OpenAI is a non-profit AI research company, discovering and enacting the path to safe
artificial general intelligence. In their search for a complex video game, the developers came upon Dota 2 as the ideal choice for their tests. It was a decision taken after a careful survey of several titles on Twitch and Steam. Dota 2 is a very complex game with a public API which helped fuel this decision.
For the average Dota 2 fan, their first brush with OpenAI was during The International 2017 at the KeyArena venue in Seattle. Watching some of the best individual mid-laners losing to Artifical Intelligence was a first for them. But it was not really a fair comparison. The matches did not play out on the Dota 2 map as it is. Instead, there were several limitations to the hero pool, the runes, the ability to use certain items in order to facilitate the AI. It was a criticism that the AI developers took in their stride as they attempt to further develop their product.
In 2018, The developers tried to have OpenAI play several 5v5 matches with limited heroes. This was an attempt to expand the scope of the Artifical Intelligence beyond a single hero and limitations. Playing as a team requires a different style of play compared to a 1v1 scenario. There is the element of making the right item choices, moving together as a team against the opponents.
Open AI did not do well at TI8
The International 2018 was the first ‘big showmatch’ for Open AI Five. OpenAI 5 is a 5 man team consisting of Artifical Intelligence controlled bots in the game. The Bots won almost all the team fights, in particular, due to their near perfect micromanagement skills. But they had a very poor perspective and decision-making process when it came to the macro decisions within the game. There were several areas where the bots could actually improve tremendously.
Warding is essential to Dota 2 as it provides vision in several parts of the map. However, OpenAI has not been able to understand the concept of warding to its full extent. They would often plant wards in areas where they are already visible. It is a waste of time and money in a game where Observer wards have an upper limit.
Ganking is a concept where heroes leave their lane to try to surprise enemy heroes and attempt to kill them. This concept enables the heroes to gain XP and Gold by killing the enemy heroes. It is a way of ensuring that your team takes a significant lead early in the game. With the early lead, a team can work on it and further increase the gap between the two teams.
But OpenAI bots usually prefer to stick to their lane and not worry about other lanes till much later in the game. The
The OpenAI bots are great at what is seemingly mechanical skill for humans. They are better than humans when it comes to team fights, mostly due to their micromanagement skills. But the final objective of a Dota 2 game is to destroy the opponents’ throne. You build your advantages, levels and items by accomplishing several smaller objectives on the way.
For the Dota 2 AI bots, this is something that is definitely absent. They need to ensure a better objective gaming and using the advantage to secure more objectives. This is the kind of ‘decision making’ that is simply not present in the OpenAI bots, yet.
OpenAI will attempt to play a no restrictions Game later in 2018 or early 2019
The OpenAI team does not see the subpar performance at The International 2018 as a defeat. Instead, it is just another step as they attempt to further develop the Artificial Intelligence. The ‘Reinforcement learning’ technique requires OpenAI to play several games against itself and against other AI’s in order to learn and develop. In addition to this, they also have a memory element so that they do not fall into a non-ending loop in their learning process.
The OpenAI will be working on their aim of defeating a five-man team by the end of 2018 or early 2019.