Equipping AI combat robots with weapons – what could possibly go wrong? How smart artificial intelligence really is is shown by an incident that is described in a new book that is due to be published at the end of February. Former Department of Defense senior official Paul Scharre reports on his previous work, which included strategies for unmanned and autonomous systems and new weapon technologies. In his book Four Battlegrounds: Power in the Age of Artificial Intelligence, he describes how a group of US Marines used kid tricks to fool a clever robot. An excerpt from the book is circulating on Twitter.
To do this, the Marines placed the robot, which was specialized in identifying people, in the middle of a roundabout. The robot was developed by Darpa (Defense Advanced Research Projects Agency), the research branch of the Ministry of Defence. The goal was to sneak up on the robot from a longer distance and touch it without it recognizing its attackers. In the end, it didn’t take any sophisticated fighting techniques or special weapons, just a cardboard box, some fir green and sporting skills to achieve this goal.
The Internet is full of hot IT news and stale Pr0n. In between there are always gems that are too good for /dev/null.
Attention, now comes a box
While some soldiers approached the robot with somersaults, others tried to hide in a cardboard box, as is known from cartoon series and computer games. Another staged himself as a “grinning Christmas tree”. The fact that the Christmas tree ran through the picture did not put the robot into an alarm mood.
In all eight cases, the robot failed to recognize the people, even though the test was preceded by a week of AI training at a test site. Even the giggles of the soldiers in the box didn’t affect the machine. The US research agency repeatedly challenges robots to test their possibilities. One thing is certain: After this flop, the robot was definitely “off in the crate”.
(mki)
