PDA

View Full Version : Machine Learns How to Play Pitfall



DA
06-25-2009, 12:12 PM
Watch the video above, and then give your nearest loved one a hug, because this could be the harbinger of humanity's ruin. As GameSetWatch reports (via Kotaku), researchers at Rutgers University have developed a new method of reinforcement learning using "Object-Oriented Markov Decision Processes." And to demonstrate this complicated-sounding breakthrough in an easy-to-understand way, there is the video above, which shows how a machine used this method to learn how to play Pitfall on an Atari 2600 emulator.
As you can see, on the first run, this frightening artificial intelligence appears to make a few mistakes as it stumbles through the obstacles in the game. After doing so, however, the program appears to remember what it has done, and uses this information to clear the obstacles with ease on the second pass. All on its own.

As impressive and horribly disconcerting as this is, the most disturbing portion of the video is at the end: After clearing the obstacles, this robotic mind appears to celebrate, making the character jump up and down over and over in what can only be described as a demonstration of joy. So not only does this thing think and learn, but it apparently also feels. This should however be considered good news, because it means that if a day comes when this machine and its kin rise up against humanity, we know psychological warfare will be effective.