One of the tropes that's always bugged us about AI and neural networks in movies and television is the idea that they can potentially miss a shot or the like - when the reality is that they never would.
They're AI. They're perfect, unyielding learning machines that wouldn't be able to aim for anything less than utter perfection. But how they would interact with something far less sinister? What about something like Mario Kart?
That's precisely what AI programmer / YouTuber SethBling set out to do, and quite honestly, the results are kind of disturbing. Using a recurrent neural network - which means that it stores information and uses that further its understanding - the system was known as MariFLOW and was placed into a number of races in Mario Kart to see how it did.
We'll say no more, except for the fact that AI and neural networks are now terrifying to us in a new way - and we didn't even think that was possible.