In the 80s, as I understand it, they used AI to look at satellite photoes from Europe to predict where Russian armor would be hiding in the forests. When Desert Storm rolled around, they tried to use it to see where Iraqi armor was hiding in the desert. It failed miserably.
AI was counting the leaves in the European satellite photos and deducing where armor was hidden based on the number of leaves. That doesn’t work in the desert.
AI has a lot of potential, but those that believe it can already “think” are mistaken, IMHO. AI is an encyclopedia thatt can open and read itself. It cannot dream new adventures.
Judging AI by what happened in the 1980s is like judging our capacity for space flight by Otto Lilienthal.
At an neural networks class I took at the Air Force Institute of Technology back in the 90s the instructor described a similar problem. They took pictures of a tank they had in various poses: out in the open, hidden in trees, partial shots, from various angles, etc . Then they took pictures of similar backgrounds without the tank. The problem was they took the tank pictures on a sunny day and tankless pictures on a cloudy day (I might have reversed those). Those pictures were then used to train the neural networks and it perfectly recognized sunny = tank, cloudy = no tank. When presented a picture of a tank sitting in the open on a cloudy day it said “no tank”. But it did have a 100% accurate response it recognizing sunny vs. cloudy.