Anartificial news ( AI)system has succeeded inmastering classic video gamesfrom the 1980s , including iconic Atari titles such as Montezuma ’s Revenge , Pitfall , and Freeway . According to its Creator , the algorithms upon which the AI is based could one twenty-four hours be used to serve robots sail genuine - world environs such as cataclysm zones . Like catastrophe zone , many " hard - geographic expedition " games present a series of obstacles that must be debar and path that must be navigated in fiat to reach a destination or destination . Previous attempt to create an AI capable ofsolving such gameshave failed , due to the complexities of free exploration . For representative , many AI practice strengthener encyclopaedism – which involves rewarding successful action – for complete a project . The problem with this approach is that rewards run to be very thin , making it difficult for a system to achieve its objective . For illustration , if a robot is call for to execute a serial publication of complex actions to make a specified location , and is rewarded only upon get at its destination , then it obtain no feedback regarding the many case-by-case stride it must take along the way . Researchers can extend more " dense " reward   – such as rewarding each step a robot study in the right direction   –   but this may then stimulate it to make a beeline for its destination and flunk to head off any endangerment that may be in the elbow room . The only way to resolve this is by creating an AI that can actively search its environment . However , writing in the journalNature , the creators of this new AI explicate that “ two major issues have obstruct the power of former algorithms to search . ”The first of these is known as detachment ,   go on when a scheme does n’t keep a criminal record of expanse it has neglect to explore . For representative , when a robot reaches a forking in the route , it must pick out one path and chuck out the other . Detachment touch on to the inability of a system to afterwards recall that there was an alternative path that might still be worth research . Even if an AI could recall such missed opportunities , it would still break away into a trouble squall derailment , whereby it continually becomes side - track by its own impulse to keep exploring . Rather than lead straight back to that promising fork in the road , it look into each side - street that it encounters on the way , and therefore never actually makes it back to the branching . To overcome all of these issues , the researchers created a “ family of algorithms ” which they have call off Go - Explore . In a nutshell , this system works by continually archiving every state it run into , thereby appropriate it to think back the route it choose to discard at each level in thevideo biz . It is then capable to immediately return to any one of these promising saved country , thus overcome both disengagement and derailment . As a consequence , Go - Explore was able-bodied to pass the average human musical score on Pitfall , a game in which old algorithms give out to score any points . It also achieve a score of 1.7 million on Montezuma ’s Revenge , smashing the puny human world record of 1.2 million point .