‘Outnumbered‘ is a book that I have been expecting for the past couple of years. Its premise: this whole algorithm, data science and AI revolution that people talk about on cocktail parties seems amazing and overwhelming, but under the hood it is rather ‘meh’.
The author, David Sumpter, is a professor of applied mathematics in Uppsala, Sweden. He writes colums for several newspapers and is famous for his book soccermatics — not to be confused with soccernomics.
Sumpter is a master in reducing key concepts within data science to a striking analogy so that no prior knowledge is needed to understand the issues at hand. As a matter of fact, if I would have to make a checklist of techniques that are popular within the field of data science, I think I’d have a bingo. Principal component analysis, recurrent and convolutional neural networks, graph theory; it is all in there and explained within the right context.
In 18 chapters, the book outlines the most common applications that are driven by data in today’s society, how they work and how effective these techniques are. From filter bubbles, to fake news, to manipulating elections, chatbots and beating humans at computer games — the author finds it all a little underwhelming. Sumpter concludes in the final chapter of his book with the following rant.
“First of all, the Facebook study showed that people who were bombarded with negative news had a propensity to write one more negative word per month, a miniscule but statistically significant effect. Secondly, the company, called Cambridge Analytica, had nothing to do with Trump’s election success. […] Thirdly, yes, computers trained on our language do make sexist and racist associations, but that’s because our society is implicitly prejudiced. […] the question of whether we can create a general AI remains wide open. We can’t even get a computer to learn to play Ms Pac Man properly. Oh, and Elon Musk is an idiot.”David Sumpter in ‘Outnumbered‘
I do not agree with everything that is outlined in the book. For example, when assessing filter bubbles, the author claims that users are exposed to all kinds of opinions — conclusion: filter bubbles don’t exist. In my view, deespite the fact that Facebook users are exposed to multiple opinions does not mean that they engage with it. Furthermore, given the razor-thin margins, I do believe the abuse of personal data contributed to Donald Trump’s election victory — contrary to the book’s conclusion that it had nothing to do with the victory.
Nevertheless, I do agree with the opinion that AI is overhyped. When a panel of Silicon Valley’s most mediatized figures are put in the same room to discuss the dangers of general artificial intelligence, what do you expect? That they’ll claim that AI and machine learning are boring statistical models? Imagine what their stocks will do.
This book offers something for everyone. If you’re working with data, you’ll be challenged to face some hard truths. While reading this book I pulled 5 scientific papers to see for myself if the author is right — e.g. here’s what I found out about using algorithms to predict crime. If data science is strange to you, not only will it soothe your worst robot fears, you’ll also pick up a thing or two to discuss at the office watercooler.
If you enjoy this book, here are some other ones you could pick up:
- The signal and the noise, by Nate Silver
- Chaos monkeys, by Anonio García Martínez
- Algorithms to live by, by Brian Christian
- Weapons of math destruction, by Cathy O’Neill