You’ve probably noticed that there are a lot of algorithms about these days, doing everything from recommending a walking route to figuring out how to beat the world’s best players at Go. But what are they, really, how do they work, and how will they change the world?
I’ve read some excellent books recently on the subject and have a few recommendations.
For a fun and memorable discussion of how specific algorithms work (even how you might use them yourself to sort out your sock drawer or find a nice apartment) then try Algorithms to Live By (UK) (US) by Brian Christian and Tom Griffiths. I enjoyed this book very much, although not quite as much as Brian Christian’s The Most Human Human (UK) (US), which is all about how to have a better conversation, whether you’re a human or a bot. It’s one of my favourite books, ever.
On the economic and social implications of artificial intelligence, I strongly recommend Prediction Machines (UK) (US), by Ajay Agrawal, Joshua Gans, and Avi Goldfarb. Since I wrote “What We Get Wrong About Technology”, I’ve been telling people not to overlook simple, cheap innovations (paper, the shipping container, concrete). Space travel and supercomputers get all the press. Just being cheap doesn’t. But being cheap can transform the world. “Prediction Machines” gratifyingly chimes with this idea: the authors argue that artificial intelligence is best thought of as a way of producing super-cheap predictions; predicting what you might buy, predicting whether a shadow on a scan is cancer, predicting what the Japanese translation of this sentence might be.
Implication 1: good predictions reduce uncertainty, and lots of things we do are a response to uncertainty. For example, freezers (uncertainty about what and when I will want to cook) Airport lounges (uncertainty about how long it will take me to get to the airport means I show up early). AI is therefore bad for airport lounges.
Implication 2: sufficiently good predictions are game-changers. If Amazon’s recommendation engine gets good enough, they can take the risk of shipping me stuff I haven’t yet bought.
Implication 3: “judgement” becomes an important complement to predictions. How bad is a false positive when I predict a fraudulent credit card transaction and annoy my platinum card holder? What about a false positive diagnosis of cancer?
Implication 4: AI rarely replaces an entire human job directly. It tends to replace specific tasks – small slices of what we think of as a job. Reimagining/reengineering workflow will be an important competitive advantage.
As a bonus, the book has lots of good examples and is written clearly. I learned a lot.
For a sceptical take on the limits and the toxic side-effects of machine learning, there’s Cathy O’Neil’s passionate, political and very readable Weapons of Math Destruction (UK) (US) or the new book Artificial Unintelligence (UK) (US) by Meredith Broussard, which I have barely skimmed but seems to contain a very good mix of storytelling, history and technical ideas. Promising.
Finally, I’ve been fortunate enough to read the manuscript of Hello World (UK) (US) – out in September – by Hannah Fry. This is really a superb overview: lots of good stories, clear explanations, and it’s wide-ranging. I think if you want a general guide to the new world of data-driven computing you couldn’t do much better than this.