Books Curated for Perspective

There exists no shortage of Machine Learning & Artificial Intelligence books for teaching technical knowledge, yet few for teaching the judgement to apply it.These are the ones that shaped how I think about evidence, and they are in the order they were written.

1. 

   Foundation Trilogy by Isaac Asimov

   Nothing else comes close when exploring the possibilities and limitations of statistical science applied to social systems. Asimov's seminal work is, at once, an inspiring and cautionary tale. It also features literature's greatest (perhaps only) statistician protagonist!

   View on Goodreads →

2. 

   How to Lie with Statistics by Darrell Huff

   Darrell is not teaching method, he is teaching suspicion. Every chapter is a technique that is arithmetically correct and still dishonest, which is the only kind of mistake that survives review. I have sat through presentations that used four of them on a single slide. Recognising one in your own work is considerably more uncomfortable than recognising it in someone else’s.

   View on Goodreads →

3. 

   The Visual Display of Quantitative Information by Edward Tufte

   A chart is an argument rather than a decoration, which makes this a book about honesty more than aesthetics. The data-ink ratio and the lie factor are the two ideas worth carrying forward, because both are measurable and neither is a matter of taste. Most dashboards fail on both counts. If a stakeholder has ever drawn the wrong conclusion from a chart, this is the book that explains why.

   View on Goodreads →

4. 

   Fooled by Randomness by Nassim Nicholas Taleb

   This book is a deathblow to the overconfidence that pervades Data Science. As humans, we overestimate causality and tend to view the world as more explicable than it really is. Our behaviors deviate from 'rational choice theory' precisely because we are so poor at assessing the world around us by ascribing success to skill and failure to chance. Only when we truly understand the role randomness plays in our lives can we build insightful and functional models and make better choices as an individual.

   View on Goodreads →

5. 

   Critical Mass by Philip Ball

   A fascinating read on the confluence of models that explain both social and physical systems. Cross-pollination of ideas between disciplines is the largest driver of innovation, and here Ball connects mathematical models from Physics, Biology, Chemistry, Economics, and Sociology. He also explores the idea of 'step-changes', where systems behave in entirely different ways after some threshold is crossed. I've used this approach many times to reconcile models which were valid for one set of data and then inexplicably invalid for another.

   View on Goodreads →

6. 

   The Wisdom of Crowds by James Surowiecki

   Diversity. Independence. Decentralization. Aggregation. The four requirements necessary for a crowd to be wise, co-incidentally also the four requirements for a large set of data to be valid. It's easy to forget the data we are analyzing originates from people and if that data isn't collected the right way or contains inherent biases, the results of our analysis are going to be as un-wise as the crowd from which they were drawn.

   View on Goodreads →

7. 

   The Signal and the Noise by Nate Silver

   Nate is perhaps the only celebrity statistician we are likely see in our lifetime (reason enough for excitement) but his message is also a powerful one: go back and check your work . How often do we build models and make forecasts that we never revisit? How committed are we to actually explaining phenomena over time and not just answering the question of the day? How deeply can we seek the truth?

   View on Goodreads →

8. 

   Weapons of Math Destruction by Cathy O’Neil

   The concern here is what a model does after it is deployed, and the argument is that scale, opacity and damage compound, so a model that is wrong about a few people quietly becomes a system that is wrong about millions. In a regulated industry this is operational rather than academic. The uncomfortable part is that most of her examples were built by competent people following reasonable incentives.

   View on Goodreads →