How Losing To AI Made Former Go Champion More Creative, And What It Could Mean For Human Ingenuity
The optimistic case for how AI, rather than displacing the human mind, may end up pushing our intelligence beyond the "four-minute barrier" to new heights.
Mind-blowing AI content is flooding the internet. It’s hard not to be amazed. But, like, it’s also hard not to be worried: is this the end of human ingenuity?
What if the answer is no? What if AI ends up boosting our creativity, pushing the human mind above and beyond what we conceived as possible?
Lee Sedol, world champion at the game of Go, leaned forward with his face in his hands. Just staring at the board. It was 77 moves into game 4, during a series of five games against AlphaGo, the AI-based computer program developed by Google’s DeepMind. For many minutes he just sat there, rubbing his chin, pondering. What was he thinking about?
The commentators explained that Lee’s recent moves were testing AlphaGo, but the algorithmic opponent was able to respond in a flawless manner. Lee might have been running out of ideas, they explained.
Then, came move 78. The Divine Move, as it is often referred to.
Lee Sedol placed a white stone right at the center of the board, driving a wedge that increased the game’s complexity. AlphaGo has still estimated its chance of winning at 70%. But then it made a mistake.
I am not really sure what AlphaGo is trying to do here, is how commentators responded to the next move. Move 78 by Lee Sedol has sent AlphaGo on a tilt, which is not something you would expect to happen to a computer. Ten moves later, AlphaGo's estimated chances of winning started dropping. It eventually resigned. The neural-network based algorithm, running on hundreds of computer processors, has been defeated at game number four. By a human, Lee Sedol.
Despite losing game number four, AlphaGo has won the series. The game of Go – orders of magnitude more complex than chess – represents the pinnacle of human intelligence; now it was mastered by a machine.
There was, though, an interesting effect at play here.
Put yourself in Lee Sedol’s shoes: You’ve dedicated your life to mastering the game of Go. You’ve been a professional since the age of 12, winning tournament after tournament. You became the best in the world. Now, this AI shows up. You lose game one. Then game number two. Then three. You effectively lost the series. The computer is the champion now.
Would you have given up?
Well, Lee Sedol certainly hasn’t. This is how he described it:
It made me question human creativity. When I saw AlphaGo’s moves, I wondered whether the Go moves I had known were the right ones. Its style was different, and it was such an unusual experience that it took time for me to adjust. AlphaGo made me realize that I must study Go more.
Losing the first three matches to a computer, has driven Lee Sedol to explore new strategies, and – eventually – to come up with “the God move”.
Perhaps this type of drive, the unwillingness to bow out, choosing instead to reach deeper into your mind and come up with even more creative strategies – maybe that’s the kind of “lizard brain” aspect of humans that AI will never be able to imitate.
AlphaGo has been constantly calculating its odds of winning. They were in the seventies when Lee Sedol came up with Move 78. Was it logical for him to even keep playing at this point? Let alone spend so much time staring at the board? Perhaps the analytical thing would have been to just give up. Yet, he didn’t.
Byrne Hobart compared it to the four-minute mile: the world record for running a mile stood at 4:01.4 since 1945. Many believed that it was the fastest a human could run. It wasn’t clear if a sub-four-minute mile was even possible. The four-minute barrier was first broken on May 6, 1954. While it had taken nine years to break the previous record, the new one was broken the following month. Once people realized it was possible, the four-minute barrier had been “broken” by over 1,750 athletes. By 1999, the record for running a mile was set at 3:43.13.
There are probably “four-minute-barriers” in so many fields. We consider them as the best possible designs for an airplane, an office building, a power plant, or a piece of software. But what if there are ways to improve on them, significantly? We just need something to push us to discover them. An incentive. A massive challenge. Perhaps that would be the impact of integrating AI into different industries.
The same goes for creative fields like arts, music, writing, films. Once AI enters the scene – sure, some people might leave. But others may rise to the challenge, and produce things that are far more impressive and interesting than humans have ever created.
This is already happening at Go. Beyond Lee’s “divine move”, researchers have found that since AlphaGo’s victory in 2016, there has been a surge in the quality and novelty of human decisions in the game of Go. After years of decreasing novelty.
Just imagine what the world would look like, if we experience a global surge in novelty and quality of human decisions, in field after field, across the board.
After winning game number four, Lee Sedol tweeted: “We cannot and should not give up on the human mind's ability to be creatively intelligent.”
Let’s not give up human intelligence just yet. Instead, let’s get excited about how far it can get, with a little push from artificial intelligence.
I remember a similar sentiment about chess. When AI showed up, people thought chess would become boring. But it has only increased in popularity! Constraints/competition foster creativity and reframes. Thank you for sharing.
This is a really thought-provoking take on AI. In my own work with LLMs, one thing that I've found is that the process of trying to automate tasks with LLMs requires me to think more deeply about what I'm doing. As Richard Feynman said, "If you want to master something, teach it."
In order to "teach" an AI to do things, I have to break down the task into clear steps, and I think this process encourages a level of understanding that goes beyond what was needed to just do the thing myself.
Something I've been thinking about recently is how AI is already superhuman in some dimensions, but we don't consider it to be AGI or superhuman intelligence because it still has things where it's far less capable than people.
An interesting thought experiment: in what types of problems have we already unlocked access to superhuman intelligence, and what is happening to human creativity in those fields? The game of Go might be one of them. Are there others?