- Former Pentagon political analyst Paul Scharre discusses global power and artificial intelligence in his upcoming book.
- He writes that the Marines trained the robots of the Pentagon’s Advanced Defense Research Agency.
- The robots, trained to identify people, were tricked by Marines who did somersaults and hid in boxes.
The state-of-the-art robots used by the Pentagon had a weakness that could easily be manipulated, according to a forthcoming book by a former political analyst: Although trained to identify human targets, the robots are easily fooled with the most outrageous disguises.
In his upcoming book, “Four Battlefields: Power in the Age of Artificial Intelligence,” former Pentagon policy analyst and military veteran Paul Scharre writes that a Defense Advanced Research Agency (DARPA) team trained its robots with a team of Marines for six days to improve their artificial intelligence systems.
Scharre did not immediately respond to Insider’s request for comment. “Four Battlegrounds: Power in the Age of Artificial Intelligence” will be released on February 28.
Shashank Josi, Defense Editor at The Economist, published a few excerpts from Scharre’s book on Twitter. In the excerpts, Scharre details how the Marines devised a game to test the intelligence of DARPA robots at the end of their course.
Eight Marines placed the robot in the center of the roundabout and found creative ways to approach it, with the goal of getting close enough to touch the robot without being detected.
Two marines did somersaults at 300 meters. Two more hid under the cardboard box, giggling the whole time. Another took branches from a Christmas tree and walked around, grinning from ear to ear pretending to be a tree, according to sources in Scharre’s book.
None of the eight were discovered.
“The AI was trained to detect people walking,” Scharre wrote. “Not people jumping, hiding in a cardboard box or masquerading in a tree. So these simple tricks, which a human would easily see through, were enough to break the algorithm.”
While it’s unclear when the exercises in Scharre’s book took place or what improvements have been made to the systems since then, DARPA robots have long faced obstacles to their performance, including poor balance and concerns about their potential to cause accidental killings due to AI acting on unpredictable way.
The Defense Department and DARPA did not immediately respond to Insider’s request for comment.