Go Home — Lee Se-dol Quits Go, does not collect $100

Light, fluffy piece today.

If you remember a bucket of months back, Lee Se-dol (LSD), Go grandmaster, has fought DeepMind’s Go-playing bot a few times now. The bot soundly trounced him in most of the games, but he did get it to bleed a couple of times. Well, he recently fought a much better version of the bot and now claims the AI won’t ever be beaten by people. And so he’s retiring.

Some folks are mad at him for giving up — after all, chess bots have been impossible to beat for decades now. Chess still goes on. And the role of AI in chess is mostly to help train people to play better. Which of course means playing more like a bot. So take a second to think about whether that’s compelling to you. Because you can take it like this: Imagine you’ve been practicing throwing baseballs as far as you can your whole life. Every day, all day. Everything is about getting the ball to go just a bit farther. People praise you for your talents, then eventually your skills. You are the best on the planet, but you’re always just a little bit better than another guy “Patrick” (whom we all hate for the way he acts outside the noble game), who sometimes gets lucky and surpasses you — but never for long. You are the best.

Then BASEBOT9000 comes along and casually, boringly hucks balls as far as you ever have. A worthy foe! You put everything you have into matching it at the tournament and you actually succeed in beating it a couple of times, though it casually, boringly beats you the other times. So you think Now that I know this BASEBOT mothasucka exists, I’m going to reach my full potential and beat it the next time we meet. A few months pass. You compete again, this time against BASEBOT9001.

BASEBOT9001 casually, boringly shatters your record, ball sailing over and out of the park, killing dozens. You still give a few throws but even your new best is a joke.

Now, you have a few choices. You can observe BASEBOT’s techniques and try to throw just like it does. Hey, you’ll probably be better than you have ever been. You know for sure that Patrick — dick — will take to this immediately. Soon you and everyone else will be playing the new game and trying and failing to perfectly emulate BASEBOT’s perfect technique. Your life will be emulation of perfection.

Your other choice is to walk away and do something else. LSD chose this route.

It might look like I’m defending LSD’s decision from you and the other wolves, but that’s not what I care about here. If anything, I’m going to hit him in a different place. LSD’s problem will be all of our problem. What’s LSD’s big problem? He thinks this is about winning. It can’t be about winning. AI will always win. That’s the point of AI. There’s never going to be a point where AI is basically the same as us. Or at least, that moment will last for seconds and then blow by us.

As I discussed in Alpha Zero — The Robots can’t even Beat us like Men, we have a primal little love of playing and winning fairly. Maybe I shouldn’t put it so pithily. After all, this desire is probably useful. If it’s primal, it served some purpose. We can readily furnish reasonable-sounding guesses as to why we thought winning was important. All the game theorists in the audience are already jumping up and down out of their seats, rambunctious tikes. Yes yes, there’s probably a zero-sum kind of game theory thing at the heart of our competitive drive. And men are probably more mindlessly driven by it than women, on average. And it probably manifests itself in stupid ways once scaled up, like — I dunno — Cold Wars. Stupid. But understandable from the position of players on the game theory board evolution put us on.

Well, it’s time we socially adjust to being hopelessly frail in comparison to AI. (And just big tech in general — I’m shortcutting with a more menacing catchall here.)

Science fiction seems to earnestly believe that this can be made clear with humanoid robots that have super-strength, know no fear, cannot be seduced, and so on and so on. They take things that look like us, make them evil-looking (red eyes help), then show us get beat up by them until a plucky human overcomes this weakness through sheer dumb luck. In a sense, we learn our lesson. We get scared. But then we log on to Facebook. Somewhere, Zuckerberg smiles.

The real killer robots aren’t robots, nor will they be. Drones already do a more brutally efficient job at the whole killing game than any foot soldier ever did and there’s no reason to believe they won’t just get better at it. And that’s just the obvious killbot type drone. The real killbots will just be hackbots that steal your information and sell you to whoever wants your life. Assuming we get there.

Just to stick with hacking for a second, you really, really don’t understand just how important cybersecurity is. It might be because of all the derpy movies made about it and it might just be the word “cyber”, which should have been abandoned in the early 90s with those horrid, staged permed family photos. Your personal information may not be worth much to you and you may be willing to sell it to save a few dollars a month (you are), but if a competent and malicious (not even evil) someone wanted to destroy your life, they could. They could plant evidence of you committing crimes, siphon away all your savings, alienate your family — this is only going to get easier for less-competent people to do, unless your government helps you out, and you help yourself out. And you will still lose if you make an enemy who really wants to end you. Consider this your depressing PSA! Back to lighter discussion — killbots!

Right, so there’s no reason to expect any one of us is strong enough to fight intelligent technology — and I’d argue (am arguing, it turns out) that we need to get used to this. Not roll over to this — just get used to it. And do what you can to stay aware and informed. But it doesn’t (read: cannot) stop at that. We need to shift whole swaths of humanity away from these optimism-laden hopes of overcoming the bots — they will take our jobs and control our economies. They will control our decisions at scale and harbour/threaten our democracies. This is inevitable. What we need is to find out where that leaves us as a species.

Like Lee Se-Dol, we need to quit expecting to win and instead look for other sources of value and support. What should humans do with their time when they can’t spend it doing manual, menial labour? Well, it turns out we have humans that have done that (nobles) and while it hasn’t turned out okay for all of them (lots of real rotten apples), some of them have turned out alright (right?).

Unlike Lee Se-Dol, we can’t quit the game entirely. Or shouldn’t. Because AI is just as capable of teaching us to do good as the wisest master on the loneliest mountain. AI will teach us to run better democracies. It will open up our understanding of our own minds. And it will free us that pointless labour. And raise billions out of poverty. And cure diseases and maximise happiness and potential. If the good things win out. But they’re not going to win out unless we stop thinking about ourselves as winners in a zero-sum game. This world is too big for individuals to do their ape-dance on the skulls of the vanquished.