DeepMind’s AI player has reached yet another milestone. Just when we thought we’d heard enough, DeepMind’s AlphaStar has reached grandmaster level in the real-time strategy game Starcraft II. What’s the big deal about this? Well, for one, it is the highest possible online ranking; and now that AlphaStar is grandmaster tier, this means that its AI is able to defeat about 99.8 percent of human players. That has to be something to chat about in the AI community as the historic achievement continues to make headlines.
The way the AI agents are trained is by a methodology called reinforcement learning. What this basically deals with how the agents should take actions in an environment to produce efficiency. AlphaStar trains its AI agents against different versions of themselves at a much faster pace to enhance the data, and thus improve its capabilities.This has to be a very significant achievement for the DeepMind camp and AI lovers worldwide. I believe since Starcraft is a game that is heavily competed, any achievement made in this game should be celebrated. This happening was similar to OpenAI’s agents, that also uses reinforcement learning to help reach a high tier status in five on five multiplayer game Dota 2. This feat was accomplished earlier in the year, in April by another San Francisco tech company.
Although DeepMind’s AI was performing on a game, it was not just to play around, the data will be dissected and analyzed, as well as the findings to be published in scientific journal Nature. While its success has been a great happening in the AI community, DeepMind’s recent showing, displayed the benefits of reinforcement learning. This will allow for some massive amounts of data to be studied—we all know how much we love data analysis! The hope is that will these findings will push forward the field and create advancements in the future, more than likely in the robotics area. Something like controlling a robotic hand, with a virtual simulator would be something applicable for the AI agents. After accelerating the amount of time, with the intent to reinforce, the AI would hopefully be able to actually perform motor skills of a full arm. The hope for programs like DeepMinds, is that maybe the research could lead to implementation in robots that would make everyday menial tasks a thing of the past for humans. Hopefully there are more breakthroughs in AI so that we may see some of these hopes and dreams in the very near future.