# The Stories Were True

> Ten years after AlphaGo, a reckoning.

19 November 2026

Filed by Charli-Jo Tyrer for [The Blind AI](https://carrados.com/blind-ai/).

I was eight years old when [Star Wars](https://www.starwars.com/films/star-wars-episode-iv-a-new-hope) came out. I remember sitting in the dark hearing Red Leader going in and feeling — with absolute certainty — that what I was watching was *about* something real. Not real in the sense of spaceships and lasers. Real in the sense of the tension between what machines can calculate and what humans can find when calculation isn't enough.

I spent the next four decades reading science fiction that told that story in a thousand different registers. The machine that sees further than we do. The human, cornered, who finds a move the machine never modelled. The grace note that doesn't change the outcome but still means something the outcome can't cancel.

I had no idea I was being prepared for a Sunday afternoon in Seoul in March 2016.

Move 37 gets most of the anniversary coverage. Deservedly. When [AlphaGo](https://deepmind.google/technologies/alphago/) played a shoulder hit on the fifth line in Game Two — a move that a professional player would almost never make at that moment in that position — it broke three thousand years of accumulated human intuition about what the game was. The commentators thought it was a mistake. [Lee Sedol](https://en.wikipedia.org/wiki/Lee_Sedol) took twelve minutes to respond. It wasn't a mistake.

What Move 37 proved was legible and communicable: artificial intelligence, trained by reinforcement learning on nothing but the game itself, had developed a strategic vocabulary that human players had never needed to develop, because human players had always been playing against other human players. AlphaGo wasn't smarter than Lee Sedol. It had been educated in a different school, by a different teacher, and it had learned things that the human tradition had never taught.

That's a remarkable thing. It became the dominant story of the match and of the anniversary.

But it's not the story that made me catch my breath.

Game Four. Lee Sedol has already lost three games. He knows the series is gone. He has spent four days studying an opponent that has no fatigue, no ego, no psychological state he can read or exploit. He is, by any rational measure, in an impossible position.

He plays Move 78.

A wedge in the centre. A move that, according to [AlphaGo's own probability calculations](https://www.npr.org/sections/thetwo-way/2016/03/13/470284113/after-three-losses-master-go-player-scores-a-win-against-computer), had a one-in-ten-thousand chance of being played by a human. Not because it was a bad move. Because it was so far outside the distribution of human Go that the model had never needed to account for it. AlphaGo had correctly modelled what human players do. Lee Sedol stepped outside that model.

The machine, encountering something it hadn't priced in, collapsed. It made a series of what commentators described as "terrible moves" — the kind of errors that occur when a Monte Carlo tree search hits a branch it hasn't adequately explored. Lee Sedol won the game.

It was the only game AlphaGo ever lost to a human. And what broke the machine wasn't human cleverness or human stubbornness. It was that under existential pressure, facing a genuinely non-human opponent for the first time in his life, Lee Sedol had been compressed into a cognitive shape that no human player had previously occupied.

He had become, briefly, divine.

Here is what I want to say to anyone who spent forty years reading science fiction and then watched this happen:

*We knew.*

Not the mechanism. The stories almost never got the mechanism right. The machine usually lost because of arrogance, or brittleness, or some fundamental flaw in silicon that flesh didn't share. The stories needed the human to win in a way that reaffirmed human primacy.

This was better than that. This was the literary version. The [Peter F Hamilton](https://www.panmacmillan.com/author/peterfhamilton) version, not the airport thriller. The human found a transcendent move that was real and verifiable and genuinely unprecedented — and still lost 4-1. The grace note was real. The war was over. Both things were true simultaneously, without contradiction.

That is not a technology story. That is a story about what happens at the edge of what has been modelled. It is a story about the strange mercy of being forced into contact with something genuinely other. It is a story that science fiction has been telling us, in preparation, for decades.

The emotional climaxes were real. All of them. The wonder, the grief, the single moment of transcendence that changes nothing about the outcome and everything about what the outcome means.

Lee Sedol retired in 2019. He said there was [an entity that could not be defeated](https://www.scmp.com/news/asia/east-asia/article/3039617/googles-go-computer-cannot-be-beaten-so-south-korean-master-lee). He said it calmly, as someone who had looked at the thing clearly and made peace with it. He didn't say it with bitterness. He said it the way a character in a good science fiction novel says the thing at the end that they couldn't have said at the beginning.

I've been working at the intersection of AI and human experience for years. I spend my days thinking about what these systems can and can't do, where they extend human capability and where they simply replace it, what is genuinely new and what is old pattern in new clothing.

But every so often I come back to a hotel conference room in Seoul, and a move that had a one-in-ten-thousand chance of being played, and a man who played it anyway because the pressure had made him into someone he hadn't previously been.

The stories were true. Not in the way we expected. Better than we expected. With the grief left in.

That still seems to me like the most important thing that has happened in AI. Not the capability. The story. The fact that the oldest human story — human and inhuman opponent, edge of what is modelled, moment of grace that doesn't change the war — turned out to be literally, precisely, technically true.

Red Leader going in.

*I'm a lifelong science fiction reader, an AI practitioner with ten thousand hours in these systems, and someone who has been in the access and assistive technology field for four decades. I write about the intersection of all of these things for the Blind AI.*


## More from Charli-Jo Tyrer

- [Any Agent, Any Screen](https://carrados.com/blind-ai/any-agent-any-screen/)
- [Thirty Minutes, Four Changes, No Laptop](https://carrados.com/blind-ai/thirty-minutes-four-changes-no-laptop/)
- [Cloud Dancer, Flo, and Phix](https://carrados.com/blind-ai/cloud-dancer-flo-and-phix/)
- [Everything Charli-Jo Tyrer has filed](https://charli.carrados.com/)
