Marines tricked an AI designed to detect humans by....not acting normal
This is an excerpt from Paul Scharre's upcoming book:
Marines tricked an AI designed to detect humans by....not acting normal
This is an excerpt from Paul Scharre's upcoming book:
@CommieGIR
Alt text:
human with a weapon looks different.” To improve their algorithm, the DARPA team spent a week with a group of Marines out at a test site. For six days, the Marines walked around and the engineers refined their algorithm for detecting people. Phil said, “On the seventh day I said, ‘We're done. We're flipping it.’ ”
Now the Marines would try to defeat the AI system. They parked the robot in the middle of a traffic circle and the Marines had to approach it undetected
1/4
@CommieGIR
starting from a long distance away. “If any Marines could get all the way in and touch this robot without being detected, they would win. I wanted to see, game on, what would happen.”
“Eight Marines—not a single one got detected,” Phil said. They defeated the AI system not with traditional camouflage, but with clever tricks that were outside of the AI system’s testing regime.
2/4
@CommieGIR
“Two somersaulted for 300 meters; never got detected. Two hid under a cardboard box. You could hear them giggling the whole time.” Like Bugs Bunny in a Looney Tunes cartoon, sneaking up on Elmer Fudd in a cardboard box. “One guy, my favorite,” Phil said, “he field stripped a fir tree and walked like a fir tree.
3/4
@CommieGIR
You can see his smile, and that’s about all you see.” The AI system had been trained to detect humans walking, not humans somersaulting, hiding in a cardboard box, or disguised as a tree. So these simple tricks, which a human would have easily seen through, were sufficient to break the algorithm.
4/4