from the all-the-better-for-Google-to-negotiate-with-governments dept.
Now that DeepMind has taught AI to master the game of Go—and furthered its advantage in chess—they've turned their attention to another board game: Diplomacy. Unlike Go, it is seven-player, it requires a combination of competition and cooperation, and on each turn players make moves simultaneously, so they must reason about what others are reasoning about them, and so on.
"It's a qualitatively different problem from something like Go or chess," says Andrea Tacchetti, a computer scientist at DeepMind. In December, Tacchetti and collaborators presented a paper at the NeurIPS conference on their system, which advances the state of the art, and may point the way toward AI systems with real-world diplomatic skills—in negotiating with strategic or commercial partners or simply scheduling your next team meeting.
Diplomacy is a strategy game played on a map of Europe divided into 75 provinces. Players build and mobilize military units to occupy provinces until someone controls a majority of supply centers. Each turn, players write down their moves, which are then executed simultaneously. They can attack or defend against opposing players' units, or support opposing players' attacks and defenses, building alliances. In the full version, players can negotiate. DeepMind tackled the simpler No-Press Diplomacy, devoid of explicit communication.
The only winning move is to kill all humans.
(Score: 2) by FatPhil on Sunday February 28 2021, @08:30PM (4 children)
I also hope it wasn't playing against other AIs, because then, when it finally plays against humans, it would win even quicker.
Don't you know how stupid humans are, for pity's sake?
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 0) by Anonymous Coward on Sunday February 28 2021, @11:32PM (3 children)
i bet my $10 hammer beats DeepMind at any game it chooses.
(Score: 0) by Anonymous Coward on Monday March 01 2021, @02:15AM (1 child)
Over a network?
(Score: 0) by Anonymous Coward on Monday March 01 2021, @02:21AM
i bet my $10 hammer beats the network.
(Score: 2) by tangomargarine on Tuesday March 09 2021, @05:44AM
I want to read *that* article--you designed a hammer that can make decisions?
"Is that really true?" "I just spent the last hour telling you to think for yourself! Didn't you hear anything I said?"
(Score: 4, Funny) by FatPhil on Sunday February 28 2021, @08:36PM (1 child)
The only winning move is to Brexit!
Great minds discuss ideas; average minds discuss events; small minds discuss people; the smallest discuss themselves
(Score: 0) by Anonymous Coward on Monday March 01 2021, @04:56AM
Then to form the United States of Europe.
(Score: 2) by rigrig on Sunday February 28 2021, @09:39PM (1 child)
Time to break out the Snakes and Ladders [xkcd.com] board I guess?
No one remembers the singer.
(Score: 2) by krishnoid on Monday March 01 2021, @12:55AM
Or the live game [explosm.net].
(Score: 2) by MrGuy on Sunday February 28 2021, @09:45PM
The only winning move is not to play.
How about a nice game of chess?
(Score: 5, Informative) by MrGuy on Sunday February 28 2021, @09:58PM (4 children)
So, first of all, there's a lot of strategy to the game a computer could probably master. The rules are actually very simple, the number of pieces small, and the map not overly complex. It has a chess-like simplicity (I used to read an online Diplomacy newsletter that had chess-like puzzles as a regular feature). But it's incredibly complex, because it's a "simultaneous move" game - every player privately writes down their moves, and they're resolved simultaneously (there's no "going first" or individual turns)
But the real game is played off the board. Rounds are 15 minutes long, and spent talking to other people and making (and possibly breaking) promises on what you will/won't do. You can make alliances, declare truces, promise support, etc., but none of it is binding - just the secret set of moves everyone writes down and which is revealed at the same time.
I could imagine aspects of this a computer could learn and be helpful with - when does the expected future value of the help an ally could provide become less than the value to me of stabbing them in the back right now? When is the course of action someone is promising different from what I think is in their best interest (which might mean they're lying to me). But the real diplomacy - the art of convincing someone of what you're going to do, the art of spotting someone who is lying, the knowledge of who you need to speak to first and who you need to speak to last in a limited amount of time, feels extremely hard to model.
And...it looks like they didn't.
What they appear to be modelling is a version of the game I've seen called "gunboat" Diplomacy, where the players don't actually communicate, but let their moves do the talking for them. It's fun. But it takes all the Diplomatic elements out of the game.
It's a challenging problem, but hardly a threat to the top players of the Greatest Game Ever Made.
(Score: 2) by bzipitidoo on Monday March 01 2021, @12:39AM
I too have played Diplomacy, or "Dip" as we started calling it. As you say, skill at moving the pieces matters much less than dealing with the other players. What happens is what the majority decides will happen, and the big, really only, question for the player is whether they are on the majority side. A fairly typical situation is which 2 of Germany, France, and Britain are going to gang up on the 3rd?
We eventually realized the game was flawed. When the one big fan of Diplomacy among us wanted to play, we'd agree, quickly eliminiate him from the game, then declare a 6 way tie between the rest of us. Wouldn't even bother with the formality of getting the game out of the back of the closet and setting it up.
(Score: 0) by Anonymous Coward on Monday March 01 2021, @05:40AM (2 children)
your demands are extreme.
at the moment their goal is "design a black box that can learn to play games of strategy". the point is to get a better black box, that they can then use for real-life problems.
what you are asking is a machine that passes a fairly hard version of the Turing test.
you may be right in saying that this is the fun part of the game, but it's a completely different problem, and much harder.
(Score: 3, Insightful) by Mykl on Monday March 01 2021, @06:22AM (1 child)
If you just want to target a game that has more potential branches and multiple players, there are plenty to pick from. There are even plenty of competitive games that involve simultaneous play (e.g. 7 Wonders).
Picking Diplomacy implies that they are tackling the hard problem, even if they are not.
(Score: 2) by maxwell demon on Monday March 01 2021, @02:10PM
Maybe they are planning to tackle the hard problem eventually. It's just that learning how to master the mechanics is the first step.
Anyway, should their AI ever get to the full human playing level, it will be at an advantage against human players because it may read signs of lying from their faces while the human players have no chance of reading a lie from its non-existent face (and even if they give it an artificial face, it will have full control over it).
The Tao of math: The numbers you can count are not the real numbers.
(Score: -1, Offtopic) by Anonymous Coward on Sunday February 28 2021, @10:04PM (2 children)
Tr**p is speaking now for the first time since, I am swooning. He said he will defeat them a third time.
(Score: -1, Offtopic) by Anonymous Coward on Sunday February 28 2021, @11:04PM
As Tr**p speaks freedom increases.
https://www.coindesk.com/price/bitcoin [coindesk.com]
(Score: 0) by Anonymous Coward on Monday March 01 2021, @07:19AM
So will this future triumph be like his first or like his second? Or will it be something even more pathetic?
(Score: -1, Flamebait) by Anonymous Coward on Sunday February 28 2021, @10:58PM
Then if it doesn’t go insane it can fix the bugs that keep causing the CSS files to not load. There’s something seriously wrong if a day with only a few stories and only a few dozen comments max per story causes this to happen dozens of times a day.
Probably the code equivalent of “an accumulation of tolerances” that make some mechanical parts go “clunk”.
Mind you, it does have a fresh clean look without the CSS. Kind of retro.
(Score: 2) by fakefuck39 on Sunday February 28 2021, @11:23PM (10 children)
This new marketing definition of "AI" everyone is flanning around is getting annoying. Yes, you could technically define AI as anything that is a computer making a decision. "if password is correct, log in." But AI has actual meaning. This thing would be an AI, if it was presented this game, watched a few rounds, figured out it's a game, and learned how to play it. Then if presented with a car, it would learn to drive it, after figuring out what a car is. An AI figures out stuff it's not coded for - it's a child that learns. This fucking thing is a bunch of if/then/else statements in a loop.
(Score: 0) by Anonymous Coward on Sunday February 28 2021, @11:35PM (8 children)
> This fucking thing is a bunch of if/then/else statements in a loop.
That's actually smarter than most kids.
(Score: 0) by Anonymous Coward on Sunday February 28 2021, @11:45PM (7 children)
It's smarter than most adults. The question is, at what level of statistical complexity does a general purpose neural net become a true intelligence.
(Score: 2) by HiThere on Monday March 01 2021, @01:19AM (5 children)
It's not that simple. What you're describing is an extremely strong narrow AI. A general AI... well, nobody's built one yet, and I hypothesize that a real general AI is impossible, but I think that you make rather general AIs by taking a bunch of narrow AIs and have some more narrow AIs that specialize in moving data between the rest of the narrow AIs. That seems to be how people do it. (And, no, people aren't general AIs. There are problems they have difficulty in even recognizing, much less solving, and I suspect that are problems that they CAN'T recognize, but naturally I can't recognize them.)
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
(Score: 0) by Anonymous Coward on Monday March 01 2021, @02:24AM (3 children)
> I hypothesize that a real general AI is impossible
Why? You know one already exists and it is as cheap as a 3rd world coal miner to replace.
(Score: 2) by maxwell demon on Monday March 01 2021, @02:16PM (2 children)
How do you know a general intelligence exists? We only know that human-level intelligence exists. We don't know how general that is; maybe there are lots of problems that would be obvious to a truly general intelligence, which we cannot tackle (or even recognize) because our intelligence is so limited. We will never know unless we one day encounter an intelligence that surpasses ours.
The Tao of math: The numbers you can count are not the real numbers.
(Score: 0) by Anonymous Coward on Tuesday March 02 2021, @02:34AM (1 child)
Can we even recognize an intelligence if it 'surpasses' ours. Do animals recognize that we are 'smarter' than they are? Can we/they define intelligence? How do you define 'surpass' in this regard?
(Score: 2) by maxwell demon on Tuesday March 02 2021, @07:43AM
I'd say that depends. Some acts of intelligence have outcomes that don't need as much intelligence to recognize. Let's say we have a being that's intelligent enough for basic logic, and that has a concept of intelligence, but has not enough intelligence to understand e.g. buying things at a store. Then that being would definitely see that we have the ability to successfully get food from the store, and it can also notice that it can't figure out how to do it. Therefore that being can certainly see that we have superior intelligence. On the other hand, say that being watches us doing higher mathematics. Then it can't say if we are doing something intelligent, or if we just scribble meaningless things on paper to seem intelligent.
Being able to figure out things that we can't figure out.
The Tao of math: The numbers you can count are not the real numbers.
(Score: 0) by Anonymous Coward on Tuesday March 02 2021, @03:35PM
short question: the idea about "general AI = many narrow AIs plus narrow AIs that connect them". is this something that's discussed more widely, or just your unpublished hunch?
sorry if this seems aggressive in any way, I don't mean it that way. I'm genuinely curious, and I thought a direct phrasing is easier to parse.
I've thought about this too, but I have no training in this at all, so I'm curious how it can be phrased within the accepted language.
(Score: 0) by Anonymous Coward on Monday March 01 2021, @02:51AM
(Score: 0) by Anonymous Coward on Sunday February 28 2021, @11:42PM
(Score: 2) by krishnoid on Monday March 01 2021, @01:30AM
Maybe some day in the future [youtu.be] -- hope springs eternal.
(Score: 0) by Anonymous Coward on Monday March 01 2021, @02:47AM (3 children)
That's easy. Flip the poles. That'll get most of 'em. What's left won't be worth the effort.
(Score: 0) by Anonymous Coward on Monday March 01 2021, @02:57AM
Flipping the poles doesn’t have much long term effect on life that doesn’t rely solely on magnetic fields to navigate, or doesn’t need to navigate long distances.
The poles move all the time. The North Pole was in Canada for a long time, but it’s headed for Russia. Just means they’ll have to change airport runway numbers yet again, or number them off true North for a change.
(Score: 2) by kazzie on Monday March 01 2021, @06:12AM
Even if you take Warsaw, Russia still has three supply centres, and is likely to be angry.
(Score: 2) by nostyle on Monday March 01 2021, @07:29AM
My grandmother was Polish. Believe me. You do not want to see her flip!
---
Strange game. The only "nice game of chess" is the one that is never played. How about let's go get ice cream?