OpenAI software has defeated a world champion DOTA team, 2-0.
Dota 2 is a multiplayer online battle arena (MOBA) video game developed and published by Valve Corporation. It has a steep learning curve and is very complex. Two teams of five players compete to collectively destroy a large structure defended by the opposing team known as the “Ancient”
An AI-powered Dota 2 team defeated the reigning world champions in the game. This is a significant milestone for the development of OpenAI. The OpenAI Five is a virtual team, built on AIs that were designed to learn how to play Dota 2 on their own. And learn it they have.
The OpenAI Five Finals Saturday wound up in a rout for OG, who fell 0-2 against their AI opponents. OG is currently the top human team in the world, having beaten PSG.LGD in the finals last year to win The International 2018.
OpenAI chairman Greg Brockman said the OpenAI Five has played 45,000 years of Dota 2 in just 10 months, each time learning how to respond more effectively to the game’s constantly shifting tactical situation.
OpenAI is general-purpose AI that the company hopes wants to use in a wide variety of applications. Instead of being built to do specific tasks, it’s meant to be able to learn how to do just about anything. The development team has used Dota 2 as a ‘test bed’ to see how far the AI can push itself.
Watch OpenAI Five Finals from OpenAI on www.twitch.tv
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
If I understand it, the AIs in these tests don’t get any more information than a player would; They’re given the image, just what the player would see, and can output to the game controls.
don’t be so foolish, just because you saw it in hollywood doesnt mean it will happen, there is a huge difference between a game and the real world.
in real life, you have to see stuff, nothing is identified for you that green and grey bit over there could be a rock, some grass, a tree, or a sniper in a gillie suit.
In games, everything is coded as a seperate entity, for instance over there is a player entity is labeled noobslayer69, that other thing overthere is listed as a shotgun, the things surounding it are health pickups, everything is pre-identified. there is no way you can plug one of these scripted AI’s to atlas and create a deathrobot, we do NOT have the computer vision to compete.
but you don’t have to take my word for it, watch one of code bullets neural networing ai videos on youtube, note how he doesn’t program his AI to see whats on the screen, but re-creates the game each time so he can directly hook up the AI’s inputs/outputs to the program itself.
you know, when I watch an DOTA or starcraft game, I’m always reminded of blernball.
https://www.youtube.com/watch?v=j5n6V33OYwc
In practice the hard part of AI isn’t the abstract thinking, it’s the dealing with a messy real world. These games are kind of a dumbed down version of reality, much less input to deal with, far more limited options for action. A good playground to get the basic problems ironed out before moving to the real world.
That’s not how neural networks work, they dont remember each game, they adjust after each game. like polishing something smoother each time. That way they find a result that works best. instead of ‘remembering’ its the train data set who serves as backup to compete against, but its not part of the config of the final ai brain.
(which is also the reason they addapt to unknown situations.
45000 years of practice. And remembered all of it. This really does illustrate the relative Mr Puniverse nature of the human intellect. At least for this particular application of intelligence.
Aim bots already exist in first-person shooters and outperform humans. There’s also much less strategic and tactical thinking in those kinds of games. A game like DOTA contains significantly more objectives and actions, and so better emulates what a computer will have to deal with in the real world.
Why do you suppose some AI companies subject their nets to multi millennium training in cartoon like games instead of games like Medal of Honor & Call of Duty?
Such a net in the body of Boston Dynamics’s Atlas would be most useful in times of trouble, as well as profitable.