become more intelligent than us and take over the world. As someone else said, it's more artificial unintelligence - brilliant at specific tasks, but lacking the core of what we would largely regard as intelligent ability. 

Inevitably, there are few things that could be done differently. The explanation of neural networks is just a bit too simple to really understand what's going on. And although Bentley doesn't whitewash the field, mentioning, for example, that it could be a long time before we accept autonomous cars, he doesn't entirely address the transparency problem - not understanding how an AI came to a decision, making it difficult to assess whether it made a mistake - or the ease with which these algorithms can be fooled. Also we are told, for example, of how brilliant AIs are now at image recognition, but Bentley doesn't really explore the way they require millions of examples to get it right, where a child can do it with a handful of examples.

There is always a danger with this kind of quick-and-simple introduction book that it feels like the author is talking down to the reader. This doesn't happen at all here - although it could comfortably be read by an interested teenager, it really does feel like a book that is addressed to adults, taking on a complex topic with a comfortable ease. Ideal for a commuting read or a relaxing evening listening to music provided by an AI system.

Hardback:    

Kindle:  

Using these links earns us commission at no cost to you

Review by Brian Clegg