How "aware" is the AI actually? A question about Tractor beams and Raid Engines

[thelonecarrot]

Can't see a thread discussing this from the search engine so I'm posting a new one.


From my experience so far the AI is a sneaky little shit. A clever girl. I give the AI alot of credit and expect it to think in quite an intelligent manner.

However the thought came to me when I thought back over an incident involving a raid engine being activated due to a tractor platform pulling some of my ships into a adjecent planet that did the AI know that?

Did it deliberatly choose that planet to pull my troops towards because it realised that the raid engine would be activated? Is the AI capable of such strategic choices?

And even if it didn't, even if it chose that world because of wormhole proximity wouldn't it make sense for it to want to activate the raid engine? It would be a great tactical advantage.

Ultimately the question becomes just how self aware is the AI of itself? Can it make decisions based its awareness of itself as a whole, rather than just as a splintered being.

[Diazo]

I don't think there's any sort of over-arching strategy, at least not for the default AI enemies. The hybrids may have (or end up with) something along those lines however.

In this specific case what I suspect happened is that the retreat threshold was hit as you killed the AI's forces and so the AI just ran away with your ships tractored and there also happened to be a raid engine in that direction.

That's one of the great things about the procedural generation AI War uses, it's different for everyone and you can get odd and interesting things like this.
D.

[Mánagarmr]

Actually, the tractor-kidnapping is emergent behaviour in tractor-using ships (Etherjets too for instance). Chris himself didn't code the AI to do that, it simply does, given the circumstances. Or rather, of course he did code it to do that, but not explicitly tell Etherjets/tractors to behave like that. The AI simply felt that works best.

So no, it's not aware it's a Raid engine that it needs to activate, but the kidnapping is "deliberate" behaviour, as far as I can remember from the discussion in the "This AI is awesome" thread.

[thelonecarrot]

Thanks for the info guys.

I wonder though, how much emergent AI is there in the game that goes unrecorded due to to how uncommon it may be to see it?

Tractor platforms legging it is an activity visible in one specific unit. But how do you then tell the behaviour of other units from coded, expected reactions and possibly emergent behaviour?

Kind of a moot point though since it doesn't change how you play the game.

[Martyn van Buren]

As I understand it, the AI really doesn't do much "thinking" at all --- each ship has a targeting procedure which is pretty much the same as your ships on FRD, there's a local "commander" who can order all the ships on a world to go after a particular target or retreat, and there's a overall AI that decides where to place reinforcements.  I could be wrong, but I think that tractor-kidnapping is actually one of the very few tactical maneuvers that was coded in intentionally; I think I remember X saying somewhere he discovered the tactic as a player and then taught it to the AI.

[Diazo]

As I understand it, the AI really doesn't do much "thinking" at all

Actually, this is a good way to look at it. There actually isn't much intelligence shown in the enemy AI, it's all reactionary behavior.

The waves and CPAs come at set intervals, everything else is in response to your actions.

It's actually one of the things I love about AI War, when you lose you look back and go "Oh, I lost because I did this." You don't go "I lost because the AI got a fleet around my left flank somehow."

Combined with the vast number of game options, it allows you to create the perfect game for yourself.

D.

[keith.lamothe]

The tractor-thief behavior was actually intentionally coded in, but it's just that little bit.  Which AI planet it runs off to with it is emergent.

In general there is no overall AI, just individual ships obeying some fairly simple rules in picking what they individually will do.

Hybrids are a different kettle of fish, but even there individual hybrids are making their own calls (often heavily weighted by "where are other hybrids planning a similar mission?") rather than the AI counting up its hybrids and saying "ok, you 10 defend that ARS, you 20 attack that human planet", etc.