Pro Players Beat Dota OpenAI 2 Games In A Row

Dota OpenAI beat plenty of pro players in the 2017 but this year the pro players were able to beat the AI. Teams from Brazil and China were able to beat the AI, 2 games in a row and hence the best of 3 was won by the humans and left no room for the third game. The team from Brazil is ranked 14th best in the world and while the team was able to beat the Dota OpenAI this time, chances are this won’t happen again.

The AI learns from its mistakes and also learns how the players play the game. This is where it has the edge. Plus it is not going to make any kind of error that can be expected from a human. OpenAI co-founder and CTO Greg Brockman congratulated the team on the win and was asked why there was no third match and he cleared things by saying that the best of three was already won so it did not make sense to have a third match.

Susan Zhang, a software engineer working on the OpenAI Dota project, explained one of the flaws. The following is what she had to say regarding the matter:

If the AI ​​meets a situation that has never been seen before, it is difficult. Adjust immediately. At the same time, during the training process, the robot will take up to 14 minutes in advance to determine what action to take. There is no mechanism for the robot to deploy a long-term strategy of more than 14 minutes.

Mentioned that there was an internal evaluation of the match and the team deduced that the OpenAI had a 50% chance of winning the game, but 2 out of 3 games is higher than that calculation. If that prediction was accurate, the OpenAI bot would have been able to win one of the games and we would have seen a third game that would decide who is better.

If you are new to the game then you can check out the Dota Plus service and it is worth mentioning that this years treasures featured in the new battle pass were the best in years. For more information regarding the matter stay tuned.

let us know what you think about the Dota OpenAI being beaten by professional e-sports players.