3 ms·

Can anyone who has a full context compare this to OpenAI's work on Dota 2? Which is more impressive, both in how far along it is and the relative game difficult

• ionforce 8y ago
  This feels more impressive since in SC2 you can pilot many units while in Dota you only (generally) have one. And Deepmind is playing (albeit a slightly older) stock StarCraft II while OpenAI had to play using a very weird ruleset (invulnerable carriers). Milestones that we have yet to see with Deepmind - generalize on maps - generalize on races - zero-style training
• b_tterc_p 8y ago
  They’re probably comparable and at similar levels of success. Talking about which game offers more entropy isn’t a good metric yet because neither AI seems to be trying to utilize the depth available to them. E.g. the Dota game had 5 AI working as a team. That feels like it should demonstrate additional competence, but it’s not really clear that they worked as a team. The Dota game allowed greater variation in starting conditions, but it’s not apparent that the Ai adapts it’s strategy to this well (e.g. hero choice). Both of them are capable of creating a basic strat and excelling at micro. Neither appears to have great depth of strategy.
• hughzhang 8y ago
  Starcraft II pros are probably much further from perfect play than Dota2 pros (simply because the game is harder). As a result, having perfect micro is a much larger advantage in Starcraft than it is in Dota 2.
• glun 8y ago
  Dota is a much harder game strategically, in Starcraft it's easier to compensate with mechanical skill. You can see this in the results. The Dota ai played a very dumbed down version of the game and still got trashed by the pros.