I wouldn’t be surprised if one day AI systems ruled the world. While there are skeptics, most people in the technological industry remain pretty optimistic. And with good reason. A system built by researchers at the Georgia Institute of Technology can recreate video games by observing them for only 2 minutes.
The team did this by training the AI on footage of two distinct types of players making their way through Level 1 of Super Mario Brothers. One that adopted an “explorer” style of play and the other a “speedrunner” style, where they headed straight for the goal.
The system managed to rebuild an accurate representation of the game with only minor deviations. It’s impressive and also far less creepy than an AI creating its own language.
“Our AI creates the predictive model without ever accessing the game’s code, and makes significantly more accurate future event predictions than those of convolutional neural networks,”
Okay, so its capabilities are still sort of spooky, but useful, nonetheless. The program, which is algorithm-based, could be vital in pattern recognition, among other things. In the end, the model is an effective training method that can also be easily controlled by users — phew!