I've noticed that game AI can tend to look smarter than it actually is. My AIs, while not being particularly complicated at the time being, will never the less give a very good impression of pushing the player into a corner they didn't really intend to push them into. I've heard that FEAR's AI didn't actually have the ability to flank the player built in. Instead, they just found cover... thanks to the map designers, there just happened to be cover behind the player. My question is: how much of creating intelligent looking AIs is smoke and mirrors? How can you give the appearance of complex behaviours with a relatively simple AI?

3 states and a plan: the AI of FEAR - there's Jeff Orkins GDC presentation that goes into a fair bit of detail about how their squad tactics are mostly smoke and mirrors.

From looking at the HL1 and HL2 SDK code, especially the HL1 grunt squad stuff it would seem that a lot of that relies on smoke and mirrors to a degree, though I can't give any specific examples right now (not near a machine with the source code on)

definatley read that "3 states and a plan" paper, good insights in it!

The HL1 AI (you can download the HL SDK, including the AI for free), had lots of hacks in it, for example, when a grunt throws a grendade, they pick a random number and half the time the grunt will aim for the last position they saw the player, and the other half of the time they will aim for your exact location, even if they dont know where you are!

Hey, Ging! I'd lost the link to that way back... all I was working from was a brief overview of the presentation. This is some damned interesting stuff. Thanks!

Just to throw an idea out there: what if there was an AI attached to the player that attempted to guess what they were up to (ie, attacking this player or blind firing) that other AIs could access? It wouldn't be quite the same as the AI actually figuring out what the player is doing from THEIR point of view, but it could be a less error prone way of going about it.

A better question is, how much of the intelligence of real people is just smoke and mirrors, or, as we are more apt to call it in the 'real world', dumb luck? I've seen people pull of some incredible stuff in multiplayer squad-based games, and invariably I ask them how they did it and the answer is always 'I dunno, we just kinda were all in exactly the right spot'.

However, in some multiplayer FPSs the teamwork IS intentional, and its not just a matter of "right place, right time". I've noticed this happens more if the game is slower paced. Games like DoD, Battlefield, or ESPECIALLY unreal tournament almost never see any kind of intentional strategy... if something awesome happens, it really IS a matter of luck. In that case the game developer would have to find ways of hinting people towards cool situations through maps or gameplay features, since they aren't going to do it themselves.

However, a slower paced game like the Half Life mod "Hostile Intent" will see tons of coordinated strategies. The mod in question results in a lot of communication over voice chat for confirmation of position, timing attacks, etc. The slower pace seems to give people the chance to actually think.

On a somewhat unrelated note, to what extent can you make AI look smarter through having it communicate it's intentions? The presentation Ging linked to mentions how much smarter the FEAR soldiers seemed because of the radio chatter. It made them appear like they were giving orders or requesting information, when they actually weren't doing anything of the sort.

Similarily, Half Life 2: Episode One had a lot of special interactions between characters. For example, Alex would kick and grapple with the zombies. If you listen to the directors notes, they point out how this wasn't actually a matter of getting Alex into the proper position to do that. Instead, they just checked if a zombie was in the proper position doing the proper action, and if those conditions were satisfied, they'd trigger the sequence.

Quote: Original post by starstriker1
On a somewhat unrelated note, to what extent can you make AI look smarter through having it communicate it's intentions? The presentation Ging linked to mentions how much smarter the FEAR soldiers seemed because of the radio chatter. It made them appear like they were giving orders or requesting information, when they actually weren't doing anything of the sort.

It isn't fully true that they weren't doing anything like that. The truth is that the squad ai was giving orders and requesting information, but the orders were directed towards selected units. This squad ai is who speaks through the soldiers. In some way, we can say that there is a certain kind of combined intelligenece in the case of real humans too when observed as group instead of seeing them as separate human entities. (The same applies to ant colonies where group behaviour is both proven and accepted.)

Viktor