Human ingenuity has caught on to artificial intelligence.
Undoubtedly, there is great potential in the use of robots and artificial intelligence for military purposes, it is no coincidence that DARPA (Defense Advanced Research Projects Agency) under the Pentagon devotes significant resources to this kind of research. One of the agency’s robots, for example, was specifically trained to recognize and identify people, but its designers did not prepare it for what happens when people behave in a non-human way, i.e. try to outsmart it.
Paul Scharre, a former army veteran and former Pentagon analyst, will only launch his book on the subject entitled Four Battlegrounds: Power in the Age of Artificial Intelligence on the market on February 28, but some exciting details from it have already appeared on social media.
One particularly funny story shared by Shashank Joshi, editor of The Economist, reveals that DARPA tested its robot’s human-recognition algorithm with marines. For six days, the marines and engineers walked, hustled and circled around the robot so that it could learn to recognize human movement. On the seventh day, the engineers placed the robot in the middle of a traffic circle and instructed the Marines to approach it and then touch it without being detected by the structure. You might think that after six days of training, DARPA’s robot would have easily spotted the best of the navy, but all eight of them successfully completed the task, even with tricks that couldn’t fool a kindergartener, but the machine could.
These aren’t the marines you’re looking for. pic.twitter.com/qMwzJblQwp
— Shashank Joshi (@shashj) January 18, 2023
Two somersaulted and approached the robot, which could not deal with its unusual form of movement, another pair sneaked closer, hiding in a cardboard box, just like how you can trick the guards in the Metal Gear Solid games. Someone was carrying a fir in front of him, and since no one had told the robot that Ents only existed in Tolkien’s novels, he ignored the walking tree.
Of course, all of this does not indicate that the robot or the algorithm is unusable, it only limits its capabilities in terms of the data it works with. And if – staying with the current example – they don’t teach him that a person can indeed get around by somersaulting, then he will never figure it out on his own.
