If you haven't tried playing Dota 2 you have to! Dota 2 creates a
unique experience for players who enjoy online strategic games. Online
strategic games usually rely on teammates who assist. Dota 2 has a similar
concept since it's played by teams of five that compete against each other.
In 2016, Elon Musk announced that an AI team was going to try out Dota
2. Very few people had expectations that the computer-developed system could beat
professional players in an online competition match.
By 2018, OpenAI Five entered its first Dota 2 competition match, but it
was unable to defeat the professional champions. However, the bots had at least
won several matches in the preliminary rounds. The affluence showed that AI
still had the potential to do the unthinkable.
Finally, in 2019, the AI bots came back and outmatched the 2018 Dota 2
professional champions. In a match that the spectators and commentators voiced
as neck and neck, the bots beat humans 2-0 in a best of three match. The win
avenged their 2-0 defeat back in January.
Human Collaboration
OpenAI Five showed that AI technology had more potential than anyone
could expect. Furthermore, after the match, the bots took part in matches with
other humans. AI showed that it could collaborate with other humans on a common
task. Participants who took part in the exhibition matches commented that the
AIs were more supportive and worked to ensure that the whole team was
supported. The bots controlled their heroes to come and assist the characters
controlled by humans.
Preparation for the Win
OpenAI Five underwent significant modifications after its defeat back
in 2018. The defeat highlighted critical aspects that the developers had not understood.
First, the AI needed to gain more experience playing for it to win
against human professionals. The practice hours drastically increased, and
during each practice session, the bots took the time to perfect the 18 heroes
it had. OpenAI Five had to perfect the heroes to their optimum levels to make
strategic planning easier.
Second, the bots had to improve their strategic planning. Efficient
planning required the perfection of their micro. The AI had to ensure that it
got the upper hand at the beginning of the game to make winning easier.
Lastly, OpenAI Five had to learn how to make decisions based on team
movements. Decision-making would be based on opponents’ real-time movements.
The bots had to calculate their next moves and respond accordingly to secure a
win.
Results
OpenAI Five had better micro compared to human competitors. The micro
focused on short-term wins that the computer anticipated. The short-term wins
ensured that the AI stayed on top throughout both matches.
The computer used unconventional methods such as buying back heroes at
the start of the game to hold an advantage. The strategic planning provided the
bots the confidence they needed. During both games, the bots calculated a
winning percentage of 90% despite the game appearing evenly matched.
Overall, humans are still better despite the loss considering that the
matches took place under a controlled environment. Some of the players did not
use their best heroes since they could only select from the pool of 18 heroes
that OpenAI Five had perfected.
