[HN Gopher] Human-level play in the game of Diplomacy by combini...
       ___________________________________________________________________
        
       Human-level play in the game of Diplomacy by combining language
       models
        
       Author : jdkee
       Score  : 106 points
       Date   : 2022-11-22 15:47 UTC (7 hours ago)
        
 (HTM) web link (www.science.org)
 (TXT) w3m dump (www.science.org)
        
       | erehweb wrote:
       | "In a game in which dishonesty is commonplace, it is notable that
       | we were able to achieve human-level performance by controlling
       | the agent's dialogue through the strategic reasoning module to be
       | largely honest and helpful to its speaking partners."
        
         | bpodgursky wrote:
         | Well, yes, "largely" honest, until the critical moment when you
         | stab to take 18 supply centers to win the game.
         | 
         | You don't win Diplomacy by _staying_ honest, you win by
         | choosing the right moment to make the move and burn that trust.
         | It 's the 5% dishonesty that matters.
        
           | wolverine876 wrote:
           | It's interesting and I think significant that people are
           | uncomfortable with positive values, such as honesty, doing
           | good for others, responsibility to others, right (not wrong),
           | etc. We shoot them down, similar to how people instinctively
           | cut off a conversation going in an uncomfortable direction,
           | often without even realizing it.
           | 
           | I can't attribute that the parent - I don't know the
           | individual - but I expected a chorus of such responses. I
           | feel like I can see them squirming in their seats ...
           | 
           | But I feel it too. There's societal pressure _against_ doing
           | good. It 's often more socially comfortable to throw a
           | cigarette butt on the ground than to pick one up and discard
           | it. That's bizarre, it's perverse incentives, and we can and
           | should change it.
        
             | svnt wrote:
             | I think what you've identified is a societal immune
             | response to predictable morality being systematically
             | exploited by amoral actors to gain wealth and standing.
             | 
             | I also think there is a big gap between what most people
             | advise and what they ultimately do that biases toward the
             | right thing. That is, people indicate they will not be
             | moral in order to dissuade others who might try to exploit
             | it, but in the end they generally behave morally.
             | 
             | Personally I will advise other people to be cautious while
             | repeatedly leaving myself open to being taken advantage of.
             | It almost never happens, and you learn to identify those
             | who will.
             | 
             | Of course this all pivots when/if you enter corporate
             | leadership and part of your job is to use the
             | morality/comfort of others as part of the container for
             | getting your role accomplished.
        
             | bpodgursky wrote:
             | When people are concerned about AI alignment, they are not
             | concerned about an AI that is a pathological liar, lying
             | for no reason.
             | 
             | They are concerned about an AI which is believable day-to-
             | day but is secretly optimizing for its own goals, which are
             | only exposed when it eventually makes a deceitful move it
             | cannot disguise, but succeeds because it now has sufficient
             | control.
             | 
             | That's literally what the AI is doing here; being honest
             | until the moment it is able to stab everyone else and throw
             | them under the table.
        
             | zemvpferreira wrote:
             | It's perfectly reasonable to advise against good values in
             | a game that's won by people who strategically abandon
             | theirs. If you want to win, that's the strategy to follow.
             | 
             | Not necessarily a lesson for life unless you view life as a
             | finite game to be won. But in that case, it's probably the
             | best strategy as well.
             | 
             | I don't view life like that and I'm guessing you don't as
             | well, but I'd advise anyone to be cautious of their
             | interactions with people who play to win.
        
         | sdenton4 wrote:
         | This is one of the key insights in playing Diplomacy well. You
         | can turn honest opposition in the early game into trust, which
         | is helpful for flipping allegiances in the late game. ('Yes,
         | I've been trying to murder you the entire game, but I've been
         | honest about it, and never lied to you once. Now I see that
         | we've got a real opportunity to dislodge Russia and sweep the
         | eastern half of the board if we work together... What do you
         | say?') Dishonesty means people don't want to work with you,
         | because you are not a dependable partner. So you gotta make
         | sure that any lies you tell pay off /big./
        
       | k647634s wrote:
        
       | Metacelsus wrote:
       | Demis Hassabis had better watch out! (Besides being CEO of
       | Deepmind he is also a top Diplomacy player)
        
       | pesenti wrote:
       | Code: https://github.com/facebookresearch/diplomacy_cicero
       | 
       | Blog: https://ai.facebook.com/blog/cicero-ai-negotiates-
       | persuades-...
       | 
       | Site: https://ai.facebook.com/research/cicero/
       | 
       | Expert player vs. Cicero AI:
       | https://www.youtube.com/watch?v=u5192bvUS7k
       | 
       | RFP: https://ai.facebook.com/research/request-for-
       | proposal/toward...
        
       | gpvos wrote:
       | dup of https://news.ycombinator.com/item?id=33706750
        
       | evanb wrote:
       | Great, let's definitely teach computers to play a game that
       | requires strategic thinking, a robust theory of mind, the
       | capacity to lie convincingly to your enemies, and well-timed
       | betrayal, with the goal of world domination. Can't see any ways
       | for that to go wrong.
        
         | wellthisisgreat wrote:
         | One more: it's Facebook or all places teaching them
        
       ___________________________________________________________________
       (page generated 2022-11-22 23:01 UTC)