Re: J-xx
Still not the same. For example, in RTS games, the NPC AI does take consideration of limited knowledge of the battlefield, unless when it is deliberately cheating.
Ok, I'll take this step by step...
The video game machine has 100% knowledge of the game universe since everything is a simulation in that game. Since a human wouldn't know everything, it "tells" the AI about the things that a human would know, and the AI acts accordingly.
The difference between this and the AI being a real robot is that there's no "lets create a map of our universe from sensor data" step. The map that the AI has is the game designer's guess of what a human in the situation would have, not what a robot in that situation would have. A real robot would have a hard time creating such a map from the available real-world (not game universe) data.
They don't need to think like humans after all. An insect for example, with vastly less computational power than a human, can perform far more complex aerial maneuvers.
No argument there. I brought this up because you claimed that a computer program is some sort of embodyment of human thought. That's not true.
Human face recognition is at the cutting edge of computer technology? Get serious. Its commonly used now even in commercial digital cameras.
No, digital cameras are able to guess that there's a face in the picture (and can be pretty easily fooled even about that). They can't recognize who that face is.
State-of-the-art software (not what's in your camera) _is_ able to recognize faces under ideal conditions. But this is cutting-edge, and pretty easy to fool. Add poor lighting, strange angles, or a little makeup, and even the most advanced software has trouble telling a human face from a monkey.
I brought up this example to contrast what computers are good at vs. what people are good at. "Chess" is an example of a problem that's pretty hard for a human, and fairly easy for modern computers. Face recognition (knowing your mother's face, for example) is something that's trivial for humans, but exceptionally difficult for computers.
And who told you that you need face recognition in dogfight AI? All you need is plane recognition, which can also be augmented by other sensors, recognition through radar or thermal signatures for example. Plane recognition through digital IIR is here already, even in missile seekers like AIM-9X. MAWS, RWRs, IFF, they already designed conclude whether a target is friendly or enemy and do not need the pilot to make that kind of decision.
Face recognition was just an illustrative example.
All of the systems you mention can be fooled, and are fooled fairly often. That's what ECM, flares, stealth, and camauflage paint are for. They're accurate enough to still be useful, but only because there's a human making some basic decisions.
I know it's an old example, and the technology is better now, but remember in 1988 a U.S. Aegis system shot down a civilian Airbus. A large commercial jet taking off (ascending) was mistaken for an F-14 on an attack run (descending) and shot down. There was a lot of human error involved here to, but it's still a pretty dramatic illustration of the limits of technology.
A modern system is a lot more advanced, but even so, this was a single plane that looked nothing like an F-14 and wasn't doing anything to try to fool the Aegis. No ECM, no stealth. Imagine how many mistakes will be made in a battlefield with lots of enemies and friendlies, each trying to fool the other.
... Ami.
