9

So, Deepmind is pushing for a human level Starcraft bot and Open AI just created a human level 1vs1 Dota bot.

Unfortunately, I've no clue what that signifies because I've never played Starcraft nor Dota nor do I have more than a fleeting acquaintance with similar games.

My question is what the difference between Starcraft and Dota is from a AI perspective and what scientific significance the respective super human bots would have.

BlindKungFuMaster
  • 4,185
  • 11
  • 23

2 Answers2

5

Those AI-learning programs may have very similar scheme. We are changing only inputs and possible actions (like "use skill" or "move here"). Starcraft AI must do a lot of actions and control many units. Dota is MOBA so bot should be good in positioning on map for example. Different opponents to destroy and target for win.

AI needs to play many games for learn to play game with some rules to find best moves in some situations/states.

Of course it's only my newbie-programmer opinion :)

Adrian Grygutis
  • 149
  • 3
  • 1
    I do think the drive at DeepMind is to morph "strong narrow AI" into Artificial General Intelligence, and games are definitely a way to approach this. In some sense AGI, without reference to "strength" can be regarded as consistent strength across a wide and ever expanding array of contexts. These context could be as simple as combinatorial games with slight difference in game mechanics, or wildly different games, such as dota2 and Starcraft. I'm upvoting because this answer is reductionist in a valuable way, getting to the core of the matter. – DukeZhou Aug 18 '17 at 15:54
4

In SC2, players have more control over every minuet mechanic (constructing buildings, resource mining and management, controlling minions...) in the game, thus putting more tactical responsibility on the burden of the player. In DOTA2, the player is only in control of the super-powered hero itself, and not much of the other aspects of the gameplay.

It is debatable if these options make the game "better" or more difficult as a result. But it is for certain that overall the search space of the problem increases much faster as the dimensions of freedom increases.

Of course, DOTA2 contains very many game mechanics as well (a ton of items that changes a lot of different stats, very many types of heroes which each have their own attacks, a variety of buildings scattered around the map, a shop for player items), but it seems that most of this complexity is focused around player engagements, which although crucial for the development of the game, lets us analyze a much shorter timespan of PvP as opposed to a whole game.

And indeed, the DOTA2 bot from OpenAI was restricted to a mid-lane fight as a singular hero with restricted items, hence restricting most of the complexity DOTA2 has to offer.

Perhaps I am overstepping with this analogy, but the OpenAI result is a bit like a solved chess endgame configuration, while SC2 would be like chess under the knowledge that neither players can see each other's pieces until they are in a position where you can capture them.

In short: SC2 is more tactical. DOTA2 is more arcade-like.

k.c. sayz 'k.c sayz'
  • 2,061
  • 10
  • 26
  • Nice analysis. Related: https://ai.stackexchange.com/a/3838/1671 *(I attempted an answer to a similar question without knowing too much about the dota2 mechanics, and took the line that "assuming both games have similar complexity in terms of mechanics...")* – DukeZhou Aug 18 '17 at 15:51