« How To Buy Videogames and Consoles | Main | More on the PSP »

May 30, 2005



I think about that when playing stealth games - how much it would TOTALLY suck if they didn't operate by simple artificial rules. I mean, if they were halfway intelligent, or say 1/10 of the way intelligent, then as soon as one of the guards saw that another guard was missing, the whole place would be on alert, and it wouldn't stop in 30 seconds - they'd keep searching until you were DEAD.

One of my favorites is when you shoot a guy (not enough to kill him) then run and hide somewhere, and 30 seconds later he's back on patrol... "hmm, must've been a mosquito."

In general, I think AI is a situational thing. I don't want fully intelligent enemies in Space Invaders, and I don't want Space Invaders to disappear. Games like that are fun, lots and lots of fun. Just because I can play an intense space shooting game against smart enemies who will maneuver like crazy doesn't mean I no longer want to mow down pitiful minions. There's no Rambo Effect in a game with enemies as smart as you. Because in real life, Rambo wouldn't live long.


The AI comment is really conditional on the type of game. I feel a strategy game SHOULD have AI that makes you cry at it's hardest level. Sports games would also benefit from better AI. At the moment, most of these games rely on giving the CPU artificial bonuses or just blatantly screwing the player without giving them an opportunity to counter it (yeah, thats a shot at EA sports)

However, as Hamumu illustrated, a fast past game like an FPS or even an RTS may become too difficult if we let the AI become too good.

Sean Barrett

Hmm, that's a weird paraphrase. The manifesto says "give us A.I. that will actually outsmart us now and then", and I agree with your rant on this. But your original summary, "AI should surprise us", is just fine without your final rephrasing of it, and until I finally clicked through on the link I was confused.

If the AI has no surprises, there's no AI--you're just playing a puzzle game. (Oversimplifying for the sake of sloganeering.)


Speaking of Manifestos: http://gbgames.com/blog/index.php?p=55

Anyway, I believe that particular manifesto was getting comments anywhere and everwhere. It's an example of how game players would make bad game designers.

Video games have alread gotten past the "let's make the AI incredibly smart and difficult" phase. More challenge does not equate to more fun.

Still, a lot in the manifesto rings true. An AI that actually surprises you by doing things "creatively" is much better than one that is predictable, and the challenge is making it appear intelligent without making it overbearing and omniscient. All the processing power in the world won't help unless someone can make the balanced combination of algorithms. AI isn't as easy as graphics.

Nathan McKenzie

"An AI that actually surprises you by doing things "creatively" is much better than one that is predictable"

(Blue in face mode initiating)

Is it?

AI that surprises you can certainly help with immersion if you're simulating humans, I can cede that.

And AI that surprises you is really important if you're making a game that simulates a competitive, equal sides game like Unreal Tournament or Chess or Street Fighter or Starcraft or Civilization. I can see that too.

But would a Zelda game REALLY be _better_ if the enemies did unpredictable things, and the bosses stopped being random pattern based? Would Resident Evil 4 _really_ benefit from enemies that did the unexpected? Or Devil May Cry? Or Ico? Or Ikaruga? Or Doom 1?

Or Lode Runner or Pac Man or Epyx's Impossible Mission or Pooyan or Robotron? Several of these would cease to be meaningful games if you tried to make the AI unpredictable. Are these game models that people are incapable of experiencing pleasure from any more?

I like the fact that in Super Mario Brothers, green turtles walk off of ledges and red turtles don't. The predictability is a big part of what let's me, as a player, make interesting plans of attack when I see those enemies on the screen. It's a big part of how the designers are able to pace my experience and differentiate what sorts of challenges I'll face, thus keeping me interested. The more that AI makes enemies unpredictable and all able to do the same stuff, the more samey a game becomes over the long run, for me, if it's a linear obstacle-course thrill ride sort of game.

It might just be my particular aesthetics with certain kinds of games, but I'm still not at all sold on the topic of "more better AI == more better game". I think they're at times orthagonal, or worse... which curiously mirrors my "more better physics == more better game" stance, a subset of my popular "more better buzzword == more better game" theory that I've been bandying about.


How about "fun_battles = enemy_intelligence * number_of_enemies"?

It's not perfect, but it's another theory. So I'm with the writers of this manifesto that it's a great thing "now and then." Granted their Doom III example is a very bad one. And in Zelda that would take the attention off of what it's famous for, it's combo of light fighting and fun puzzles. It's just like what was earlier said; it can destroy the Rambo fantasy in many games. (Or maybe better, Hot Shots: Part Deux.) But that doesn't mean that it's a universally bad idea.

For instance, Shadow of the Colossus is going to have twelve monsters. If they're all as easy as Ganondorf (dodge, dodge, attack, dodge, hit weakpoint, repeat) I'm going to want my money back. And I'm fine with the badguys having a weakpoint. It's that I don't want battles to largely be finding a process that works and using it against all of the enemies. At least not in every game I play.

Imagine if there was a formula that takes account enemy intelligence/capabilities and the number of enemies, and tells you how fun a game is. Most games by far and large use the "dumb badguys, plenty of them" approach. There are bound to be other combinations that work.


AI today: Two guards are standing outside. The player sneaks up to them in the bushes, jumps out and slices up the first guard with a katana. The other guard says a lame one-liner like "You will die" or "I'm gonna get you for that" and proceeds to stand on the same place shooting at the player. The player then kills the other guard.

AI the way gamers want it: Two guards are standing outside, smoking cigarettes and talking about the football game that was on yesterday. The player sneaks up to them in the bushes, jumps out and slices up the first guard with a katana. The guard then reacts like a real guard would, for example by getting scared and running away, or maybe by dropping his weapon and pleading for his life, or in some other realistic way. The player would then react accordingly, maybe tie him up or deal with the situation as he wishes.

Why would anyone want to play a game where the opponents are nothing more than glorified gunnery targets? If the game becomes too challenging with too smart AI, then either give the player better equipment/superpowers/whatever, and give the enemies poor equipment/training/morale. The reason people play games instead of watching movies, is because they want to be in control of the events. I can't even count the times I've thought "I would've handled that differently." while watching a movie.

Jason Booth

I'm with nathan on this one: Here's what I posted in response to the sister thread of this one on Damions Blog:

You know, whenever realism in AI, or even realism in general come up I cringe. AI is there to provide a challenge; and that challenge does not necissarily imply realistic behavior in any way what so ever. For instance, have you ever heard anyone complain about the AI in Mario Brothers? Would Mario Brothers be improved if it had more “realistic” AI? Of course, the answer is no.

What is important is that the “laws” of the world are maintained. We cringe at the “coming straight at you” mentality of bad shooter AIs because we expect that the laws of a world based on our reality make that a bad tactic. However, given a 2d world as in Mario Brothers, and our expectations are not as such because we do not directly corelate those tactics into our reality.

Mario’s AI should be of particular interest to MMO developers, believe it or not. I know it sounds far fetched to look back to the 8-bit days for AI, and to use such a simplistic example, but let me ask you something: Did Mario’s AI provide you with a fun and rewarding challenge? Do the AIs of current MMPs provide you with a challenge of equal quality? Personally, I don’t think they do.

The reason is that Mario’s AI is believeable within the Mario world; you don’t question it. Additionally, those extremely simple behaviors and situations combine to form new challenges. In most MMPs, fighting an orc is always the same battle, and if you are significantly higher level than the orc, you will win. The experience does not change significantly, nor does it bare repeating if not for the grind imposed on you. Yet in Mario, every combination of platform and mushroom yields a slightly different challenge which your brain must process; even if you’ve played the game for years, you’ll still find challenge in a new scenerio (or, really, combination of elements).

This moves you very quickly towards an n^2 model of content instead of an n++. This is a desirable place for an MMO developer to be in, because you simply cannot produce enough content in the n++ model.

Even if you are producing “realistic” AIs, there’s a lot to learn from this approach. Give each oponent a different tactic and make that difference clear to the user, and you’ll quickly have a lot more depth to the experience. The brain is quick to identify and break these patterns, often as developers we try to fight against this using randomness and other means. But I generally have found that doesn’t produce a satisfying result. Instead, let your users see and master the base level patterns (this will make them feel smart); then combine them to produce the real challenge. These combinations will keep them busy far longer than any single element can.


Hmm, everyone here seems to have already said all the cogent stuff. I'll throw in my two cents anyway:

- AI that is totally "unpredictable" is often frustrating to play against, because it creates a total lack of what Doug Church calls Perceivable Consequence - behavior patterns that the player can observe and base future strategies upon. I think there will always be games that use very simple, almost 100% deterministic "AI", because they create very easily perceivable consequences for player actions, and provide a type of challenge that is very well-liked and well-established (Mario, Ikaruga, etc etc).

- I think what the guy is bitching about in the editorial is AI that sucks in ways that destroy immersion. We expect a certain level of intelligence-seeming-ness from enemy guards and even hellish monsters, and when one of them runs in place endlessly like a jackass, we're made acutely aware that he's just a bundle of (poorly implemented) algorithms. Which, I agree, sucks.

- The "calvary charge" attack pattern, has the added effect of making for very boring combat - chop the big dummy down as he runs towards you. It's maybe okay for silly looking floating skulls, and maybe for the scary crispy zombies from HL2, but anything more human-oid deserves better (even if it's just a slightly more complex, 100% predictable pattern).

- I don't think it's paradoxical at all to have enemies that *seem* very human (or beastly, or alien), appear to have a conscious mind that thinks about things, but also have them be predictable enough for the player to strategize around and defeat. I'd say the Thief games are one of the very best examples of this. The guards have so many little human touches - they carry on conversations with one another, convey what they're thinking via audio and animation, clearly announce transitions from different states - yet they're still dumb as dogs and easy to outsmart once you understand how they work.

- I do think that a game with a much more complicated (thus, generally, "unpredictable") AI could work very well... if the game were structured around this complexity. I'd like to see a game that pits you against a single, fairly canny individual in a trap-filled maze or suchlike, rather than the usual massive hoard of zombie morons. He might know how to sneak around, follow you from a distance, set traps for you. You'd have to consider what he knows about the situation relative to what you know, and be fairly paranoid and watch your back. Could be cool. Anyways, yeah, just wanted to point out that even the Perceivable Consequence thing isn't a hard-and-fast rule - even if it's a Good Idea for the types of games most people are making these days.


It's strange to compare Mario's flying turtle to a space marine. They may both represent creatures metaphorically but serve different purposes gameplay-wise. That turtle may as well be a bouncing boulder rather than an animal. Perhaps an alternative way of viewing AI is that their purpose in the game's universe is to entertain the player by creating gameplay and/or immersion. Along those lines, it's odd to distill all the AI in games to any formula. It's like saying dishes with the most oregano are the best tasting.

As for AI randomness, it depends highly where that randomness is applied. I think a Halo developer once mentioned that sticking a grenade on a monster produced the deterministic result of them running away but the randomness was applied to the location that they ran to.


The goal behind designing a better AI is not only to improve the experience in today's games but also to make new things possible. A smart AI is not necessarily controlling ennemy units but could be an ally as well, or could even be neutral and subject to diplomatic moves.

Imagine starcraft when you could say "Lieutenant, take 10 mens and 2 medics and bring this small base down while I am preparing the final assault"

Imagine a wargame where you would be at the head of a military hierarchy where your officiers may be smart enough to handle a battle decently but where only you would have the psychology needed to defeat your human opponent. Wouldn't it be great ?

Benjamin Graner

I think what he's talking about are enemies which feel real. A good example (for me, you may disagree) were the pig-guys from Wind Waker. They were the one's that had the spears and lanterns.

They weren't challenging, and I wouldn't consider them 'smart' but they did have pretty believable behavior. They walked around, they stopped and sniffed for you, they reacted to you in a believable manner, etc. They behaved pretty much how I would expect a low-level pig-guy to behave.

I think gamers are just getting frustrated with AI that is so clearly a finite-state autonoma. When he says surprise us, I don't think he's saying outwit us, but rather suprise us by making their reactions and behaviors more lifelike.


I think Kasparov said it best. And I quote from http://www.abc.net.au/science/news/stories/s789071.htm

He stressed the importance of psychology in chess between one human player and another, and described the "psychological discomfort" involved in adapting to a confrontation with a machine.

In chess with humans, "you're always attempting to impose your decisions on the personality of your rival. A game is always an exchange of errors, of imprecisions, it's psychology, there's never complete exactitude or purity in a game of chess," he said.

"But playing against a machine, beyond a certain point, to win or even to save the game you have to play with absolute exactitude, which is not a human quality. Knowing this specificity of your rival creates a psychological discomfort which is very difficult to overcome."

The comments to this entry are closed.

The Games

  • Energy Hook
    3D grappling-and-swinging-and-running-on-walls-and-doing-tricks ... with a jetpack ... for style!

Jamie's Bragging Rights

  • Spider-Man 2
    The best superhero games of all time Game Informer
    Top five games of all time Yahtzee Croshaw
    Top five superhero games of all time MSNBC
    Top 100 PS2 games of all time Official Playstation 2 Magazine
    1001 Games You Must Play Before You Die Nomination for Excellence in Gameplay Engineering Academy of Interactive Arts & Sciences
  • Schizoid
    Penny Arcade PAX 10 Award
    Nominated for XBLA Best Original Game
    Nominated for XBLA Best Co-Op Game