Book review: Range: How Generalists Triumph in a Specialized World

I had to wonder. In my previous computer science positions, my coworkers were rarely computer scientist majors. They had a varied background, like chemistry, and I have myself an odd background (majored in signal processing, digital electronics and automation, then music and also a PhD in machine learning in partnership with a neuroscience lab).

In finance, lots of people are finance first and only, and they take everything at face value. Then came Epstein’s book. Could it explain what I was seeing?

Discussion

Contrary to my other reviews, I will not talk about each chapter, but about the general content of the book.

I think the book gives answers to different questions, not just “should you generalize”, but it also has some shortcomings, but I’ll come back to this later.

What hit me in the first half of the book is how successful are some people that are generalists. They can take a step back and understand what was going on (which is very important in finance, where lots of people don’t have that breadth of vision). And if you look at Epstein’s reasons, it is obvious that another parallel should be drawn, this is time with artificial intelligence. I loved the chapters that cover the discovery that concepts are center to what we count as intelligence.

And this is were it is important (and where I digress). When presented with a just few samples, the human brain can find the concept underneath very quickly. You don’t need million of miles to know how to drive safely. Of course, this skill is due to our instruction from the time learned how to crawl.

This is in contradiction with the current trend of ML/AI where you throw even more similar data and hope that it will generalize. Unfortunately, it won’t, because the architecture of the algorithms (neural networks as well) are not suited for that. For instance, let’s reuse one of the examples in the book about a circle and a disk, and people asked to categorize shapes like these. They would not put them together because for them a disk would be the moon and the circle something else, which were fundamentally distinct. For a neural network, it’s the same. It will not be able to understand that they are the same unless you train it to say they are the same.

The brain has a mechanism to assimilate concepts while we sleep. And as I’ve said, we can learn how to drive also because we learned how to walk, cycle… It was integral to this learning process to see other things.

Another parallel is interleaved learning. Epstein says that it’s far more efficient in humans because it helps generalizing. Same for neural networks, you know that you need to interleave. Still, it’s the struggling part that he says is important, which leads me back to reinforcement learning where we need even more data for a result that still doesn’t lead our models to concepts. In humans, such training does. I remember reading math books when I was 15 years old, and I couldn’t make sense of them. Fast-forward to “classes préparatoires” (French elite learning system) where you are struggling every minute of every day of every week for two or three years to learn math, physics, engineering science, French, English. This was the perfect setup to struggle, and to learn concepts. It also confirmed my feeling that learning something by heart (like a language -> Duolingo is basically useless and a waste of time) doesn’t help. Learning history, a language by heart doesn’t work. You have to immerse yourself so that you struggle.

I like the book until the last couple of chapters, as he still seem to respect generalists (the last couple ones seem to be very harsh to specialists, which I find unfair and shortsighted). Current research requires mastering different skills. I chatted a few days ago with a researcher, and he said he now needed to be a good coder, a good statistician and an excellent neuro-psychologist to be able to succeed. But you need to rest on shoulders of giants. Without them, you won’t solve anything. You need the statistician expert to help you design your tests. You need the expert coder to get your results in time.

The book also sometimes feels like anyone can be a famous artist, and that we should all do that. But not everyone can. The current society cannot work with everyone being a celebrity. Or an influencer…

Conclusion

The book is very interesting, with lots of interesting examples. Still, it seems to idolize generalists when specialists are also required. Without them, a breadth of knowledge would be inaccessible. Both are required, this is something that may not transpire at the end of the book as much as it should.

And obviously; not everyone can be Steve Jobs. Even if you generalize, your intellectual ceiling will probably be far lower than Jobs’. For every genius generalist, how many standard generalists and mediocre specialists like me?

It would be great for everyone to achieve their ultimate goal, spend time to find it, but let’s face it. What would this mean? If Epstein followed his train of thoughts during the book, that should be the last question of his book: Can our society survive with billions of genius generalists?

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.