We're going to find out what happens when we give rocket launchers to AI
Machine learning takes a step up
We've been battling bots in video games for many years, but now we're taking things to the next level. An upcoming competition will see participants unleash their AI in a battle that relies entirely on what they see, with no special knowledge of the game's internal data.
Google's DeepMind has already built a program that can navigate an environment designed on Doom, making in-the-moment decisions based on what it sees.
The next step is to give these types of AI a rocket launcher and pit them against other violently-inclined AI, which is what will happen as part of the Visual Doom AI competition. Participants will need to use machine learning methods to create a "controller" that can navigate the environment and take down its opponents without knowing any of the game's secrets as a normal video game bot would.
AI will also be able to gather medikits and ammunition during each 10-minute round, with whoever kills the most enemies declared the winner.
There'll be another round that makes different weapons and items available, with a final test to take place on three maps unknown to the participants beforehand, just to make things extra tricky.
The results of the showdown will be announced between September 20 and 23 this year, during the IEEE Computational Intelligence and Games Conference. Want to get involved? You can read more here.
Via Popular Science
Get daily insight, inspiration and deals in your inbox
Sign up for breaking news, reviews, opinion, top tech deals, and more.
Hugh Langley is the ex-News Editor of TechRadar. He had written for many magazines and websites including Business Insider, The Telegraph, IGN, Gizmodo, Entrepreneur Magazine, WIRED (UK), TrustedReviews, Business Insider Australia, Business Insider India, Business Insider Singapore, Wareable, The Ambient and more.
Hugh is now a correspondent at Business Insider covering Google and Alphabet, and has the unfortunate distinction of accidentally linking the TechRadar homepage to a rival publication.