OpenAI is a non-profit organisation whose mission is to create a distributed safe Artificial General Intelligence(AGI), its purpose being to help and extend human capabilities. Although the founder of the project, Elong Musk, also founder and CEO of SpaceX and CEO of the electrical car company, Tesla, believes that AGIs are a threat to humankind, he believes their creation to be inevitable. So, in order to prevent a doomsday scenario, along with Sam Altman, he started this open source endeavour, in order to make it as widely available as possible, I guess similarly to the principles of blockchain technology.

Since their inception in 2015, the project has employed 60 researchers and engineers to work on the project. They recognise the rise of AI technologies, but predict that the first AGI will overshadow any emerging technologies these specialised systems can bring to the forefront. Thus, they intend to create the future, in order to predict it. To that end they attend and publish all information in conferences and blog and publish their open source software tools to ensure transparency and safety. Among their top contributors are Chief Technology Officer, Greg Brockman, and Research Director, Ilya Sutskever, machine learning expert, formerly at Google (thank goodness these pioneers are censorious Alphabet Inc employees).

Unlike most AIs, which borrow human expert knowledge and learning methods to improve in skill, OpenAI is a general intelligence. Thus, theoretically it can master any field without human input given enough time. One of the first logical steps would be to let it learn a complicated game. DOTA 2 was chosen, and it was initially dropped in the game world with no knowledge whatsoever about the game to duke it out with the game's AI and learn the basics on March 1st, this year. Of course, it failed miserably at first.

Above, you have the TrueSkill ranking (similar to ELO ranking) evolution of the bot. Human distribution is as follows, about 15% of players are below 1.5k MMR (match-making rating), 58% are below 3k MMR, and 99.99% are below 7.5k MMR. Upon learning the basics of the game, it was pitted against itself, thus having no human interaction. As you can see the progression is fairly linear, meaning its expertise has actually increased exponentially (as getting better is harder and harder).

It's timeline is this:

• March 1st - first bot interaction with the game, learning to kite vs hardcoded DOTA2 bots
• May 8th - a 1.5k MMR tester reports that he is learning faster than the bot
• Early June - the 1.5k MMR tester is defeated by the OpenAI bot
• June 30th - the bot wins most of the matches against 3k MMR testers
• July 8th - bot inches a first win against an 7.5k MMR tester
  DOTA 2 International participation
• August 7th - bot beats former pro-player Blitz (6.2k MMR) 3-0, pro-payer Pajkatt (8.5k MMR) 2-1 and pro-player CC&C (8.9k MMR) 3-0
• August 9th - bot beats top tier player Arteezy (10k MMR) 10-0
• August 10th - bot faces 1v1 world champion Sumail and wins 6-0. Bot also goas against the August 9th version of itself and wins 2-1. Sumail claims the bot is unbeatable, feeling human, but also something else.
• August 11th - bot goes against former world champion Dendi and wins 2-0 against the crowd favourite. OpenAI also achieves a 60% win rate against the August 10th version of itself.
  
  Below you can watch the 1v1 match against Sumail, the world champion. It lasts less than 3 minutes, so give it a watch :)



I am by no means a DOTA 2 expert or anything, but I understand the concepts behind it as I do play a very similar game, League of Legends. This to me seems like an extraordinary achievement, probably more so than Google's AlphaGO beating the GO world champion a while back. This is for two reasons, Google's AI has been expertly trained by, thus is a specialist AI, not an AGI, and DOTA 2 has far more complex systems and rules to govern those systems than GO has. Even if there are more possible distinct GO matches than there are atoms in the universe, I believe DOTA 2 to be far more complex, there is not only just one way to beat an enemy, but there are a lot of modifiers to those tools, and there are out of game modifiers as well. Not to mention entities outside of any player control, and the fact that all those entities evolve according to player's choices. This is both amazing and scary at the same time, we might have just created Skynet, though it might not come at us in metal casings, just all across the electrical grid.

What are your thoughts? Is this wonderful or ominous? Share your opinions and feedback below, it is greatly appreciated!




Thank you for reading and keep on steemin'!


(clicking on any image will take you to its source)