Dominion Strategy Forum

Please login or register.

Login with username, password and session length
Pages: 1 [2]  All

Author Topic: DeepMind for Dominion  (Read 8087 times)

0 Members and 1 Guest are viewing this topic.

crj

  • Saboteur
  • *****
  • Offline Offline
  • Posts: 1477
  • Respect: +1644
    • View Profile
Re: DeepMind for Dominion
« Reply #25 on: May 24, 2021, 08:28:29 am »
+3

Go is more complex that Dominion.  Starcraft too (as AI can still only just crack the top 99.8 of players) But Go is (was) known as the most complex game in the world with the most possible changes in strategies and permutations.  However, due to its RIDICULOUSLY intricate set of rules Magic The gathering has recently passed Go as the most complex game in the world. 
I think you -- and perhaps also the article -- are conflating several concepts, here.

The Magic result shows that the game is uncomputable. That is, it proves that there are circumstances in which you can show a computer two potential states of a Magic game and it will be not just unlikely, not just very hard, but impossible for the computer to say whether or not state B can be reached from state A.

This was achieved by embedding a Turing machine in a game of Magic, because that's the thought-experiment machine Alan Turing constructed to prove the Halting Problem cannot be solved in all cases.

Go and Chess are not uncomputable in this rigorous sense. There are fewer than 20,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000,000 possible positions in Go, so a computer can work through them all one by one. Maybe not before the end of the universe, but that's beside the point.

That's a cute, possibly even important, theoretical result. But it says nothing about how any real-world game of Magic would ever play out.


The complexity of a game's rules are a separate consideration. Here, it's pretty clear in an intuitive, human, sense that Go is simpler than Chess, is simpler than Dominion, is simpler than Magic. The rules of Go fit in a couple of paragraphs, whereas the Magic rulebook is 250 pages. Edward Lasker once said "While the Baroque rules of Chess could only have been created by humans, the rules of Go are so elegant, organic, and rigorously logical that if intelligent life forms exist elsewhere in the universe, they almost certainly play Go."


A third question is how competitive computers are against human players. We know that computers now reliably beat humans at Go and Chess. We also know that machine learning can figure out how to beat humans at Go and Chess. We know computers aren't beating humans at Dominion yet; I don't personally know how good computers are at playing Magic right now. On the other hand, I'm not sure having computers beat humans at Dominion has received anything like as much research attention or computer power as the iconic problems of beating a human at Go or Chess. My intuitive sense, as a player mainly of Dominion and Chess but not Magic or Go, is that Dominion is a lighter, easier, less mind-bendingly cerebral game.
Logged

Honkeyfresh

  • Minion
  • *****
  • Offline Offline
  • Posts: 617
  • DSF lowest ratio upvotes-posts 14 yrs & counting
  • Respect: +228
    • View Profile
Re: DeepMind for Dominion
« Reply #26 on: May 26, 2021, 08:04:54 pm »
0

I imagine you can train an AI to play a particular kingdom really well. The trouble comes from variability in the setup. Chess an Go always have the same moves available; I don't know much Starcraft, but I imagine that the pool of units is still significantly more limited than the pool of Dominion cards.


Starcraft is kind of like World Of Warcraft on steroids.  It's got strategic elements but is in a complete different realm of strategy game than dominion and MTG.

from wiki

Quote
Blizzard Entertainment's use of three distinct races in StarCraft is widely credited with revolutionizing the real-time strategy genre.[6] All units are unique to their respective races, and while rough comparisons can be drawn between certain types of units in the technology tree, every unit performs differently and requires different tactics for a player to succeed.

The psionic and technologically adept Protoss have access to powerful units and machinery and advanced technologies such as energy shields and localized warp capabilities, powered by their psionic traits. However, their forces have lengthy and expensive manufacturing processes, encouraging players to follow a strategy of the quality of their units over the quantity.[7] The insectoid Zerg possess entirely organic units and structures, which can be produced quickly and at a far cheaper cost to resources, but are accordingly weaker, relying on sheer numbers and speed to overwhelm enemies.[8] The humanoid Terrans provide a middle ground between the other two races, providing units that are versatile and flexible. They have access to a range of more ballistic military technologies and machinery, such as tanks and nuclear weapons.[9]

Although each race is unique in its composition, no race has an innate advantage over the other. Each species is balanced out so that while they have different strengths, powers, and abilities, their overall strength is the same. The balance stays complete via infrequent patches (game updates) provided by Blizzard.[10]

StarCraft features artificial intelligence that scales in difficulty, although the player cannot change the difficulty level in the single-player campaigns. Each campaign starts with enemy factions running easy AI modes, scaling through the course of the campaign to the hardest AI modes. In the level editor provided with the game, a designer has access to four levels of AI difficulties: "easy", "medium", "hard", and "insane", each setting differing in the units and technologies allowed to an AI faction and the extent of the AI's tactical and strategic planning.[11] The single-player campaign consists of thirty missions, split into ten for each race.
Logged
"Rap game Julio Franco, Chuck Norris, Texas Ranger/ Ice on my fingers look like I slap-boxed a penguin." -- Riff Raff Proverbs 4:20

"Sometimes I say some things people may think are just outlandish, but I'm going to have the last laugh." -- Riff Raff  Exodus 6:66

Honkeyfresh

  • Minion
  • *****
  • Offline Offline
  • Posts: 617
  • DSF lowest ratio upvotes-posts 14 yrs & counting
  • Respect: +228
    • View Profile
Re: DeepMind for Dominion
« Reply #27 on: May 26, 2021, 08:08:28 pm »
+1

At this point, with the number of different things AI's been made to do, I think it's highly likely that you could make an AI to do Dominion too.

"But at what cost" is the relevant question. A hobbyist tinkering at some code on the weekends - yeah, probably not. Google gives the project $50 million dollars and a team of AI PhDs? Sure, of course, I don't see why not.

I have no idea whether Google's "DeepMind" is the AI to do it or if you need a different one, I don't really have enough technical know-how about how exactly DeepMind works to judge how well it would fit Dominion.

My brother actually until last year worked on Deep Mind.  He's now running an AI lab at a  major university.  I was thinking of asking him to assign this kind of a think to his students as a class project.  will report back if i anything comes of it...
Logged
"Rap game Julio Franco, Chuck Norris, Texas Ranger/ Ice on my fingers look like I slap-boxed a penguin." -- Riff Raff Proverbs 4:20

"Sometimes I say some things people may think are just outlandish, but I'm going to have the last laugh." -- Riff Raff  Exodus 6:66

Honkeyfresh

  • Minion
  • *****
  • Offline Offline
  • Posts: 617
  • DSF lowest ratio upvotes-posts 14 yrs & counting
  • Respect: +228
    • View Profile
Re: DeepMind for Dominion
« Reply #28 on: May 26, 2021, 08:14:01 pm »
0

Yeah, it feels like there are three separate AI challenges, here:
  • Learn the rules of the game
  • Learn how each card works
  • Learn to play well
You could sidestep the first two by baking the rules into the AI. Or you could bring in some natural-language-processing AI and feed it the rulebooks and card texts. Or you could have it figure them out by trial and error.

Baking in a strategy for playing well would clearly defeat the point of the exercise. And I'm assuming one would want it to figure out strategy by trial and error, rather than by knowing how to read the dominionstrategy wiki. (-8
a lot of people learn by watching great people play.  I imagine that AI could do that.  Only instead of "watching" every game that has ever been played on dominion and spending millenia doing so, we  could just feed it all the game #'s/logs ever and process that in about 5 minutes.  I imagine just doing that alone should make an AI good enough to tangle with even fairly advanced players.  I imagine in all the data a lot of the things we think of as strategy get conveyed as data that AI could use to make the most optimal play...
Logged
"Rap game Julio Franco, Chuck Norris, Texas Ranger/ Ice on my fingers look like I slap-boxed a penguin." -- Riff Raff Proverbs 4:20

"Sometimes I say some things people may think are just outlandish, but I'm going to have the last laugh." -- Riff Raff  Exodus 6:66
Pages: 1 [2]  All
 

Page created in 0.405 seconds with 20 queries.