Skip to main content

Artificial Intelligence - Melanie Mitchell *****

As Melanie Mitchell makes plain, humans have limitations in their visual abilities, typified by optical illusions, but artificial intelligence (AI) struggles at a much deeper level with recognising what's going on in images. Similarly in some ways, the visual appearance of this book misleads. It's worryingly fat and bears the ascetic light blue cover of the Pelican series, which since my childhood have been markers of books that were worthy but have rarely been readable. This, however, is an excellent book, giving a clear picture of how many AI systems go about their business and the huge problems designers of such systems face.

Not only does Mitchell explain the main approaches clearly, her account is readable and engaging. I read a lot of popular science books, and it's rare that I keep wanting to go back to one when I'm not scheduled to be reading it - this is one of those rare examples.

We discover how AI researchers have achieved the apparently remarkable abilities of, for example, the Go champion AlphaGo, or the Jeopardy! playing Watson. In each case these systems are tightly designed for a particular purpose and arguably have no intelligence in the broad sense.  As for what's probably the most impressively broad AI application of modern times, self-driving cars, Mitchell emphasises how limited they truly are. Like so many AI applications, the hype far exceeds the reality - when companies and individuals talk of self-driving cars being commonplace in a few years' time, it's quite clear that this could only be the case in a tightly controlled environment.

One example, that Mitchell explores in considerable detail are so-called adversarial attacks, a particularly AI form of hacking where, for example, those in the know can make changes to images that are invisible to the human eye but that force an AI system to interpret what they are seeing as something totally different. It's a sobering thought that, for example, by simply applying a small sticker to a stop sign on the road - unnoticeable to a human driver - an adversarial attacker can turn the sign into a speed limit sign as far as an AI system is concerned, with potentially fatal consequences.

Don't get me wrong, Mitchell, a professor of computer science who has specialised in AI, is no AI luddite. But unlike many of the enthusiasts in the field (or, for that matter, those who are terrified AI is about to take over the world), she is able to give us a realistic, balanced view, showing us just how far AI has to go to come close to the more general abilities humans make use of all the time even in simple tasks. AI does a great job, for example, in something like Siri or Google Translate or unlocking a phone with a face - but AI systems still have no concept of, for example, understanding (as opposed to recognising) what is in an image. Mitchell makes it clear that where systems learn from large amounts of data, it is usually impossible to uncover how they are making decisions (which makes the EU's law requiring transparent AI decisions pretty much impossible to implement), so we really shouldn't trust them with important outcomes as they could easily be basing their outcomes on totally irrelevant inputs.

Apart from occasionally finding the explanations of the workings of types of neural network a little hard to follow, the only thing that made me raise an eyebrow was being told that Marvin Minsky 'coined the phrase "suitcase word"' - I would have thought 'derived* the phrase from Lewis Carroll's term "portmanteau word"' would have been closer to reality.

There have been good books on the basics of AI already, and excellent ones on the problems that 'deep learning' and big data systems throw up. But without a doubt, Mitchell's book sets a new standard in giving an understanding of what's possible and how difficult it is to go further. It should be read by every journalist, PR person and politician before they pump out yet more hype on the AI future. Recommended.

* Polite term
Hardback:   

Kindle:  
Using these links earns us commission at no cost to you
Labels:

Comments

Post a Comment

Popular posts from this blog

God: the Science, the Evidence - Michel-Yves Bolloré and Olivier Bonnassies ***

This is, to say the least, an oddity, but a fascinating one. A translation of a French bestseller, it aims to put forward an examination of the scientific evidence for the existence of a deity… and various other things, as this is a very oddly structured book (more on that in a moment). In The God Delusion , Richard Dawkins suggested that we should treat the existence of God as a scientific claim, which is exactly what the authors do reasonably well in the main part of the book. They argue that three pieces of scientific evidence in particular are supportive of the existence of a (generic) creator of the universe. These are that the universe had a beginning, the fine tuning of natural constants and the unlikeliness of life.  To support their evidence, Bolloré and Bonnassies give a reasonable introduction to thermodynamics and cosmology. They suggest that the expected heat death of the universe implies a beginning (for good thermodynamic reasons), and rightly give the impression tha...
Post a Comment
Read more

Humble Pi - Matt Parker ****

Matt Parker had me thoroughly enjoying this collection of situations where maths and numbers go wrong in everyday life. I think the book's title is a little weak - 'Humble Pi' doesn't really convey what it's about, but that subtitle 'a comedy of maths errors' is far more informative. With his delightful conversational style, honed in his stand-up maths shows, it feels as if Parker is a friend down the pub, relating the story of some technical disaster driven by maths and computing, or regaling us with a numerical cock-up. These range from the spectacular - wobbling and collapsing bridges, for example - to the small but beautifully formed, such as Excel's rounding errors. Sometimes it's Parker's little asides that are particularly attractive. I loved his rant on why phone numbers aren't numbers at all (would it be meaningful for someone to ask you what half your phone number is?). We discover the trials and tribulations of getting cal...
Post a Comment
Read more

Quantum 2.0 - Paul Davies ****

Unlike the general theory of relativity or cosmology, quantum physics is an aspect of physics that has had a huge impact on everyday lives, particularly through the deployment of electronics, but also, for example, where superconductivity has led to practical applications. But when Paul Davies is talking about version 2.0, he is specifically describing quantum information, where quantum particles and systems are used in information technology. This obviously includes quantum computers, but Davies also brings in, for example, the potential for quantum AI technology. Quantum computers have been discussed for decades - algorithms had already been written for them as early as the 1990s - but it's only now that they are starting to become usable devices, not at the personal level but in servers. In his usual approachable style, Davies gives us four chapters bringing us up to speed on quantum basics, but then brings in quantum computing. After this we don't get solid quantum informat...
Post a Comment
Read more