Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

Immersion.

Challenge.

Complexity.

Ability to run on any system.

Undecided.

All are equally important.

Other.

Don't care.

View poll.

[Leafsnail]

Cheating AI is kindof weird.  I mean, if you take the Black Ops Veteran AI, they can hit from miles away with the most innaccurate guns and no sights of any kind, and can wheel around 180 degrees to shoot you as if they had their sensitivity on full, they're still fundamentally very stupid.  Me and my friend managed to camp in a room for an entire game against 10 hostile computers, simply because they never tried to grenade us out or make a coordinated attack.

So... in many cases, it's probably the case that AI needs to cheat in order to be a real challenge.

[Toady Two]

EU3 AI might not be the best example. Ideally we would want the AI there to behave like the country it is representing as opposed to trying most efficiently to conquer the world with every minor like a human player would.

As I watch the Starcraft videos I begin to wonder how much of the Overmind AI success in the tournament comes from the fact that it uses exclusively Mutalisks in a manner that is confusing for a person and probably debilitating for another computer.

Really good AI shouldn't be a one trick pony and should instead diversify it's main strategy according to the situation.

[de5me7]

I think what an AI should do depends a LOT on the game in question.

A Strategy AI should go for challenge, complexity, and adaptability.

A Combat AI should be focused more on being interesting and providing an acceptable level of challenge.

i agree with this in that there are two purposes of AIs imo

purpose 1 is offline target practice for playing humans online.

purpose 2 is simulating a human because you cannot play a human in said game

so in unreal tournament, i dont expect alot of variations from the bots apart from weapon pref, and possibly an aggression setting. They exist as target practice and training for online play. On very hard modes, i dont even mind that much if they cheat.


in a total war game, i want to feel like im locking heads with great military minds, so the purpose of the AI is to simulate that. If the AI; cheats, performs impossible feats, does things that a human would never do, evades the traps a human would fall into etc, then its failing.

I think some AIs should actually simulate human flaws. The total war AI is a pretty good example of a mediocre ai, especially on the battlefield but also on campaign map. It always knows where hidden troops are when a human wouldnt, it almost always uses the same strategy, its rarely indecisive (expect in Medival 2 where there was an infinite indecision bug) etc. Historically different generals had differnt approaches, and became known for their ways of doing things, AIs should seek variety to make you feel like you fighting different oponents.

[Draco18s]

As I watch the Starcraft videos I begin to wonder how much of the Overmind AI success in the tournament comes from the fact that it uses exclusively Mutalisks in a manner that is confusing for a person and probably debilitating for another computer.

Really good AI shouldn't be a one trick pony and should instead diversify it's main strategy according to the situation.

Oh certainly.  The Overmind is brutal with the few units it does know how to use, and commands them well, mainly because each one is acting as its own independent entity (with slight oversight from the core AI to coordinate attacks on a single unit, etc).

And yes, AIs shouldn't be one trick ponies.  The problem is, AIs are very good at their single pony tricks.  By trying to spread out the AI gets weaker (either because its being too generalized) or too strong (because it's able to alter strategies in microseconds, switching from one trick pony to another faster than a human can react).

[GaelicVigil]

I suspect that you're only going to get "immersion" as the top number in a poll done on a site like Bay12.  Go ask the same question on a TBS or RTS forum and you'd get Challenge as easily the highest number.

This is what pisses me off to no end - when people confuse a good AI with a challenging AI.  Galactic Civ 2 had this problem.  With a difficulty setting at anything higher "normal" and above, the AI was absolutely, mind-bogglingly, difficult to beat.

The sucky thing is, is that game developers seem to fall into the trap of basing the quality of their AI on the feedback they get from the most hardcore players (read: forums).  Brad Wardell loves to listen to what people are saying on his forums, which, of course are the most avid fans.  Unfortunately, for the more casual player who spends only a few hours a month playing, we get screwed by an impossible AI.

[Draco18s]

This is what pisses me off to no end - when people confuse a good AI with a challenging AI.

This is also true.  While I want my AIs to be challenging (i.e. goal) I want them to be challenging because they're good and make smart decisions.

[hemmingjay]

I voted "other" because i believe the focus should be on adaptive ai. regardless of how challenging the ai needs to be(set by difficulty level in options or by level design in code) adaptive and responsive to the player's technique. If this exists then all other categories will be fulfilled.

feel free to flame me for my simple answer, but I believe it is correct

[Akhier the Dragon hearted]

    First of all calling what is used in games an AI is incorrectly describing it but only because we do not have the appropriate word to do so. The current best AI we have is below animal level thinking while saying AI conjures up thoughts of human level intelligence. Now that that's off of my chest let me answer the question asked in the original post.
    The Goal should be fun on the player side. It does not matter if it thinks in a way humans would or makes decisions like us. What matters is that it appears to do so while being fun to play against. Honestly if a person spent enough time watching how people play a game against other humans then makes up a bunch of "if then" statements you could theoretically make the computer play like that person. The problem is this takes a lot of time and fine tuning so most games are released before they would have been able to do so. A good set of articles on game AI can be found at Ascii Dreams where the maker of Unangband talks about the Angbands AI and how he uses it in his game to make monsters fun to fight while appearing intelligent.

[Gantolandon]

The main problem with developing advanced AIs is that it very rarely pays off. It's much easier (and cheaper) to hardcode most of the choices you can conceive and throw some random variables than design a complicated algorithm which could make those particular choices and more. Even when made, this AI cannot surprise its makers too much - otherwise how would they even test it, let alone tweak it? And, of course, it should be transparent enough to the player. Otherwise he won't even understand what's going on under the hood and become frustrated very quickly.

There are more ways to arouse the player's interest and most of them are much cheaper, with a higher chance to actually succeed. That's why most of the game developers never go beyond finite state machines.

[Akhier the Dragon hearted]

    That is not a bad thing in and of itself really. The only thing about it is that its not AI. A game just needs the enemy's to be fun and seemingly smart. An example of this is in the Ascii Dreams articles:

One of the most effective AI actions in F.E.A.R. is when the player has killed all but one of their enemies and they hear the radio message 'Man down, I need support. Calling for reinforcements.' broadcast by the last remaining survivor.

The equivalent AI action is: do nothing at all. No additional backup arrives. The designers didn't need to worry about developing this code. The player perception is that the backup is coming, so they have an incentive to hunt down the last survivor. And the designers know that there are more soldiers around the next corner, which the player will perceive as the reinforcements arriving a little to late. It may sound like a cheap trick, but it is a huge saving in development cycles and CPU time.

Its a simple thing to code and it adds a depth that otherwise would take a lot of coding to make work.

[Gantolandon]

That is not a bad thing in and of itself really. The only thing about it is that its not AI.

It is - a bit. More advanced AI helps the game to be more fun when played several times.  A good AI could create more fun and interesting situations than its developers could think of. So it's a goal worth pursuing - unfortunately, not really attainable by now.

[Draco18s]

    That is not a bad thing in and of itself really. The only thing about it is that its not AI. A game just needs the enemy's to be fun and seemingly smart. An example of this is in the Ascii Dreams articles:

One of the most effective AI actions in F.E.A.R. is when the player has killed all but one of their enemies and they hear the radio message 'Man down, I need support. Calling for reinforcements.' broadcast by the last remaining survivor.

The equivalent AI action is: do nothing at all. No additional backup arrives. The designers didn't need to worry about developing this code. The player perception is that the backup is coming, so they have an incentive to hunt down the last survivor. And the designers know that there are more soldiers around the next corner, which the player will perceive as the reinforcements arriving a little to late. It may sound like a cheap trick, but it is a huge saving in development cycles and CPU time.

Its a simple thing to code and it adds a depth that otherwise would take a lot of coding to make work.

That's an example of an AI that is doing the right thing.  There's no reason to "add more units" because they're already there.  The idea is that the player thinks something happened.

[Akhier the Dragon hearted]

No that is an example of a "if then". If you are the last in you squad Then shout about back up. There is no intelligence in it. You can easily add it to any game with random spawns and players will think the next enemy's they meet are the back up. True Intelligence would do so because it actually had an effect on game state. In the end though the difference is only in the code because whether the unit is actually requesting back up or just playing a prerecorded message will not be visible to the player.

[Draco18s]

No that is an example of a "if then". If you are the last in you squad Then shout about back up. There is no intelligence in it. You can easily add it to any game with random spawns and players will think the next enemy's they meet are the back up. True Intelligence would do so because it actually had an effect on game state. In the end though the difference is only in the code because whether the unit is actually requesting back up or just playing a prerecorded message will not be visible to the player.

Its a very simple AI that does what's best for player experience.
The AI is playing a different game than the player is, so it runs by different rules, and the rules it runs by make for a good player experience.

[Akhier the Dragon hearted]

AI means Artificial Intelligence. That is a finite state machine. For the thing to even come close to being considered an AI it would need to add detection of units it could call for back and a system to allow that backup to come to it. I would need 2 lines of code for the example which would roughly be:

If squadmates = 0
  Then say "Man down, I need support. Calling for reinforcements."

Its that simple.