Are Game AI's getting worse? Options

[Warlordtheft]

Is it me or am I just getting older? Most newer games just seem like button mashers rather than requiring any real thought or planning to play. I would think AI's would have gotten better over the years but have been sorely unimpressed with the AI's of the recent total war series, Hearts of Iron III, and others. Or is it just that I'm so used to try an outhink human opponents (from my many years of wargaming) that any AI wouldn't challenge me.

What have your all's impression of the challenge of AI's in newer games?

[Tanegar]

I think it's an inevitable result of increasing sophistication in games. It's easy to write a program that makes smart decisions within simple parameters, but when the rules and variables get more complicated, it becomes much harder for an AI to evaluate them all and make good decisions.

[CanRay]

Explains the proliferation of Multiplayer as well... And why designers are more interested in that than first-person storylines and challenges.

[capt.pantsless]

I think it's an inevitable result of increasing sophistication in games. It's easy to write a program that makes smart decisions within simple parameters, but when the rules and variables get more complicated, it becomes much harder for an AI to evaluate them all and make good decisions.

Or better put, it's harder for wimpy human programmers to figure-out a good AI algorithm that evaluates all the parameters intelligently.

Often times a huge number of imputs can be a major advantage for a well-thought-through AI. E.g. if there's a complicated mathmatical model to finding the optimal action, the AI can find that optimal path in a couple thousand clock-cycles.

One of the bigger factors is game-design studio's not dedicating enough programmer-hours to the AI. Since it's not a feature you can take a screenshot of to put on the box, it's not going to get as much attention from the marketing dept.

[Wounded Ronin]

Shogun: Total War 2 has a better AI than earlier games in the franchise.

[Bigity]

Shogun: Total War 2 has a better AI than earlier games in the franchise.

Now if only my PC didn't cry in terror when I even think about trying to run that.

[Warlordtheft]

Or better put, it's harder for wimpy human programmers to figure-out a good AI algorithm that evaluates all the parameters intelligently.

Often times a huge number of imputs can be a major advantage for a well-thought-through AI. E.g. if there's a complicated mathmatical model to finding the optimal action, the AI can find that optimal path in a couple thousand clock-cycles.

One of the bigger factors is game-design studio's not dedicating enough programmer-hours to the AI. Since it's not a feature you can take a screenshot of to put on the box, it's not going to get as much attention from the marketing dept.

Might be why computers are good at chess....it is fairly predictable. But yes there is the factor of more complicated strategy games (I'd put HOI3 in that boat) where the AI can't make a decision because of the number of inputs and options that are available. However, processors have gotten powerful enough that a more robust AI should be possible.

Also, begs the question if a scripted AI would be better in some respects than a fully autonomous one would be.

PS:I've not gotten Shogun Total WarII yet, but a friend of mine has said it suffers from the same Battlefield AI idiocy that made Napoleon and Empire total war less than challenging.

[CanRay]

I knew I should have put my money into Artificial Stupidity Futures in the '80s...

[Blade]

In the past, a lot of games used to cheat: FPS bots knew exactly where you were and in a lot of strategy/tactics game, the AI just started with more resources than the human. Nowadays it's less common.
But even back then, I remember disliking RTS because the AI were just so simple to beat that they turned into auto-satisfaction games where all you had to do was spend enough time to see your unstoppable army destroy your opponent.

In some cases, playability is an issue. For some games it's easy to do an optimal AI that nobody will be able to defeat, but doing so would make the game less interesting. So you have to program an AI that will be challenging (and preferably human-like) yet not too good. And since the overall difficulty level of video game has been steadily going down for a long time (with some exceptions), AI might look like they're getting worse.

[Bigity]

You're on to something here, IMO.

Developers can't spend months making a kick-ass AI for a FPS game, because the Xbox kids won't be able to 'win' by mashing buttons.


I mean, look what the consoles did to the Rainbow Six series.

These days games have to cater to 30+ year olds that remember when video games were brutal and utterly unforgiving and 10 year olds on an Xbox shouting profanities into the mic 24/7 and thinking they are cool for doing so.

[Kagetenshi]

Part of the problem is that developers have decided to focus on AI too much. Games like DOOM were challenging but not unfairly so despite having very small amounts of monster logic—primarily due to player-monster asymmetry (monsters not nimble, mostly non-hitscan weapons, encountered in massive numbers). This wasn't guaranteed to maintain an enjoyable experience (Rise of the Triad's hardest mode was simply unfair), but Unreal started a nasty trend of having opposition resemble humans more, at which point you really lose the middle ground of "difficult but able to be overcome with effort/study".

Plus, AI problems tend to be hard. One of my favourite game AI stories is about how the AI in F.E.A.R. was made to appear much smarter than it actually was—the AI would appear to flank the player, but would actually just switch between "attack" and "take cover" modes, with the flanking achieved by designing the combat areas such that iterating the "take cover" mode would naturally lead to a flanking-style movement. Similarly, the AI would yell for reinforcements while being massacred—these calls were noops, but players would tend to assume that the next batch of enemies they encountered were responding to the call.

So I suspect the problem is that game AIs are getting better.

~J

[KarmaInferno]

The voice clips in FEAR 3 are a little annoying.

It's kinda nifty to hear the AI opponents going "Holy shit, did you see that?" and "That's impossible!" early on in response to your uber-combat ability.

Hearing it for the ten-thousandth time, well, I shoot them faster just to get them to shut up.




-k

[Blade]

The voice clips in FEAR 3 are a little annoying.

It's kinda nifty to hear the AI opponents going "Holy shit, did you see that?" and "That's impossible!" early on in response to your uber-combat ability.

Hearing it for the ten-thousandth time, well, I shoot them faster just to get them to shut up.

It was fun in Deus Ex to hear security guards in the area 51 say, a minute after an explosion "Must have been a cat" or "Probably a bum".

[CanRay]

Yeah, they really need to aim the voice systems a lot better. More voice acting!!!

[KarmaInferno]

I am reminded of this.





-k

[Wounded Ronin]

Might be why computers are good at chess....it is fairly predictable. But yes there is the factor of more complicated strategy games (I'd put HOI3 in that boat) where the AI can't make a decision because of the number of inputs and options that are available. However, processors have gotten powerful enough that a more robust AI should be possible.

Also, begs the question if a scripted AI would be better in some respects than a fully autonomous one would be.

PS:I've not gotten Shogun Total WarII yet, but a friend of mine has said it suffers from the same Battlefield AI idiocy that made Napoleon and Empire total war less than challenging.

I mean it's not like a little Subotai is living in your processor or anything, but it's a VAST improvement over Medieval II.

[Fyndhal]

I mean it's not like a little Subotai is living in your processor or anything, but it's a VAST improvement over Medieval II.

Just give it some time. Things will get better. (IMG:style_emoticons/default/biggrin.gif)

[Tanegar]

So, we're making AIs by teaching them to play a game where world domination by military supremacy is a valid goal? What could possibly go wrong? (IMG:style_emoticons/default/cyber.gif)

[CanRay]

NO! Not TVTropes! Don't teach military AIs about TVTropes! THEY'LL RULE THE WORLD!!!

Rage is getting some interesting reviews on their AI. Going to have to check that out.

[Kagetenshi]

So, we're making AIs by teaching them to play a game where world domination by military supremacy is a valid goal? What could possibly go wrong? (IMG:style_emoticons/default/cyber.gif)

On that topic…

~J

[Socinus]

Part of the problem is it's far easier to make an AI that will cheat more as the difficulty goes up rather than one that has better strategy.

The basic AI tactic, the bull rush, is effective if the computer can build units twice as fast as you and can harvest more resources than you.

That becomes more problematic in games where there is specific mechanics in place to beat out a rush or you can develop defenses against a rush quickly. At that point, massed units are basically only really good for boosting the player's score. If the AI has formidable defenses, either as emplacements or hordes of units, you can just wait until the computer makes a bad move or until their resources run out or you can get a force in position to the damage needed to knock them out of the game.

I'm not a programmer, but I wonder how difficult it would be to put a major push into a program that could analyze what the player did to win a game, then employ that strategy or defenses against it and simply build as more games were played.

As I said, most game designers take the short-cut of having the AI cheat more as the difficulty goes up. Which does present it's own tactical challenges, but bludgeoning someone to death with mobs of units only works so long or is only fun for so long when you have this great tactical framework that isnt utilized.

[Warlordtheft]

As I said, most game designers take the short-cut of having the AI cheat more as the difficulty goes up. Which does present it's own tactical challenges, but bludgeoning someone to death with mobs of units only works so long or is only fun for so long when you have this great tactical framework that isnt utilized.

This is one of my basic complaints about difficulty levels in many strategy games. The higher difficulties only mean that the AI's units are given bonuses, they get more strategic resources, and might be given a better strategic position. But that still won't prevent the AI from continuing to do stupid stuff like the mongol fort trap in Medieval Total War II

[Wounded Ronin]

Part of the problem is it's far easier to make an AI that will cheat more as the difficulty goes up rather than one that has better strategy.

The basic AI tactic, the bull rush, is effective if the computer can build units twice as fast as you and can harvest more resources than you.

That becomes more problematic in games where there is specific mechanics in place to beat out a rush or you can develop defenses against a rush quickly. At that point, massed units are basically only really good for boosting the player's score. If the AI has formidable defenses, either as emplacements or hordes of units, you can just wait until the computer makes a bad move or until their resources run out or you can get a force in position to the damage needed to knock them out of the game.

I'm not a programmer, but I wonder how difficult it would be to put a major push into a program that could analyze what the player did to win a game, then employ that strategy or defenses against it and simply build as more games were played.

As I said, most game designers take the short-cut of having the AI cheat more as the difficulty goes up. Which does present it's own tactical challenges, but bludgeoning someone to death with mobs of units only works so long or is only fun for so long when you have this great tactical framework that isnt utilized.

That is why the next great game should be Redcoats vs. Shaka Zulu. Just have the AI use charging bull formation every time.

Better yet make it a squad level FPS. Think of how hardcore it would be to work the action on your rifle in a realistic manner while a hundred guys with spears suicide rush you.

Use dated graphics so the computer can keep track of hundreds of bodies piling up on the ground and forming obstacles.

[Warlordtheft]

Better yet make it a squad level FPS. Think of how hardcore it would be to work the action on your rifle in a realistic manner while a hundred guys with spears suicide rush you.

Back in Medieval toal War (the original), there was one spot that the Mongols invaded through. I was the Byzantines, and had stacked my armies there in preparation for the MOAB. THe battle timer was off for this game. It took about 3 days realtime to resolve that battle and the corpses (especially at my defensive line) was about 6-7 deep. It was one of the most epic battles I've ever had on a PC.

[Blackb1rd]

We are at an interesting point in the technological world where the potential for huge advancements are present but we do not have the resources, abilities, or skills to make those advancements. Give it a few years and the AI in games will blow you out of the water, however right now AI in games is far from stupid or predictable. Alot of it depends on the way YOU play the game because any decisions the AI makes are based off of what you do.

I'd assume this thread is speaking manly of AI in strategy games? I ask because AI in RPG's has advanced significantly over the past couple years to the point where there is the potential for some very realistic behavior coming from in game characters.