Author Topic: 1st Chess , then Go , now US Air Force , when D&D ?  (Read 1397 times)

Offline awaken_D_M_golem

  • Epic Member
  • ****
  • Posts: 7299
  • (un-) Amazingly Unproductive
    • View Profile
Re: 1st Chess , then Go , now US Air Force , when D&D ?
« Reply #40 on: December 07, 2017, 05:06:22 PM »
here We GO


A generalized version AlphaZero just got put out.
It learns on it's own, without previous human sources.
It beat the best a.i. versions of Shogi in 2 hours , Chess in 4 hours , and "AlphaGo Zero" in 8.

Fancy paper ---> https://arxiv.org/abs/1712.01815
News writeup A ---> https://venturebeat.com/2017/12/06/google-deepmind-repurposed-its-alphago-ai-to-beat-the-best-chess-and-shogi-bots/
News writeup B ---> https://www.theverge.com/2017/12/6/16741106/deepmind-ai-chess-alphazero-shogi-go

I don't pretend to understand ... but ... anything "quantize-able" (to paraphrase PLZ) about d&d, it'll get'r done.  Maybe soon too.
avatar#3 , gravitational lensing edition ... I'm way on the other side of the universe but look like pretty rings

Offline Archon

  • Full Member
  • **
  • Posts: 209
    • View Profile
Re: 1st Chess , then Go , now US Air Force , when D&D ?
« Reply #41 on: December 08, 2017, 09:09:33 AM »
Interesting. I wonder how much these systems invent their own internal meta-game, and if that effects how good they are at dealing with some more human-intuitive scenarios? Independent of the different strengths and weaknesses of the human and computer player.

The problem with getting a AI to play D&D is that D&D is fundamentally poorly defined. Both in rules (which often require rulings or disambiguations) and in objectives (which are kinda ??? in computation terms). Once computers have empathy and natural language and speech ect. down (i.e. notime soon), they will be able to take a crack at D&D.

Offline awaken_D_M_golem

  • Epic Member
  • ****
  • Posts: 7299
  • (un-) Amazingly Unproductive
    • View Profile
Re: 1st Chess , then Go , now US Air Force , when D&D ?
« Reply #42 on: December 08, 2017, 04:50:13 PM »
Stuff like i.p.proofing and anything simple math like a +1 blanket ability vs a +2 small use ability , is definitely fair game for a.i.

But yeah, so much of the game is semantics, english language logic + usage quirks, leadership, divinations, etc.


Chess based sites now have up longer write-ups about the AlphaZero games.
Tossing so much 1000+ year tradition aside, for tactics unimagined until these games happened.
avatar#3 , gravitational lensing edition ... I'm way on the other side of the universe but look like pretty rings