Re:OpenAI at the Dota 2 World Championships
by AndrewKemendo on Friday August 11, 2017 @09:06PM
(#14995744)
These types of MOBA games are a good precursor of military unit management.
Not really.
There may be some limited use cases where this beats traditional sandboxing, wargaming or milsim. However, the primary distinction is that DOTA and the likes have fixed rule sets (albeit large and complex), and a near infinite number of simulations that can be run, wherein the actual "battle" will conform to those rule sets and play variability.
It may not seem like it, but we (military) only have a tiny slice of possible data from the past few hundred years, that would be appropriate to train a DNN on. Not only that, there are few wars that conform to previous war patterns. There are some constants, and we study those, but they don't really give you what you need for troop movements and tactics the same way as is done in DOTA.
Not really.
There may be some limited use cases where this beats traditional sandboxing, wargaming or milsim. However, the primary distinction is that DOTA and the likes have fixed rule sets (albeit large and complex), and a near infinite number of simulations that can be run, wherein the actual "battle" will conform to those rule sets and play variability.
It may not seem like it, but we (military) only have a tiny slice of possible data from the past few hundred years, that would be appropriate to train a DNN on. Not only that, there are few wars that conform to previous war patterns. There are some constants, and we study those, but they don't really give you what you need for troop movements and tactics the same way as is done in DOTA.