Stupid AI vs. disengagement penalties in combat

[Mr. Magniloquent]

Many moons ago I was given a rather common assignment of scripting an AI that would be able to path through a "field of pit-falls". The object wasn't merely to have the AI memorize where it had fallen in, but begin to proactively predict where pits may exist and navigate around any potential harm. Everyone is given a procedurally generated "map", and then the class saw who's program could get the furthest in the fewest attempts.

 

Challenging, indeed.

 

I imagine that scripting an AI to avoid engagement/disengagement penalties would be very similar, but slightly easier. Knowing explicitly where the hazards are and what each hazard is capable of will make evaluating the least harmful outcome to the stated goal much easier. This is where the emergent AI that Endroz was mentioning would come into play. With today's computing power, it should be rather feasible. Execution is always more difficult than theory though. Having the AI NPCs perform these actions in such a manner that would not be easily exploited would be quite a task.

[PrimeJunta]

Pack animals coordinate extremely well. A spec-ops platoon has nothing on a pack of African wild dogs. I'd be bummed if they nerf the combat AI for animals just because they're animals.

[Lephys]

Pack animals coordinate extremely well. A spec-ops platoon has nothing on a pack of African wild dogs. I'd be bummed if they nerf the combat AI for animals just because they're animals.

Hear hear! Just 'cause they don't think like people doesn't mean they're imbeciles (although some are, just like people. 8P)

[JFSOCC]

good AI isn't intelligent, it is predictable.

 

And interesting challenges can be made by combining enemy types in different compositions, increasing combat complexity.

 

So at first you find an enemy which debuffs your speed and stays at range, the next time you find that the same enemy type is combined with an enemy which does AOE damage. You now have a more complex challenge as you're not just trying to close range with the speed debuffer but also need to stay separated enough while crawling to the ranged guy to not die of AOE damage. Next up, the encounter is indoors, and the way through is blocked by a fighter, adding yet another level of complexity.

 

You still know that the first enemy type always does ranged damage and debuffs speed, but your encounters have evolved to become more interesting tactical puzzles nonetheless. You don't need intelligent AI for that, you need predictable AI and intelligent placement by the devs.

[Hormalakh]

Predictable AI has one huge flaw: once you learn its tactics, it no longer poses a challenge. Without any sort of randomization, it becomes a puzzle that repeats itself. Dev placement only adjusts it slightly. This is why you have "cheese tactics" that occur with certain AIs. If you know an AI will always follow you and trickle in one at a time you always stand behind a door and shoot them down one at a time.

 

I disagree with the notion that "good" AI is predictable. Maybe "non-frustrating" AI is predictable. But definitely not "good" AI.

[JFSOCC]

for gameplay purposes, yes, predictable is better. You can manage encounters by having a varied enough roster. And I don't think the second time you get the same monster it will pose no more challenge, the idea is that you have a bestiary large and varied enough that having the same encounter more than three times is unlikely.

 

Bosses can be intelligent, especially with the predictable guys backing him. I think that gives enough space for variation to never have a dull encounter.

[Lephys]

for gameplay purposes, yes, predictable is better. You can manage encounters by having a varied enough roster. And I don't think the second time you get the same monster it will pose no more challenge, the idea is that you have a bestiary large and varied enough that having the same encounter more than three times is unlikely.

 

Bosses can be intelligent, especially with the predictable guys backing him. I think that gives enough space for variation to never have a dull encounter.

*shrug*... I'm with Hormalakh. Ideally, you don't need to fight a different creature (no matter if the bestiary contains 30 or 3,000 for you to choose from) just to experience an unpredictable challenge.

 

A good AI is mostly predictable. It always predictably "chooses" quality tactics/actions, given the circumstances, but doesn't always arrive at the exact same decisions. It's not unpredictable in that it's just going to do completely random stuff, or you're not going to know what abilities it has in its repertoire. It's just often going to have multiple "good" options to go with, in terms of "what action should I take right now?", and you're not going to know exactly what's going to occur, so you still hav e to react.

 

Think of an attack resolution system that has no chance involved, versus one in which you've introduced the possibility of critical hits. In the former, you know that, if the enemy does 10 damage, and he attacks, he's going to deal 10 damage. So you can always calculate the best course of action. However, if he CAN do 17 damage, but you don't know WHEN he's going to do 17 damage, you now have a SIGNIFICANTLY less predictable outcome. You can no longer say "hey, this guy has 13 HP left, and there's only one hit to go to kill this beasty, so I can DEFINITELY send him in to kill it and the beast's 10 damage won't kill me." But, you also can't say "I HAVE to do something else, because he's going to do 17 damage." You don't know if he's going to do 17 damage, or 10.

 

Now, apply that to behavior. You know roughly what the thing's capable of, and what it's sort of going to do, but you can't just foretell it's every action and/or the specific outcomes of those actions.