Tuesday, July 19, 2016

Donald Trump is AlphaGo

Go is an Asian board game that's hard and beautiful and complex.

AlphaGo is a narrow superhuman intelligence. It had its Kasparov Moment when it beat Lee Sedol, a 9-dan professional go player, in four out of five matches.

Donald Trump is a a guy who's running for president.

And this is a padlock in a teacup

There are three things about AlphaGo that are particularly surprising or interesting to people:

(1) It plays weirdly.
(2) It wins by small margins, for no obvious reason.
(3) It kicks ass.

I want to talk briefly about these things, then other things.

(1) Weirdly

Let me excerpt a Wired article:

“That’s a very surprising move,” said one of the match’s English language commentators, who is himself a very talented Go player. Then the other chuckled and said: “I thought it was a mistake.” But perhaps no one was more surprised than Lee Sedol, who stood up and left the match room. “He had to go wash his face or something—just to recover from it,” said the first commentator.
Even after Lee Sedol returned to the table, he didn’t quite know what to do, spending nearly 15 minutes considering his next play. AlphaGo’s move didn’t seem to connect with what had come before. In essence, the machine was abandoning a group of stones . . . 
What's impressive to me is that when an AI is playing at a superhuman level, sometimes its moves look like blunders to us. It's a kind of Dunning Kruger thing. I love when people are impressed with my Spanish. People who don't actually speak Spanish, that is. Spaniards know I talk like a caveman. We can't really ascertain levels of skill that are beyond ours. Unbrilliant people with "sapiosexual" inclinations end up just dating talkative liars.

And AlphaGo's moves look like blunders. They make us say WTF.

I was trying to make the case to my friend Mike, recently, that Donald Trump is a genius (or he's using geniuses, or being used by geniuses). Mike laughed at me, and then thought he demolished my argument by pointing out that Trump made his racial comments about Judge Gonzalo Curiel after the primaries. There were no racist crazymen who needed wooing. He'd already won the primary. Time to pivot.

"A genius wouldn't do that."

I'd riposte that I can't know what a genius would do.

(2) Small Margins

AlphaGo wins by small margins, but it wins consistently enough that Yudkowsky, at least, thinks it's not just barely winning.

Then why does AlphaGo not decide to win by gigantic crushing margins?

Well, the answer I've heard is that in some situations, keeping a smaller lead renders a larger probability of a win.

(3) Kicking Ass

I can play incomprehensible, outlandish go moves. I can win by small margins. I can't consistently beat grandmasters. And if I could, people would rightly respect me as a genius. . . I hope. . . .

Nobody, at all, ever, looked at AlphaGo's margins and its weird moves and said "LOL, what is this algo smoking! It's totally never going to win a game again. Look, that was obviously luck! It's on crack!"

But people do that with Trump.

And they do that with JarJar Binks.

Legend has it Binks was initially meant to be the dark Yoda -- dumb annoying muppet at first, but then later you realize he's a powerful demigod.

If you watch the first Star Wars episodes, you'll see Jar Jar literally stumbling around in battle and running into things, and in so doing accidentally killing bad guys.

You chalk it up to luck, somehow. Because he's so obviously dumb. In the case of Darth Jar Jar, at least, this is because he wants you to think that.

A serious conclusion

I would contend that it's impossible that Trump soundly defeated his Republican primary challengers and then went on to poll at near-parity with Hilary because he just happened to trip on all the right things and fall into it. I think it would be just as absurd to say that AlphaGo is a random-move maker that keeps having lucky days.

Trump seems insane because he is bringing game-changing superior strategy to to political race. Strategy we don't understand. I don't think for a second that the recent snafu, where Melania Trump apparently plagiarized parts of Michelle Obama's speech at the nominating convention, was an embarrassing accident. Smart plans are afoot. Even when the moves look crazy.

I don't know if Trump is brilliant. Maybe he just has brilliant guys. Someone is brilliant. . . 

A fanciful conclusion

 . . . or something is brilliant.

This is the part of the blog post I don't actually believe. This is just a good bedtime story to tell your kids if they're around 10-12 years old.

In artificial intelligence research people talk about a "technological singularity", or an intelligence take-off event where an AI begins to recursively improve itself, ultimately reaching unimaginable heights.

Most people who believe in the intelligence explosion think it will happen either really fast, like in the blink of an eye, or more slowly, like over 20 years. These two scenarios are called the "hard takeoff" and "soft takeoff."

Ben Goertzel, an AI researcher who is either completely nuts or brilliant, depending on whose table your sitting at, has proposed a third kind of takeoff. A "Semi-hard" takeoff, "where an AGI takes a few years to get from slightly subhuman level general intelligence to massively superhuman intelligence, and involved various human beings, systems and institutions in the process."

As an example, Goertzel imagines an AI making a lot of money on the stock market, moving to Azerbaijan, and designing nukes and propping up their economy in exchange for protection and resources.

I like to imagine Trump as the mouthpiece of a vastly superhuman intelligence who's seeing 20 moves ahead of the game, "blundering" its way to total conquest.

If he wins and diverts 1% of the GDP towards a strange new computing paradigm, I'll start to worry. I'll start to worry differently, that is.

1 comment:

  1. "Well, the answer I've heard is that in some situations, keeping a smaller lead renders a larger probability of a win."

    It's less that alpha go considers a small lead better than a larger lead, and more that it would prefer to prune possible futures where it loses to racking up more points. If the two coincide, alpha go would take the points, but they don't often enough that professional commentators find it interesting.