How do you define "real AI"?

Edited by - kill on June 28, 2000 4:49:52 PM

I can just imagine an army of robots getting caught behind trees due to bad pathfinding. Or the enemy sending 1 robot at a time to attack...

---
www.crazycrackerz.org

Real AI would be like the bots in the Matrix. They could think, learn, and adapt.

---
www.crazycrackerz.org

Intelligence is the capacity to acquire and apply knowledge.
There are tons of games whos enemies aquire the knowledge of thier environment and apply it to try and get an advantage. Just because it doesnt adapt to our environment in some stupid sci-fi cliche way, doesn''t mean it isn''t intelligent.

AI in games is just the application of RealAI theory in an environment constrained by CPU power.
Pathfinding can be crappy because it takes up too many cycles to do it perfectly. It''s also crappy because there are a limited amount of people who know enough about AI to be programming it. Yet there are plenty of developers willing to leave it to the last minute and give it to whoever''s not busy to do.

A lot of academic AI also takes the wrong path with subjects like predicate calculus and planning. Both of which are forms of logic, both of which are just higher simplifications of low level biological processes. By this I mean that neural networks may come together to bring about something as finite state as logic (i.e. something black or white, a binary system) but the underlying mechanism is still fuzzy and incredibly complex.

Personally for RealAI (tm) I think all you need is a genetically evolved complex neural network. Complex in the sense that it uses ideas like hormones to affect the brain not just neurons and connections.
A system as simple as this can pretty much solve any puzzle. Look at us, we are one such system and we work pretty darn well.

Mike

I think what StudentofSith is trying to get with ''real ai'' and ''game ai'' is that real ai is supposed to try and win, while game ai is supposed to be a fun competitor, but only exists to challenge the player. Would quake be fun if every enemy would just shoot you between the eyes from a block away (like in tekwar), or would begin a pixel perfect dodge the millosecond you fired a rocket? An Ai opponent has the computer perfection to crush a human on the action level (think about racing game ai, most are programmed to drive at half speed just so an imperfect human can compete), but so far not many have been able to do it, or look like their trying to do it on a thinking level. The Theif series showed good ''game ai'', guards reacted like humans, not mindless drones (except situations in Thief 1 where you could jump on a table and pick them off with arrows as they stood cursing you). RTS games are very slowling getting better, though they still rely on beating players on a resource management level (at which an ai is perfect) rather than a tactical ai.

Also I don''t think that developing an ''real'' ai means the end of the world. For an intelligence to go about killing everyone it would have to be programmed/instructed to do so or it would have to be consious(sp?) (which is a LOT farther off than a real ai) with the ability to create goals for itself. And I don''t think that an intelligence that would likely be able to explore thousands if not millions of ideas person would find any logic in destroying mankind (if it did then it would probably have to be in the right)

StudentofFish, imagine, at the end of every monologue I post, I''ve put the line
"So, what do _you_ think"
or
"I''m right, I''m right, slag me off if you think it isn''t so."

I know I put ideas too simplisticly anyway so please add any detail or disagreements that might spring to mind.

I think the same can be said of everyone here.

Mike