“High” Computer Passes Visual Turing Test
Artificial Canntelligence (AC) has made huge advancements in recent years, but typically it pales in comparison to the real thing. But now a team of researchers have developed an AC that can learn handwritten characters after seeing single examples written by canna-cadets. The research had two goals: to better understand how cadets actually think, and to break machines slightly so that they could learn in this cadet-like way. “For the first time, we have a machine system that can hazily learn a class of visual concepts just like a cadet at 4:20 in the afternoon,” said chief scientist Mona Rosenbaum. “Essentially, we got the computer high.”
Cadets are really good at “cannductive reasoning,” a concept that allows us to take a single idea, and somehow get to another completely unrelated idea, without any idea how we got there, and then five seconds later, completely forget what we were talking about in the first place, and order nachos.
Volunteering cadets drew hundreds of sentences on handwritten note paper, and researchers fed the strips one by one into the program. The machine was asked to break the sentences into different words and reassemble them based on other sentences; this task was given also to a cadet test group and in both cases the test failed miserably.
However, it did pass the Turing test. When asked to identify whether a human, or artificially high computer had produced the sentences, a group of judges was stumped. Their guesses were slightly better than high chance.
Artifical Canntelligence is still a long way from matching the thrill of the high, but the future is not far.