OpenAI’s Dota 2 Bots Lose First Match Against Professional Team

Spread the love

A team of five OpenAI bots trained for the game Dota 2 were defeated in a match against a professional team of human opponents. Two more matches against other professional teams will follow in the coming days.

The Brazilian team called paiN Gaming cleared the bots of OpenAI Five in a 51-minute match defeat. The bots did well at taking down opponents; they managed to score quite a few kills through coordinated attacks. The bots also knocked out more opponents than the winners by the end of the match.

However, the team of professionals turned out to be better on a strategic level and partly because of that managed to win the match. An AI researcher from the British University of Falmouth said in a short comment on Twitter that the bots are very good at the moment-to-moment actions, but according to him they are still quite bad at macro decisions.

As with previous, recent appearances by the OpenAI bots, which saw them beat semi-professionals and a team mostly made up of professionals, the match against paiN Gaming was subject to a number of limitations. That is still necessary, because in the game Dota 2 there are still too many combinations and possibilities for artificial intelligence.

One of the limitations is that the bots reaction time is set to 200ms, which is mainly done to give the human players more chance. Also, only eighteen of the more than a hundred heroes could be chosen. In addition, the engineers behind the OpenAI bots agreed with paiN Gaming that the teams chosen should be balanced; the heroes were already chosen before the match.

OpenAI’s battle against paiN Gaming was a singles match that is part of a best of three. In the coming days, the bots will once again face two other professional Dota 2 teams. The OpenAI Five matches take place as part of The International, the annual Dota 2 tournament.

OpenAI Five actually consists of five neural networks. The bots have been trained for several months by playing against themselves every day; in an accelerated manner, that was good for 180 years of daily playtime.

You might also like