In March, the A.I. platform AlphaGo beat a human champion at the (notoriously complex) game of Go. Later this year, at the 2016 Computational Intelligence and Games (CIG) Conference, bots will leverage machine-learning algorithms to play “Doom,” the classic first-person shooter. Now comes the next stage in the machines’ inevitable overshadowing of humans’ game-playing prowess: Google’s DeepMind has learned to beat “Montezuma’s Revenge” (screenshot above). If you’re a child of the 1980s, you may remember “Montezuma’s Revenge,” a difficult side-scroller in which an explorer navigated through an ancient temple, jumping over rolling skulls and figuring out how to grab keys to unlock new levels. Unlike board games such Go or chess, side-scrolling video games demand split-second timing and a little ingenuity if you want to succeed. How did Google’s DeepMind division, tasked with building A.I. platforms that rapidly learn and adapt, build software capable of guiding a digital character through all manner of obstacles without splattering everywhere? Simply put, it built a sense of “artificial curiosity.” If the software “wants” to survive and explore, it will keep trying and dying until it figures out the patterns necessary to make it through. The benefit of older games like “Montezuma’s Revenge” is that every element is pre-programmed, as opposed to procedurally generated (as with many modern games); that means a rolling skull will always roll across a particular floor with the same timing, no matter how many times the player encounters it. If you tried using DeepMind’s software to beat a procedurally generated game with random elements (such as level maps or enemy movement patterns), the software’s avatar would likely have a difficult time surviving. If you’re interested in learning more about machine learning, check out Dice’s article on breaking into the field. Data scientists and software engineers are particularly suited for jobs that involve building or refining machine-learning algorithms; soft skills are also a must, as a lot of A.I. projects are also cross disciplinary. Gaming skills may also become a nice-to-have; you’ll never know when your current A.I. project will want to challenge you to a match. Google A.I. Is Getting Better Than You at Gaming
In March, the A.I. platform AlphaGo beat a human champion at the (notoriously complex) game of Go. Later this year, at the 2016 Computational Intelligence and Games (CIG) Conference, bots will leverage machine-learning algorithms to play “Doom,” the classic first-person shooter. Now comes the next stage in the machines’ inevitable overshadowing of humans’ game-playing prowess: Google’s DeepMind has learned to beat “Montezuma’s Revenge” (screenshot above). If you’re a child of the 1980s, you may remember “Montezuma’s Revenge,” a difficult side-scroller in which an explorer navigated through an ancient temple, jumping over rolling skulls and figuring out how to grab keys to unlock new levels. Unlike board games such Go or chess, side-scrolling video games demand split-second timing and a little ingenuity if you want to succeed. How did Google’s DeepMind division, tasked with building A.I. platforms that rapidly learn and adapt, build software capable of guiding a digital character through all manner of obstacles without splattering everywhere? Simply put, it built a sense of “artificial curiosity.” If the software “wants” to survive and explore, it will keep trying and dying until it figures out the patterns necessary to make it through. The benefit of older games like “Montezuma’s Revenge” is that every element is pre-programmed, as opposed to procedurally generated (as with many modern games); that means a rolling skull will always roll across a particular floor with the same timing, no matter how many times the player encounters it. If you tried using DeepMind’s software to beat a procedurally generated game with random elements (such as level maps or enemy movement patterns), the software’s avatar would likely have a difficult time surviving. If you’re interested in learning more about machine learning, check out Dice’s article on breaking into the field. Data scientists and software engineers are particularly suited for jobs that involve building or refining machine-learning algorithms; soft skills are also a must, as a lot of A.I. projects are also cross disciplinary. Gaming skills may also become a nice-to-have; you’ll never know when your current A.I. project will want to challenge you to a match. 
