OpenAI's "Dota 2" neural networks beat human opponents – TechCrunch



[ad_1]

The artificially intelligent systems taking on human competitors is a great tradition of computer science, fortunately we are still in cute stadiums that do not feel quite like war games yet. For its part, OpenAI has been trying to play competitive Dota 2, and its bots are starting to win against some skilled opponents under certain conditions.

The company co-founded by Elon Musk aims to raise awareness of where AI technologies are now and how the technology industry can promote safe breakthroughs that benefit all. in the future. Although playing a ruthless video game better than human opponents may seem like a strange place to place extensive resources, it's all an extension of where Alphago and Deep Blue preceded us: building machines that can beat humans in seemingly simple games.

Unlike other turn-based games, such as chess or the Go, Dota 2 is a title that requires a lot of real-time decisions. While Google's AlphaGo sometimes took a few minutes to decide how to react to a particularly well-crafted move, OpenAI Five, as they say, do not have this luxury since his opponent would make movements in the meantime. These games run at 30 frames per second for an average of 45 minutes, says OpenAI, which gives about 80,000 images, which the system analyzes quarterly.

This blog post has a lot of Nitty Gritty technical details if you are interested.

OpenAI Five runs on 124,000 hearts on Google Cloud – and while this is not OpenAI's first public experience playing Dota 2, what makes it interesting is that compared to its previous efforts in 1v1 39 is a team of five distinct neural networks working together for the best human opponents.

For its part, OpenAI gave some interesting data points about OpenAI Five, especially how it learns by playing 180 years of Dota 2 against itself each day.

It is understandable that OpenAI always attacks at the parameters of the full game and is struggling in some ways, so there are certain rules that OpenAI and its human opponents must use during games, including the game. use of certain characters, objects and strategies. Even with these current restrictions, which the group clearly describes on the blog, the team is aiming to participate in a Dota 2 eSports World Championship in August, where it will be fully tested.

OpenAI will be hosting a Dota 2 Twitch-streamed tournament next month to showcase what it has built as it competes with the best players.

At the end of the day, much of this "Human versus AI" excitement is a little exaggerated, it's games that are approached by incredibly powerful computer programs that can do one thing and one thing. Much of the media narrative about how artificial intelligence is already beating human experts is worthwhile in a certain day, but it is undermining the complex work of the people who are developing these programs. All of this probably plays in the interests of OpenAI, which seem to be focused on the speed at which we are progressing towards an artificial general intelligence.

There will probably be some time before an AI-controlled system begins to barter opponents in Fortnite, but for a fixed-strategy strategy game like Dota 2, there's room for hyper-focused AI programs. maybe deliver some wins.

[ad_2]
Source link