Can AI systems experience illusions as in human vision? For instance, look at the picture below (click on the image to zoom in.)
Most people see a hole, shifting to a pyramid, back and forth over time. Because of the shape of the shadows and the bright wall on the left, the brain "knows" that it is a pyramid, yet the sensors (the eyes) report conflicting information. Would an AI object recognition technique provide a different answer each time it "sees" it? Or after more training data is added to the system? I would think AI algorithms can be fooled, but in a different way than biological intelligence (the human brain.)
This picture was posted here, though I don't know if it is a real object, or an AI-generated image. Clearly, the illusion is caused by loss of data (or data reduction as statisticians call it), after projecting a 3-D object onto a 2-D image representation. The solution would be trivial if the angle, in this view, was different, or if we we offered two pictures from two different angles, or a video showing the object rotating.