Kasparov's Law on Productivity in an A.I.-Driven Future

Anyone who's asked me for any recommendations lately knows how hard I've been banging the drum for David Epstein's seminal 2019 book Range: Why Generalists Triumph in a Specialized World (Chinese title:《成长的边界》).

Without exaggeration, this is the most influential book that I've read in the past 15 years. It is difficult to give it a concise description, as it covers and analyzes such a broad expanse of topics. I gave a copy to our company CEO, and his key takeaway (i.e. logic from other fields can be extrapolated to solve new problems where direct experience doesn't exist) was completely different than mine (i.e. learning in the most challenging environment often results in the slowest progress but deepest retention), and yet these were still just two among many fascinating topics covered at great length in the book.

On a regular basis, I see or hear something that makes me think of one anecdote or another from the book, most recently when I was going through the four episodes on NVIDIA from the Acquired podcast. These episodes essentially are a mini-audiobook on the history of Jensen Huang and NVIDIA, and I strongly recommend a listen to anyone interested in that field (that's right, two call-outs in one article, hashtag engagement, hashtag gimme all da likez).

In the third episode, co-host Ben Gilbert described Huang's notion that every application in the future could have a GPT front-end interface, allowing humans to interact with computers more naturally. Ben says, "I think [Jensen] means everyone can become a programmer, but the programming language is English." This would jive with Huang's statements that programming should no longer be considered a critical skill, and that humans should put their focus on higher-level domains like biology or education and leave the minutae of coding to A.I. What intrigues me is that a microcosm of this problem has already been explored by another genius whose destiny was upended by A.I.: Garry Kasparov.

Despite being the greatest chess grandmaster of his era and a key opposition figure in Russian politics later in life, Kasparov unfortunately may be most remembered as the first world champion to have his ass handed to him by a computer in the game he devoted his life to playing, when he lost to the IBM supercomputer Deep Blue in a series of chess matches in 1997. What is talked about far less is what Kasparov did after his defeat, which is detailed in Range, and which I will paraphrase/plagiarize for your benefit below.

There is a saying that chess is 99% tactics, or short combinations of moves to achieve an immediate advantage on the chessboard. There are patterns to these tactics, and grandmasters spend their lives memorizing these patterns. Yet pattern recognition and memorization are as easy to computers as eating and breathing are to us, making them tacticaly flawless compared even to the best humans chess players. Today, even the chess app on your phone could beat Kasparov, according to Kasparov himself.

However, the other supposed 1% of human chess-playing is strategy, which involves the bigger picture planning of getting your pieces into good positions to deploy your tactics. What Kasparov noticed in his games against Deep Blue was that, while it was insanely stronger than humans at tactics, it was comparatively weak when it came to strategy. Well, what if we just combined the two brains, allowing humans to make the strategic plans and outsourcing tactics to computers?

In 1998, Kasparov helped organize the first chess tournament where each human player paired with a computer. Suddenly strategic creativity became much more important than the decades of memorization and repetition on which grandmasters relied. A month prior, Kasparov had trounced another player four games to nil in traditional, human-only matches, but in this tournament, he had to settle for a 3-3 draw against the same player.

In 2005, the first "freestyle chess" tournament was held featuring teams that could consist of multiple humans and computers. This was a tournament that featured not only teams of ranked grandmasters and their computers, but also Hydra, the then most advanced chess supercomputer with an estimated Elo rating of over 3000, significantly above what any human has been able to achieve. Yet the winner of this tournament came from a much more pedestrian background. Two amateurs from New Hampshire, named Steve and Zack, crushed grandmasters and supermachines alike using three PCs and $60 worth of software that anybody could pick up at Best Buy. They were just good enough chess players, but where they excelled was in knowing when and how to use their computers to aid them.

This shocking result led Kasparov to invent Kasparov's Law, which states that a tactically weak human with a computer and a strong process is superior both to a strong computer working alone and a strong human with a computer but a weak process. The "process" in this case refers to having informed humans who understand the rules and limitations of their A.I. partners, can form the right prompts to get the appropriate outputs from the A.I. and then utilize them in the best way on the battlefield.

I believe this is the type of reasoning that Jensen Huang, Sam Altman and other "stop learning how to code" futurists are using to try to convince us that working with A.I. will help humanity unlock a new level of productivity that, in the narrow view, could potentially replace all the old lost jobs with newly invented ones and, in the grand view, could potentially lead us towards an epoch of economic surplus.

Ultimately, it doesn't matter whether or not you believe them. All the pure skill-based jobs that A.I. does better than humans do will disappear at some point in the future, likely in our lifetimes. Right now, the hysterics commenting on every LinkedIn post with an A.I. video crying, "But what about the artists?!" are acting as futilely as the people at the turn of the 20th century crying, "But what about the horses?!" as the first combustion engine cars rolled onto the streets. That future is coming, and every day you spend trying to turn it back is another wasted day that you should have spent preparing for it instead.

The winners in this future will be those who can create and implement the strong processes, the ones who can interface properly with our A.I. partners, at least according to Kasparov's Law. As it so happens, though, there is another topic in Range that has something to say about this as well. In a different chapter, the author describes how people with extensive experience in domains like chess, golf and playing a musical instrument generally tend to become masters, but possessing extensive experience in other domains like war, politics and medicine has not demonstrated the same correlation to mastery.

The book theorizes that this is because domains like chess are kind learning environments, where patterns constantly repeat and feedback is extremely quick and accurate. However, other domains are wicked, where rules are unclear or incomplete, patterns do not repeat or are not obvious, and feedback is often delayed, inaccurate or both.

There is great danger in treating a wicked world as if it were kind. Does a human-A.I. hybrid being the most successful chess-playing combination mean it will be the most successful combination in other domains? Certainly in some, but could you neuralink a charismatic avatar with GPT-5 and beat Donald Trump in an election? I think the world might be too wicked a place to put all our faith into that convergence just yet.