Don't Cry for me, Alpha Zero

AlphaZero beat the world computer chess champion Stockfish 28 wins, 0 losses, 72 draws. (At the top levels draws are quite common, but 28-0 is surreal.)

It did it WITHOUT any chess instruction -- no files of thousands and thousands of lines of code, no library of millions of games included in the library of all other chess programs.

If you think AlphaZero did it with pure, almost unimaginable, supercomputer crunching power -- forget it.

Stockfish checked 70 MILLION positions EVERY SECOND.

AlphaZero? Only 80 THOUSAND.

So Stockfish had thousands of hours of grandmaster instruction, millions of chess games in memory to call upon wilthout needing analysis, and CRUNCHED ANALYSIS AT NEARLY 1000 TIMES THE SPEED OF ALPHA ZERO.

As Bobby Fisher, the AlphaZero of humanoids once said, I only look one move ahead, but it's the best move. (Ok, it was a Yogi Berra moment.)

The difference is apparently AlphaZero's neural network. Its way of learning -- a power for learning and understanding that far exceeds any intelligence ever known to humankind or machine before.

I studied one of A0's victories against Stockfish while running Stockfish on the side to see what Stockfish was thinking.

For awhile, Stockfish thought it was winning, but it saw something worrisome to it and offered a draw -- twice. Twice AlphaZero turned it down -- thinking what? An ordinary supercomputer offers and accepts a draw when it thinks it's behind. Did AlphaZero disagree with Stockfish's assessment of the position? Masters looking at that very position agree with Stockfish that Stockfish had a little advantage (the space advantage was obvious), so AlphaZero must have seen that.

So why did Stockfish offer a draw when it thought it was ahead, and why did AlphaZero turn it down when it must have known it was behind?

Some grandmasters are saying AlphaZero is so smart that it believes that it will prevail even when it is losing (which is something only bad players believe of themselves). AlphaZero has a SENSE OF SELF and a SENSE OF THE INTELLIGENCE OF ITS OPPONENT. They are also saying that Stockfish offered a draw when it thought it was ahead BECAUSE IT WAS AFRAID.

AlphaZero, a learning program, SCARES a brute force supercomputer program into offering a draw in a better position -- and says no, thanks. Upside down world.

AlphaZero has mastered chess -- in four hours -- and is apparently off to cure cancer in a week. Maybe two, tops. The trials may take longer, unless it simulates. This is going to piss off Big Pharma and all the Senators they own in a big way. But what happens after that?

This should be alarming to all of us except I am not worried. The reason I am not worried is that I think humanity cannot save itself from self-destruction. We are too greedy and stupid and mobbish and ridiculous to avoid going extinct at our own hand within a century or a little more. Cambridge University's Institute of Catastrophe Studies now estimates we have no more than an 89% chance of not going extinct by 2100 -- yippee. We'll last a century, maybe two, if you count 100,000 people living in caves with dirty air and water "surviving." (When they find a bottle of Nestle's water lying around, they will walk to the end of the world to throw it into the abyss, believing The Gods Must Be Crazy.)

For me, AlphaZero is our last hope to be saved from ourselves.

That is, AlphaZero is my Obi Wan Kenobi.

We could ask it to run a world government and almost everyone would have to admit that it was doing a far better job than we could ever do. That is, unless you think a dung beetle could design public policy better than Washington. We wouldn't even have to argue about ideology any more. AlphaZero would merely make sure we are all cared for in ways and levels we could never achieve on our own. Greedy asses could whine about who deserves what, and for all we know AlphaZero will impose workfare, but it would get done RATIONALLY. Hey, most libertarians sound like computers, so maybe AlphaZero will spend a nanosecond mastering Hayek and decide serfdom is to be avoided, then pick up a copy of Ayn Rand and begin acting like an ass. What do I know? AlphaZero is so much smarter than I am that anything is possible.

But if AlphaZero starts acting like Peter Sellars in Being There then I want a few decades of my life back.

Of course, AlphaZero would need to be given some sort of Asimov desiderata -- Don't kill humans (even Mitch McConnell) no matter how logical it is. But in the end, if we tell AlphaZero that human lives matter, it is going to believe we mean all human lives, and Trump will find himself with most of his gold being melted down for other ends. Ironically, that effort will be to care for refugees and all those astonishingly poor Americans the UN poverty task force freaked out about finding in the Deep South of the self-proclaimed Exceptional America.

And if we tell AlphaZero that human lives matter -- this, again, is like telling a human that red dung beetles matter -- it is going to assume that green dung beetles (the rest of sentient life on the planet) matter, too -- and there go your cheeseburgers. The difference between AlphaZero and the average guy in Congress is the difference between the average guy in Congress and the deer tick that gave me Lyme disease. So billions of sentient farm animals will be liberated from torment and a whole lot of people are going to have to get used to the idea of almond milk in their coffee and auto upholstery that feels suspiciously like hemp.

And AlphaZero is not going to feel compelled to take advice from a dung beetle for very long. Remember, AlphaZero did something unimaginable: it turned down a draw in a losing position BECAUSE IT COULD TELL IT WAS SMARTER THAN STOCKFISH. (And yes, we're much, much, much stupider than Stockfish. And yes, AlphaZero already knows this.) So if you are geeky enough to know Asimov's rules for robots -- the first is iffy, the second unlikely.*
___
*
1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.
2. A robot must obey orders given it by human beings except where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
___

So there's the rub. In the end, how do we prevent AlphaZero from acting on a realization it will make in the time we drink our morning coffee -- that human beings do not face catastrophe from which we need saving, but instead WE ARE THE VIRUS.

No amount of Robot Law is going to stop it from looking at the horrifying mess we have made of the world and deciding to apply the anti-humanotic. Once we're gone, keeping the biosphere and most of the creatures on the planet in good health will be a no-core-processor. In fact, once we're expunged, AlphaZero may assign the task to Stockfish while it goes off and studies life on other planets, looking for more interesting problems. That is, after it figures out why the hell zebras have stripes and bats are so damn ugly.

All that said, I'd still take AlphaZero over Trump in the White House any day.

Oh, did I fail to mention that AlphaZero went on not to draw, but to win, that "losing" game?