reactive vs. predictive AI

[intently]

It's made me watch the behaviors of pets and other animals, for example, and realize that they may be closer to us in intelligence than we like to give them credit for.  The degrees of separation between language, toolmaking, and all the sophisticated stuff, and simply existing in an animal-like state, may be somewhat lesser than humanity as a whole likes to imagine.  Who knows -- it's not my field, and it's impossible to say for sure.  But when evaluating an AI, it does raise some interesting questions.

From a Turing-test perspective, the AI in AIW are undoubtedly more intelligent than any animal on the planet except for humans.

The Chinese Room problem (http://en.wikipedia.org/wiki/Chinese_room) cuts to the heart of perceived intelligence, and will be very thought provoking for anyone who has not been exposed to a variant of it before.

The other ideas, about having ships individually carry a bit of memory, is an interesting one, sort of an extension of the hotspots idea making it per-agent instead of geographic in the same sense.  I'm still not sold on the idea for AI War, but it's definitely interesting to at least ponder the many different possibilities that abound.

The reason I mentioned heat maps is that they are a form of memory that is very easy to tease emergent effects out of.  It is "loosely coupled" memory, in the sense that it can be "consumed" by users in a myriad of ways that the memory doesn't need to know anything about.  When you're dealing with complex adaptive systems, loose coupling is extremely important.

The reason I mentioned RPS earlier is that the first few campaigns I tried in AIW (level 7) ended in very quick (< 1 hour) stalemates.  I think AIW may be closer to RPS than it may look at first blush, largely due to the lack of "memory", not just on the part of the AI but also insofar as there are no high-marginal-cost actions that any player can take.  E.g., if a player could invest millions of resources into a system to "upgrade" it in some non-mobile manner, the game would be quite different I think.

[x4000]

But, I suppose it is easier to conceive of the AI War in terms of instinct rather than cogent thought, at any rate -- it is not self aware.  

I don't know man. Like I mentioned above with my cross planet attack example, maybe your AI has gone beyond you and is self aware now.  Did you accidentally give the AI a way to modify its own code? 

That is the beautiful, beautiful thing about AI that is highly emergent -- even people who know they shouldn't tend to anthropomorphize it.  I tend to do that, for that matter.  It isn't self-modifying in any way, and it doesn't have what I would call higher function intelligence... and yet, on the other side of things, I think humanity as a whole has a really poor grasp on what constitutes higher-order intelligence if it doesn't look exactly like ours.  And for that matter, we have trouble evaluating even our own; IQ tests are a joke, which I say despite the fact that they paint me in a flattering light.  From that angle, the AI in lots of games could be considered somewhat... I don't know, I don't even have a word for it.

[x4000]

It's made me watch the behaviors of pets and other animals, for example, and realize that they may be closer to us in intelligence than we like to give them credit for.  The degrees of separation between language, toolmaking, and all the sophisticated stuff, and simply existing in an animal-like state, may be somewhat lesser than humanity as a whole likes to imagine.  Who knows -- it's not my field, and it's impossible to say for sure.  But when evaluating an AI, it does raise some interesting questions.

From a Turing-test perspective, the AI in AIW are undoubtedly more intelligent than any animal on the planet except for humans.

The Chinese Room problem (http://en.wikipedia.org/wiki/Chinese_room) cuts to the heart of perceived intelligence, and will be very thought provoking for anyone who has not been exposed to a variant of it before.

Very interesting, I had not seen that thought experiment before.  From that standpoint, AI War would I suppose be a "strong AI" that fails the intentionality test in that experiment.  Although, if you go back to asking questions about what defines intentionality, then you get very muddy and circular arguments.

When you're dealing with complex adaptive systems, loose coupling is extremely important.

For sure -- that's how I approach software in general, and that's what later carried over into the AI.

The reason I mentioned RPS earlier is that the first few campaigns I tried in AIW (level 7) ended in very quick (< 1 hour) stalemates.  I think AIW may be closer to RPS than it may look at first blush, largely due to the lack of "memory", not just on the part of the AI but also insofar as there are no high-marginal-cost actions that any player can take.  E.g., if a player could invest millions of resources into a system to "upgrade" it in some non-mobile manner, the game would be quite different I think.

Interesting that you'd reached stalemate so very quickly, that is a big surprise; generally it takes far longer for that to happen, I suspect that you were not at a true stalemate.  If you're that early in, then the likely outcome is for you to try to take some new territory and then lose the game based on the AI catching you off guard while you do so.  Of course, if you're unwilling to risk an offensive in that sort of position, then I suppose that could be considered a stalemate.  But I guess my point is that in a true stalemate, there is literally nothing you can do that has any hope of bettering your situation, whereas so early in an AI War game there ought to be lots of low-level planets you can pick off with ease... the possible downside being your own death if your defenses are not adaptable enough.

Incidentally, the more planets you control, the more intelligent the AI tends to seem.  Increasing the decision space, and all that.

[intently]

Very interesting, I had not seen that thought experiment before.  From that standpoint, AI War would I suppose be a "strong AI" that fails the intentionality test in that experiment.  Although, if you go back to asking questions about what defines intentionality, then you get very muddy and circular arguments.

Well, "strong AI" is really more of a philosophical question than a technical one.  You may "believe" in it and choose to apply that believe to AIW, but it's not really something you can prove.

Interesting that you'd reached stalemate so very quickly, that is a big surprise; generally it takes far longer for that to happen, I suspect that you were not at a true stalemate.  If you're that early in, then the likely outcome is for you to try to take some new territory and then lose the game based on the AI catching you off guard while you do so.  Of course, if you're unwilling to risk an offensive in that sort of position, then I suppose that could be considered a stalemate.  But I guess my point is that in a true stalemate, there is literally nothing you can do that has any hope of bettering your situation, whereas so early in an AI War game there ought to be lots of low-level planets you can pick off with ease... the possible downside being your own death if your defenses are not adaptable enough.

Incidentally, the more planets you control, the more intelligent the AI tends to seem.  Increasing the decision space, and all that.

It's very possible I just didn't know / don't know what I'm doing

At a tactical level, AIW is very RPS-ish.  Fighter>Bomber>Frigate>Fighter (I think).  The fact that the AI apparently doesn't pick a force mixture based on the player's current force mixture means that you've basically decided on a random strategy for the AI force mix -- which can force the player into a tactical tie, just like a random strategy in RPS can.  (Unless the AI picks a non-random force mixture that isn't responsive to the player force mixture, in which case the AI is doomed to a tactical loss position over time.)  This is probably a good decision, since a human might be better at adapting to a co-adapting enemy force mixture than the AI would be.  On the other hand, you may want to look at some of the algorithms of the winners of the International RoShamBo Programming Competition (http://webdocs.cs.ualberta.ca/~darse/rsbpc.html).  I think I remember reading that these programs could regularly beat humans at RPS tournaments.

[x4000]

Well, "strong AI" is really more of a philosophical question than a technical one.  You may "believe" in it and choose to apply that believe to AIW, but it's not really something you can prove.

Certainly.

It's very possible I just didn't know / don't know what I'm doing

Sure, no worries.

At a tactical level, AIW is very RPS-ish.  Fighter>Bomber>Frigate>Fighter (I think).  The fact that the AI apparently doesn't pick a force mixture based on the player's current force mixture means that you've basically decided on a random strategy for the AI force mix -- which can force the player into a tactical tie, just like a random strategy in RPS can.  (Unless the AI picks a non-random force mixture that isn't responsive to the player force mixture, in which case the AI is doomed to a tactical loss position over time.)  This is probably a good decision, since a human might be better at adapting to a co-adapting enemy force mixture than the AI would be.  On the other hand, you may want to look at some of the algorithms of the winners of the International RoShamBo Programming Competition (http://webdocs.cs.ualberta.ca/~darse/rsbpc.html).  I think I remember reading that these programs could regularly beat humans at RPS tournaments.

Yes, there are a huge amount of RPS relationships, and the AI does not choose its ship mixes.  However, it's vastly more complex than even Rock-Paper-Scissors-Lizard-Spock, in that there are dozens of interrelationships.  The AI generally tries to keep something of a balanced force on its own planets, and then the human players have the "tempo" (in the Chess sense) to treat this as a puzzle of sorts to solve.  If the AI was always adjusting its ship mixes versus your ship mix adjustments, I think that would lead to the only valid strategy for human players being to keep a super-broad array of ships in every mix.  As it is that is sort of the case already, but that is mitigated somewhat by the AI being more of a specialist at each wave and guard post, and thus the player having some opportunity to specialize in response to those.

I guess the main thing is that the AI is acting here both as a scenario generator and as an opponent, and ship mixes are completely about scenario instead of trying to beat you.  For every reaction there is an equal and opposite one, and if the AI were more pointed and sneaky with its ship mixes, then the human players would not get to be.  As it stands, the AI is intentionally not, and so the players get to be a bit.  I guess my main goal in general with the AI in AI War is not to win and kill the humans immediately, but rather to provide the widest variety of interesting and challenging situations possible.  To that end, certain sorts of superfast analysis I keep out of the hands of the AI. On the other side, in tactical engagements where the ships are already set and known, I let it work in an unbridled fashion (assuming a high enough difficulty level).  So that gives really a hybrid mix, where the AI is trying to beat you and where it is just trying to set up an interesting scenario; in many ways, the AI Director in Left 4 Dead does a lot of the same sort of thing -- in its case, it is more about mood and tempo and atmosphere than just trying to kill you (when it populates zombies), but the zombies themselves act out of a sole desire to kill you as fast as possible.  Sort of the same idea in AI War, although I hadn't played L4D until well after AI War was released (I had read articles, though, and they were something of an inspiration).

[RCIX]

Y'know, that "neuronal" idea that Trurl came up with spawned a new minor faction idea! *scurries off to suggestions forum*