It's the tale of John Henry for a new generation, with a bit of a twist: a new artificial intelligence system has beat a visual Turing test, essentially proving that it can more or less trick humans into thinking it's one of them.
Developed by scientists at MIT, the AI system was designed to understand and digest myriad data sets, all of which aided the robot in its ultimate task for the experiment: no, not wielding a hammer to beat a steam-powered drill as John Henry did, but drawing a letter.
Created by Alan Turing, the father of modern computer science, the Turing test was designed to test the capabilities of AI programs—and how human-like they could be—and is meant to generate identical responses through the simple medium of conversation. Or, in this case, nontraditional communication, more or less a take on Turing's original proposed rubric.
This particular test was more of a two-parter, which was overseen by an evaluator. For the first section, a number of respondents were selected to view a drawn character—that is, a character that looks like a letter from the alphabet, but isn't—and to copy it, adding their own variations to the figure. For the second section, the participants were asked to come up with their own letter that fit in with the grouping they had already been given.
What the respondents didn't know? Which among them was the MIT AI robot.
After their figures were completed, the group was then asked to discern which of the letters were drawn by humans and which were drawn by the AI system. In the end, the odds were 50/50 that the participants guessed correctly—the same luck one would have as flipping a coin.
"In the current AI landscape, there's been a lot of focus on classifying patterns," said system co-developer Joshua Tenenbaum, a professor in the Department of Brain and Cognitive Sciences at MIT, in an official statement released by the university. "But what's been lost is that intelligence isn't just about classifying or recognizing; it's about thinking."
"This is partly why, even though we're studying hand-written characters, we're not shy about using a word like 'concept,' " he added, continuing to explain the derived authenticity of the robot's shapes. "Because there are a bunch of things that we do with even much richer, more complex concepts that we can do with these characters. We can understand what they're built out of. We can understand the parts. We can understand how to use them in different ways, how to make new ones."
The research was presented in the journal Science, authored by Brenden Lake, now a post-doc student at New York University who earned his Ph.D. in cognitive science from MIT last year while on Tenenbaum's team, along with Tenenbaum and Ruslan Salakhutdinov.