Centaurs: The Most Powerful Thing You Can Become
A computer beat the world chess champion. Then the champion found something better than beating it.
Let me tell you my favourite story about AI, which is actually a story about chess, and which I think is the most useful thing anyone can understand about the next ten years of their career. In 1997, a computer called Deep Blue beat the world chess champion, Garry Kasparov. The headlines were exactly what you would expect. Humans, they declared, were finished at chess. Game over. The machines had won. And then Kasparov did something far more interesting than losing gracefully. He asked a better question.
The better question
Instead of human versus machine, he wondered, what about human with machine? What if a player could consult a chess engine during the game, the way you might consult a brilliant but occasionally clueless advisor? He called it Advanced Chess, and other people called it centaur chess, after the mythical half-human, half-horse, and the results turned out to be genuinely strange and wonderful.
The best chess player in the world stopped being a human, and stopped being a computer. It became a human and a computer, working together.
In one famous tournament, the winners were not the best humans or the best machines. They were ordinary players with ordinary computers, but a brilliant process for combining the two. They beat grandmasters. They beat far stronger standalone engines.
Why the centaur wins
The reason is simple once you see it. Humans and machines are good at almost opposite things. The machine is tireless, precise, encyclopedic, and utterly without judgment about what actually matters. The human is slow and forgetful and emotional, but knows which lines are worth exploring, senses when something is off, and understands the difference between a move that is technically fine and a move that wins. Put them together and each one covers exactly the other's weakness.
The skill is the combining
This is the template for the entire AI age, and it is enormously good news. The future is not human versus AI, a contest you will lose. It is human plus AI, a partnership you can get extraordinarily good at.
The value is not in the human or the machine. It is in how well the human steers.
You do not need to be the best human in the world, and you do not need the best AI. You need to be the best at putting the two together, and that is a game almost anyone can learn to win.
"But the centaurs got beaten in the end"
Here is the objection an honest chess fan will raise, and it is a real one. The centaur era did not last. Engines kept improving until, by most accounts, a strong human added to a top engine no longer helped and often just introduced noise. The machine alone became best. If that is how it went in chess, why expect human-plus-AI to be the durable answer anywhere else, rather than a brief transitional stage before the machine solos past us again?
Concede it completely: in chess, the human did become a liability. But notice what makes chess the easy case. It is closed and fully specified. The rules are fixed, the board is fully visible, and above all the goal is given, checkmate, a win condition no one has to define. When the objective is handed to you, raw search eventually wins, and the human is just slower search. Almost no real work looks like that. The objective is not given, it is the hardest part. What counts as a good strategy, a good essay, a good decision is contested, contextual, and changes with who it is for. There the machine cannot solo past you, not because it lacks horsepower, but because it has nothing to optimize toward until a human decides what winning means. The centaur loses exactly where the goal is fixed. It keeps winning everywhere the goal is the real question, which is almost everywhere that matters.
Your move
Knowing what to hand the machine and what to keep. When to trust its answer and when to override it. How to combine its tireless power with your irreplaceable judgment. The future belongs to the centaurs. Go and become one.
I write about AI, data, and learning regularly at pinaldave.com, and I have been teaching this hands-on in my AI workshops.