In a competition against some of the best players in the world, Elon Musk backed research group OpenAI has come out victorious in head-to-head combat in the popular DOTA2 game. It's a great achievement and occurred in short order for the group, but the question remains, how significant is this for AI research in general? Is this more impressive than DeepMind's accomplishments in the game of Go? I would argue no.
While DOTA2 is a fast-paced and complicated game, there were some key advantages that OpenAI had in getting their bot to this level. First, they had pre-programmed the items that the bot would use. With a build in place, it significantly reduced the search space for the bot and made training easier. Another thing, they only had success in 1-v-1 matches. DOTA2 is often played as a 5-v-5 game which requires coordination and teamwork that a 1-v-1 greatly simplifies. The AI system only worked for one character - Shadow Fiend - out of the dozens available. And finally, the bot had a distinct advantage by being patched into the game system rather than interpreting the game via the screen (like the Atari game accomplishments). This enabled the bot to have direct access to distances and other information that would not be available to human players and, most notably, enabled the bot to execute attacks as soon as it got within range of its opponent.
All in all, the OpenAI feat still remains a huge achievement, but I don't think it represents the same level of progress that we have seen in other areas. Still, the team is working on this game to this day and will likely be able to build on this early success in the future to claim DOTA2 in all of its complexity for AI.
artificiallyintelligent1@gmail.com
