The Gordian Tomapple
Reading Wired for War; not sure if I’ll end up reviewing it or not – 100 pages in it doesn’t seem to be too germane to my project. Had to quote something funny though from his “Robotics for Dummies” chapter and its subsection on AI. Discussing how bad AIs are at pattern recognition, Singer uses this analogy:
A good illustration is what some describe as the Apple-Tomato test. For a computer to tell the difference between an apple and a tomato is actually quite tricky. It could resort to all sorts of visual analysis, comparing the size, shape, and color. But soon the machine would find that in certain cases there would be overlap, so any and all tests, no matter how rapid, would be inconclusive. It could next proceed to taking samples, such as capturing its chemical makeup via a smell test, and then comparing the data to other known test subjects. Ultimately, it could only be sure beyond a reasonable doubt with a DNA sample, which would occupy a massive part of its processing power.
Then again, you could give it a knife and let it cut the thing in half.