ActionAttention!
Earn from 0.001 to 0.5 bitcoin!
Earn bitcoin easily. Invite partners and earn from 0.001 to 0.5 bitcoin. Get from 5% to 70% in bitcoins from your investment partners. Sign Up - Register

You can see previous news in the old version of the news blog. Watch

AI is very easy to fool: soldiers were able to outsmart a US combat robot by hiding under a box.

Published: 2023-01-20

The military is more active than others are testing artificial intelligence technologies in an effort to reduce casualties among personnel. Significant resources are invested in such technologies by DARPA, one of the key research bodies of the US Army. But so far, the result, to put it mildly, is not encouraging, as experts involved in the tests admit. AI is very easy to mislead with the simplest tricks.


Sources say that a book by former Pentagon military analyst Paul Scharre is being prepared for publication. In the work Four Battlegrounds: Power in the Age of Artificial Intelligence, the author cites many cases of questionable behavior of robotic platforms with AI in the process of training and simulating combat work.


In particular, a case is described when one of the DARPA combat robots was trained to recognize moving fighters for a week. Marines and engineers were in motion next to the robot for six days, and on the seventh they came up with a game: the robot was placed in the center of the circle, and the task of the fighters was to get close to it from a distance of about 300 meters unnoticed by recognition systems and touch it.


All eight fighters who took part in the experiment were able to solve this problem. The two Marines approached, performing somersaults towards the robot, movements that the AI had not seen before. Another Marine "disguised himself as a spruce tree and walked like a tree" approached the robot, whatever that meant. Two other fighters approached the robot, hiding under a cardboard box. The latter tactic has already been seen in Metal Gear Solid by the notorious Hideo Kojima.


As the author summarizes, the term "artificial intelligence" is used heavily with a stretch. The AI platform only knows what is put into it and is not capable of any intellectual conclusions beyond the data received. “AI is only as useful as the data we provide it with,” the author writes.