Outnumbered - David Sumpter ***

This book contains some impressive and important content - so I struggled initially to understand why I found it difficult to get on with. More on that in a moment.

Applied mathematician David Sumpter takes apart our current obsession with algorithms, information bubbles, AI and fake news, showing that all too often what we read about it is more hype than reality. Whether he is dealing with the impact (or otherwise) of Cambridge Analytica on elections, or the ability of algorithms to out-think humans, he shows that we have too often assumed that sales pitches were a reality: at the moment AI and its algorithms are rarely as good as we are told.

It might seem that this is the work of an academic with an axe to grind about the other mathematicians who are coining it in, but this is no unsubstantiated polemic. In many cases, Sumpter describes constructing a model to simulate the workings of an algorithm and demonstrates how feeble it really is. It was also fascinating to discover the way that an algorithmic presentation of 'also liked' amplifies (mathematical) chaos to bring out near-random winners - responsible, for example, for those YouTube stars where no one can understand their success.

I absolutely loved one section where Sumpter is trying to assess the intelligence levels of current AIs. Clearly they can't match humans. How about dogs? No. Bees, maybe? No. He shows that in reality, current machine learning struggles to match the intelligence level of an advanced bacterium.

Everything about what's in the book (apart from Sumpter's enthusiasm for football) seems a perfect match for someone deeply interested in algorithms and AI. So why did I not find the book particularly compelling? In part it's because it has quite a dry presentation. Unlike Sumpter's previous title Soccermatics, the style here is very measured and near-academic, presumably to add weight to the content, but the result was that some of it proved a dull read. 

It's not all like that, I ought to stress. I loved the line when considering what the Cambridge Analytica model promised: 'Democrats... could focus on getting the vote out among Harry Potter fans. Republicans could target people who drank Starbucks coffee and people who go camping. Lady Gaga fans should be treated with caution by both sides.'

I think the other issue was the 'negatives don't engage' syndrome. While it's important to know that algorithms and AI are far less powerful than we are generally given to believe in the news (and some books), it's hard to get too excited when told about something not being the case. It's a bit like the news headline 'War does not break out.'

The last thing I want to do is put people off this book. It really was interesting to learn how relatively ineffective AI is at this stage of its development, given how much news coverage has been given particularly to Cambridge Analytica, but also to the dark power of algorithms. It's an important message. I just wish the way it was delivered had been more engaging.

Hardback:  

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

Review by Brian Clegg