Jump to content


Photo

Will artificial intelligence develop to the point where it can beat a human in stratego?


  • Please log in to reply
43 replies to this topic

#41 TemplateRex

TemplateRex

    Sergeant

  • Members
  • PipPipPipPipPipPip
  • 449 posts
  • Coat of arms
  • Silver Captain

Posted 11 June 2018 - 08:48 PM

This is my question to the 'know hows', is it possible to teach an ai to respond to known knowledge of his opponent? If so then the combinations of information a computer could learn could indeed beat the best of humans if it was designed by the best players. I'm sure it would take an atrocious amount of time to create but it could be done. Imagine an ai Nortrom that never suffered any type of mental fatigue or over stimulation. Even more amazing, think of the code? If Nortrom made this, the code would hold his deepest Stratego secrets. A genuine chest of Stratego treasure.

 

In other games, hand-coding human knowledge has proven to be almost impossible to give super-human performance. Older computer programs in chess and other board games were very efficient number crunchers with rather simplified evaluation functions to assign a judgment value to a position. Adding knowledge to such program has always slowed them down too much to benefit from better judgment.

 

AlphaZero showed that no special knowledge is necessary. Just lots of computers, learning from self-play (many more games than played in the history of humankind). The knowledge is represented as a hugely complicated non-linear function. Works for Go, chess, Shogi and also poker. No reason why stratego won't be next. 


I hereby grant explicit permission to all my opponents to record and publish my games as they see fit.


#42 Morx

Morx

    Lieutenant

  • WC Online Team
  • 684 posts
  • Coat of arms
  • Platinum Colonel

Posted 12 June 2018 - 07:49 AM

Yes responding to the known pieces is needed to achieve this and it is possible.The rest of what you are saying is much harder to program.

#43 TemplateRex

TemplateRex

    Sergeant

  • Members
  • PipPipPipPipPipPip
  • 449 posts
  • Coat of arms
  • Silver Captain

Posted 25 June 2018 - 07:21 PM

Today, OpenAI made an announcement that they have made significant progress on getting world class 5v5 Dota2 player:

https://blog.openai.com/openai-five/

 

The current version is not yet playing the full game, but in August they will challenge the world's best pros. Some nice quote which sounds ominous for Stratego (bold emphasis mine):

 

In the first games, the heroes walk aimlessly around the map. After several hours of training, concepts such as laningfarming, or fighting over mid emerge. After several days, they consistently adopt basic human strategies: attempt to steal Bountyrunes from their opponents, walk to their tier one towers to farm, and rotate heroes around the map to gain lane advantage. And with further training, they become proficient at high-level strategies like 5-hero push.

 

They used over 100,000 CPUs in parallel to get 180 years worth of training *per day*. I wonder what they could do to Stratego :)


  • Unladen Swallow likes this

I hereby grant explicit permission to all my opponents to record and publish my games as they see fit.


#44 Nortrom

Nortrom

    Colonel

  • Moderators
  • 1,990 posts
  • Coat of arms
  • Platinum Marshal

Posted 25 June 2018 - 07:30 PM

In the first games, the heroes walk aimlessly around the map " 

 

Sounds like the allies I get on that game all the time.


  • Unladen Swallow and TemplateRex like this
"Rock is overpowered, paper is fine" - scissors




0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users