A new problem in social choice theory

S.L., a loyal MR reader, asks:

Suppose aliens came and demanded one or more chess games, one move per
day, with humanity (maybe these aliens):

http://gambit.blogs.nytimes.com/2010/11/18/interview-with-the-president-of-the-world-chess-federation-part-3/

where humanity would be destroyed on losing, and a draw would lead to a replay.  One move per day.  What would be the best way for humanity to run our decision process?  How would we actually run our process?  ("The committee must include at least one grandmaster from each UN country…")

I opt for Kramnik, Anand, Kasparov, and Carlsen, using Rybka and two or three back-up machines, such as Junior (which seems to find different moves).  The Grandmasters vote, with 3-1 or better required for a move, but all four deferring at least initially to the single human agenda-setter and Rybka as tie-breaker.  They would be playing a version of advanced chess.  The goal is to have some different perspectives, but not so many that it becomes unmanageable.  Most of the time the computer would get its way, but the four would consider those instances where Rybka suggests misleading moves.

Or is there a better decision-making process?  Could a decision-making process like this be used for other dilemmas?

Comments

I vote for tipping the board over if it looks like we're losing.

Are we white or black?

Kasparov wrote an excellent article about computer chess (http://www.nybooks.com/articles/archives/2010/feb/11/the-chess-master-and-the-computer/?pagination=false ) in which he describes a competition where two weak human players with weak computer assistance were able to beat both very strong chess machines, and also very strong humans working with strong machines. He suggested that this showed the importance of the decision-making process. In his words:

"Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process."

Slightly off topic, you might enjoy Kasparov's talk at Google:
http://www.youtube.com/watch?v=9hf31xOhchY

Draw after draw would be akin to winning because the rest of us could carry on with our lives while they duke it out. In fact, draw after draw would be what I would aim for.

Should they play to draw? one move per day with computer chess improving at a rate of about 50 elo a year that means if you could draw a few long games in a row you would fairly rapidly increase your rating. Add to that that after a few months you would have every supercomputer and spare cycle on a desktop in the world working on the chess game.

So at least initially should you find the best way to draw a 100 move game to bring more resources to the problem in game 2?

It's not inconceivable that given all the world's computing power and a serious effort by the AI community we could use computers to play a perfect game even without any human contributions.

Surely the aliens would have more advanced computer technology than us, so turning it over to computers might just guarantee defeat. Maybe our best weapon would be a really good, really unpredictable human player--and not necessarily the best human player--because the aliens would probably not have a deep understanding of human psychology.

If they were illegal aliens, and they landed in Arizona, I would ask for identification and send them back to where they came from.

Otherwise, we'll have space aliens challenging us in games of chess and taking over good paying jobs.

Some of them might even get pregnant and their children would become US citizens.

I've seen it in the desert, or so the governor said.

Build that g-damn fence.

It seems probable to me that in such a situation:

1) The aliens would already know they were certain to win, or
2) The aliens would turn out to be horrible at chess and we would easily win.

The likelihood the aliens would challenge us to such a match without either 1 or 2 holding would be infinitesimally small, therefore we would be better off not wasting all our computer and intelligence resources and just using the best of what we've already got. Other threats against mankind's existence should probably be treated similarly.

The game should be decomposed into two sub-games with different objectives.

2 grandmasters compete against each other, one playing the Alien and the other the human representative--that group plays to win

2 other grandmasters play each other, one playing the Alien and the other the human, with the purpose to draw.

Whichever team is the most probable winner as voted on by both determines the move the group will make.

A lot of people are reading too much into the chess aspects of this. I'm guessing SL is more interested in the "How would we actually run our process?" part which has some more realistic applications with climate change, north korea, etc. E.g. do you think China is going to let 2 Russians, an Indian and a European waltz in and call all the shots?

The process would probably end up something like this:

After the aliens move, an initial (from the previous days analysis) response is immediately chosen.
A few hours of initial computer and grandmaster analysis and the interim response may be updated.
As the day progresses teams of the top x hundred led by top 20 players review ongoing computer analysis and update the interim response as needed.
At the 20th hour the top players available that day (if it goes on very long people will need holidays to recover) convene for a few hours. The final move is chosen.
Something like the un security council (but with no veto) votes to accept the move.
As the world watches, Hu Jintao (or other world leader) makes his way with the official earth response toward the alien spaceship. Hu hands the move to the small figure. It's Trig Palin! He runs up to the spaceship to deliver the move to the aliens.

Should we even be trying to win? Is there a reliable strategy for forcing a stalemate? If so why not just play an abritrarily long series of games each ending in a draw?

Do we get to destroy the Aliens if they lose?

Salem

Great idea , wonder why no one has ever come up with it before. " Of the chess players , by the chess plaers for the chess players " has an immortal ring to it.

RR

I said "We can’t find BETTER way to play the chess in the economy."

I mean

We can choose variables (players). But the economy's performance will be inside (or on) of its own transformation supply curve.

I keep picturing this as a scenario where the aliens turn out to be an alien child playing around. The world is terrified by each move the alien child makes because the child's moves look worse and worse, convincing our chess experts that the intelligence behind them must be even greater than we could possibly imagine. As we near an endgame, which it looks like we will win, the child is called for supper by his parents and the spaceship takes off never to return to earth. We assume it was a forfeit, our lives have been spared and rejoice.

This whole discussion has a very "The Player of Games by Iain M Banks" feel to it.

Tyler,

The only reason your heuristic and the many that the readers have put through would work is because, at least until the moment of the aliens' arrival, the art of chess has been isolated from corruption due to political pressure.

The much more difficult task is if the aliens had given this challenge to humanity and said that they would return in 150 years to begin their series of games.

Then, all of a sudden, the game of chess has these political overtones that it didn't have before. How do you keep that from corrupting the system?

For your problem, I would more or less go with a committee kind of solution with thorough inputs from computers. We cannot do much better.

For my problem, I think that only a friendly AI will be able to resolve it.

Hey Tyler, for you, is talking about chess a form of signaling, or do you just do it because you find it fun?

We should go for that 4 move checkmate!

with distributed computing, there will be tons of security problems. Nuts / alien viruses interfering with the reported results.

And what if one of the grandmasters is replaced with an alien doppleganger? Obviously we need an oversight committee. Or maybe just stream all of their discussions online. I guess theoretically, it won't matter if the aliens overhear it.

I'm guessing SL is more interested in the "How would we actually run our process?"

Oh, in that case what will happen is that there will arise a new political movement that will see the aliens as the creation of a cabal that wants to create a world government, usher in universal socialism and erode American liberties.

A market would be a terrible mechanism for determining the next move. First, the average market participant won't be very good at chess, much less grandmaster good. They would therefore have to rely on other signals (price, analysts) to make their purchase, not what they thought was best. Also, the best move may be based on seeing 20 moves ahead when most people cannot see that far. As a result, it's possible that the cheapest move would be the best move as only the best of the best are able to see that it's the best.

Also, the rational strategy would simply be to buy the highest trading move, since under the rules that move is most likely to win, causing the price to skyrocket instantly. This strategy would most likely lead to instant death, as the market pursued a four-move mate.

Markets are also unlikely to be rational when the fate of human civilization is on the line, as profit expectations/desires would be warped.

Same goes for opinion polls.

Ray, it's an open question what an alien would want because they are aliens and we should not assume much about them.

You might have a good point, it certainly could go the way you suggest.

Or they could feel it was ethical for them to exterminate us after they beat us at our own game.

Or they could want to give us time to get resigned to the idea, and a good long time slowly losing a chess game would do that.

If we were the sort to put all our attention into a chess game instead of prepare for violent resistance, that could be convenient for them. If we for example exploded a bunch of doomsday bombs we might make the earth unsuitable for whatever purpose they had for it after we were gone.

They might feel this was an interesting way to get to meet us. They threaten us with destruction and then we win the game and they promise never to genocide us, and maybe then we'd feel better about them than we would if they'd never made the threat and then lost the game.

But these are all human ways to think. Aliens might intend something less familiar. Something more alien.

I don't know enough about chess to say what structure we should have for the decision process, but I think if we survived any real amount of time in the game we'd develop chess playing programs that would rapidly surpass all human play. If much of defense and research spending was devoted to making better chess algorithms we'd certainly make quicker progress. If we were already close enough to the aliens' abilities to stand a chance, we'd soon surpass them. Unless of course the aliens have solved chess, which I understand to be theoretically possible but computationally intractable, or have incredibly powerful chess computers, in which case we'd probably lose and be wiped out in a few months.

The aliens have advanced quantum computers. They've solved chess.

One thing's for sure: if we managed to draw the game out for long enough, we'd end up with better chess-playing computers by the end of it. Human decision-making would be obsolete.

Are we using the fifty-move rule?

"When all the worlds computers are being used to calculate chess moves, the programs can be much smarter than they are now."

It's been a long time since I've studied game playing AI, but I don't think today's dedicated chess-computers are processor-limited in a such a way that commandeering people's desktops would be that helpful. Although Mario's idea about running monte carlo simulations every day is a good idea and would make use of the extra CPU cycles. Chess becomes a very different problem when it's not real-time.

As to TC's post-ending question ("Could a decision-making process like this be used for other dilemmas?") I would refer people to matters of diversity in heterogenous ensembles in the Machine Learning literature. Maybe something by Kuncheva or Dietterich perhaps.

I'd lose and double down on Candy Land! :).

Comments for this post are closed