Home

 › 

Articles

 › 

Scientists Teach Mario To Think, Feel, and Speak For Himself

Scientists Teach Mario To Think, Feel, and Speak For Himself

A.I. Researchers at the University of Tubingen in Germany have found a way to make a video game character think for himself. The character in question is Mario, moving through a simple 2D Mario world.

Mario was programmed with a few simple desires and emotional states. If he is hungry, for example, he will seek out coins to get full. If he is curious, he will attempt to interact with his environment, going places he hasn’t before, trying to figure out what is in power blocks, and more. They also track his happiness and several other emotional metrics as well, which effects actions he will take while “idle” for example.

Mario collects information about his environment as he explores. For example, if he lands on a goombah and it dies, Mario internalizes that the reason the goombah died might have been because he landed on it. He can then test this out by jumping on other goombahs and trying to interact with goombahs in other ways, until he eventually figures out with 100% certainty that yes, when a goombah is jumped on, it dies. He does the same with things such as power-up blocks, realizing that he can get coins or power-ups from them by jumping into them.

Then, Mario examines this raw information he has stored and uses this to parse actual English language sentences. This is harder than you think, as he needs to examine his internal probabilities to use words like “certainly” or former recorded trajectories to use words like “land” and “jump.”

It’s a very fascinating study. To see more about it, check out this Youtube video detailing the project.

Source: Science Alert

To top