Google Uses StarCraft II Game as Testing Platform for AI

Google’s DeepMind has been making use of Blizzard’s StarCraft II Game as a testing platform for AI as well as important machine learning research. It’s not just Google that can benefit from this platform either; it’s open to anyone who wants to make use of it and is available worldwide. The company commented, “We’ve worked closely with the StarCraft II team to develop an API that supports something similar to previous bots written with a ‘scripted’ interface, allowing programmatic control of individual units and access to the full game state (with some new options as well).”


The two companies are still working together now, this time creating “curriculum scenarios.”  These are tasks that get more difficult as they are completed and are designed to allow AI researchers to get the system up and running, benchmark algorithms, and advance. DeepMind says that video games are one of the best ways to both develop and test AI and are on a mission to develop programs that can solve any problem no matter how complex.

DeepMind says, “Games are the perfect environment in which to do this, allowing us to develop and test smarter, more flexible AI algorithms quickly and efficiently, and also providing instant feedback on how we’re doing through scores.”  As well as using StarCraft II to test their AI, the GoogleMind team has also used 2D games on the Atari as well as more complex games like Go. But, with StarCraft II, in particular, the AI has to solve real-time problems such as when and how to farm minerals and gas; when and what buildings to erect; and exploring unseen areas of the map and remembering the route.

Researchers at DeepMind have also been busy developing an AI that can store knowledge such as the London Underground so that it can easily navigate within it. It uses a differential neural computer (DNC) to train its memory through a process that involves comparing the correct answers to its own answers and each time getting one step closer to getting the answer right itself the first time.  The DNC actually has an accuracy of almost 99% after the external memory has been added.

AI is moving forward, and that is partly down to GoogleMind and its fantastic team of researchers.  As well as working on all of the previous projects mentioned, the team is also in the process of developing a deep neural network that can produce a human-like text to speech system. GoogleMind’s machine learning will now be used in partnership with the U.K.’s National Health Service (NHS) to experiment using it with radiotherapy treatments to free up more time for doctors.



More News To Read

Comments

comments