Skip to main content

A Citizen's Guide to Artificial Intelligence - John Zerilli et al ****

The cover of this book set off a couple of alarm bells. Not only does that 'Citizen's Guide' part of the title raise the spectre of a pompous book-length moan, the list of seven authors gives the feel of a thesis written by committee. It was a real pleasure, then, to discover that this is actually a very good book.

I ought to say straight away what it isn't - despite that title, it isn't a book written in a style that's necessarily ideal for a general audience. Although the approach is often surprisingly warm and human, it is an academic piece of writing. As a result, in places it's a bit of a trudge to get through it. Despite this, though, the topic is important enough - and, to be fair, the way it is approached is good enough - that it deserves to be widely read.

John Zerilli et al give an effective, very balanced exploration of artificial intelligence. Although not structured as such, it's a SWOT analysis, giving us the strengths, weaknesses, opportunities and threats of AI. Of course we get the concerns that have been repeatedly raised in books such as Weapons of Math Destruction that artificial intelligence and big data can result in opaque decision making that influences our lives and that can have unintentional biases baked into the systems. But we also see the potential benefits of AI and rather than just bemoaning the dangers, there is real consideration of the checks and balances that can be put in place to make use of it without suffering from its unwanted side-effects.

Some aspects really jump out at the reader, for me particularly around what is and isn't possible as far as transparency goes, and making the very important point that we should not judge AI in isolation but have to weigh it up against the lack of transparency and biases that human decision makers also have. Similarly, for example, when talking about self-driving cars, there is a discussion of the challenging aspect where a famous ethical puzzle, the trolley problem, is brought to life: how should a car judge priorities if, say, it had the choice of saving the driver or a cyclist, or has to choose between the life of the driver or a group of children on the pavement. As Zerilli et al point out, we all might favour saving the children in principle, but would you buy a car that is prepared to intentionally kill the driver?

The book's academic origin comes through in the care with which it drills down into things we tend to take for granted. So, for instance, there is a box explaining the difference between 'appeal' and 'review' in responding to legal and governmental decisions that some considers incorrect. That particular example was quite interesting, though overall this approach does contribute to the parts of the book that are quite hard going.

Despite being relatively heavyweight reading, this is a different take on AI to any I've read before. It focusses on how AI will affect our lives and how we as a society should react to it. At the very least it should be recommended reading for those in government who are having to make decisions in this area - and deserves a significantly wider readership too.



Using these links earns us commission at no cost to you
Review by Brian Clegg


Popular posts from this blog

Grace Lindsay - Four Way Interview

Grace Lindsay is a computational neuroscientist currently based at University College, London. She completed her PhD at the Centre for Theoretical Neuroscience at Columbia University, where her research focused on building mathematical models of how the brain controls its own sensory processing. Before that, she earned a bachelor’s degree in Neuroscience from the University of Pittsburgh and received a research fellowship to study at the Bernstein Center for Computational Neuroscience in Freiburg, Germany. She was awarded a Google PhD Fellowship in Computational Neuroscience in 2016 and has spoken at several international conferences. She is also the producer and co-host of Unsupervised Thinking , a podcast covering topics in neuroscience and artificial intelligence. Her first book is Models of the Mind . Why science? I started my undergraduate degree as a neuroscience and philosophy double major and I think what drew me to both topics was the idea that if we just think rigorously enou

The Science of Can and Can't - Chiara Marletto *****

Without doubt, Chiara Marletto has achieved something remarkable here, though the nature of the topic does not make for an easy read. The book is an attempt to popularise constructor theory - a very different approach to physics, which Oxford quantum physicist David Deutsch has developed with Marletto. Somewhat oddly, the book doesn't use the term constructor theory, but rather the distinctly clumsier 'science of can and can't'. The idea is that physics is formulated in a way that is inherently limited because it depends on using mechanisms that follows the progress of dynamic systems using the laws of physics. This method isn't applicable in circumstances where either something may happen, but won't necessarily, nor where something isn't allowed to happen (hence the science of can and can't, which probably should be the science of could and can't if we are going to be picky). Deutsch and Marletto have proposed a way of using 'counterfactuals'