The Escapist Presents: Escapist Report: AI Innovation

 Pages 1 2 NEXT
 

Escapist Report: AI Innovation

In depth, expert interviews on where video game A.I. is going.

Watch Video

I've always thought AI could be truely magnificent if there was a great movement towards making it better, like Graphics or Physics have had in the past.
Ultimately though there's a big barrier between what the AI could be capable of and what the player expects. Really if the shooters shoot and the talkers talk nobody really faults it.

holy cow in the sky, a serious show on the escapist?, really great report and i am truly looking forward to the videogame war where they try to up each others AI and story.

First time I saw the show, but I liked it, felt a lot like a documentary, but from a gamers perspective, which is nice, because I think it may offer some more insight into the "problems" that needs to be overcome to attain any "real" progress with the video game medium.

They should have included S.T.A.L.K.E.R though, that game has some amazing AI, with packs of animals migrating, eating, sleeping etc, human stalkers who organize patrols, search for artefacts, sleep, play guitar and so on, and it's all dynamically generated, nothing is scripted.

Alas, CSG Game World finished the feature, but never included it into the "vanilla" Shadow of Chernobyl, most likely because it risked having important story-characters wander of and get killed, rendering the "main quest" impossible to complete. However, the dynamic AI (A-Life) have been reactivated in several mods, and I think it is present in S.T.A.L.K.E.R: Clear Sky as well.

I've had the sense for years now that the end was nigh for graphics being the all important stage in the development of a game, beyond maintaining the status quo, and it was the opponent a game presented you with that would make the difference in the desirability to play that game. A good opponent makes for a great game and depending on the type of game, it is very desirable to have a variety of opponents to keep things exciting.

Awesome video, but come on....

Terminator much? :(

/scared

way too overly dramatic. good concept, but seriously, there was nothing informative about this video. the information included was minimal, and it seemed the video was just a litany of obstacles rather than an expose on AI in games. also talk to the guys over at monolith about the fear AI. it was radically different than normal AI and made that game standout among a million other shooters. its probably the best AI ive ever encountered in games.

The problem is that scientific research into AI has proven tricky too. The idea that we could develop dynamic, complex, unique, good stories through AI without a REAL sentient program inside to design such a world is unlikely at best. Procedural levels/content has and continues to lack anything better than either a cookie-cutter world, or a random world with no real unique feeling/characters/etc.

Spore, though a bad game, might have a glimmer of answer in its use of the players as designers, but the gameplay itself was very cookie cutter.

Nice video, but it would have been cool to hear about some breakthrough instead of how stuck we are in AI development.

Artificial Environmental approach to AI is where it's at with the cool cats.

randommaster:
Nice video, but it would have been cool to hear about some breakthrough instead of how stuck we are in AI development.

It would have been nice to report such a breakthrough. Sadly, most AI research doesn't go toward developing better games, and the development that does go toward game AI is grossly underfunded and underappreciated.

I tend to think of AI in games as it relates to the movie industry, and the saddest truth of the world of entertainment (in all forms) is that the blockbuster system dictates "more, faster" rather than "better." Game AI is a lot like practical effects like makeup and puppetry - they are only implemented when the content demands it, and the quality can swing wildly from "amazingly well done" to "slapped together in 20 minutes."

Any chance that transcripts will be made available for these things?
I like reading. Weird, in this day and age, I know.

What is up with that narrator? It almost sounds like you used a text-to-speech program to read a script.

Really nice report and a good one too. One of the areas ( AI ) I'm most interested in concerning design since this is what games are lacking but also one of the most challenging parts of it since you challenge a human. It's like building a chess computer to beat a chess master. You're challenging human abilities with a machine and that's most probably the trickiest job you can do. Let's see what future has in stock for us. Hopefully more interesting and better games through better AI.

CaptainCrunch:

randommaster:
Nice video, but it would have been cool to hear about some breakthrough instead of how stuck we are in AI development.

It would have been nice to report such a breakthrough. Sadly, most AI research doesn't go toward developing better games, and the development that does go toward game AI is grossly underfunded and under appreciated.

I tend to think of AI in games as it relates to the movie industry, and the saddest truth of the world of entertainment (in all forms) is that the blockbuster system dictates "more, faster" rather than "better." Game AI is a lot like practical effects like makeup and puppetry - they are only implemented when the content demands it, and the quality can swing wildly from "amazingly well done" to "slapped together in 20 minutes."

I know that advances in AI are slow and I really appreciate all the work that goes into making these videos. I just thought that this only covered the current state of AI, which isn't going anywhere fast. I would have had a more documentary- or report-like feel if there was some discussion about the major advances in AI.

Or you could, you know, completely revolutionize the field of artificial intelligence so that you haave some cool stuff for the video. That seems totally plausible.

AI. No matter how well you program it, AIs still get stuck or walk into a wall. AIs can probably be programmed much better then "run here to here, shot people who you see and go to power-ups." With some effort, you could program an AI to recognize its surroundings and calculate what objects are around it and "make" decisions on what to do. Unfortunately this would suck up a lot of processing power. On more advanced modern systems I guess that's possible but it still seems like every time computers each out another section of power, its immediately taken in order to produce slightly more real ripple effects that on a puddle of water. I think before AI can evolve beyond what are still mostly very dumb bots with simple AIs we need to first decide that were not going to funnel all our advances in processing speed on adding to graphics and instead focus more on AIs. Graphics are largely really good now and I hope that really good is good enough and we can work on other things because I foresee a diminishing return value on continued graphic advances. Its getting so good that making it better offers minimal changes in the visuals but takes more power While AI can still be vastly improved and even small advances can make a real difference in how good it is.

I cant belive that this "AI Innovation Report" didint included an interview with BIA, kinda disappointing

randommaster:

CaptainCrunch:

randommaster:
Nice video, but it would have been cool to hear about some breakthrough instead of how stuck we are in AI development.

It would have been nice to report such a breakthrough. Sadly, most AI research doesn't go toward developing better games, and the development that does go toward game AI is grossly underfunded and under appreciated.

I tend to think of AI in games as it relates to the movie industry, and the saddest truth of the world of entertainment (in all forms) is that the blockbuster system dictates "more, faster" rather than "better." Game AI is a lot like practical effects like makeup and puppetry - they are only implemented when the content demands it, and the quality can swing wildly from "amazingly well done" to "slapped together in 20 minutes."

I know that advances in AI are slow and I really appreciate all the work that goes into making these videos. I just thought that this only covered the current state of AI, which isn't going anywhere fast. I would have had a more documentary- or report-like feel if there was some discussion about the major advances in AI.

Or you could, you know, completely revolutionize the field of artificial intelligence so that you haave some cool stuff for the video. That seems totally plausible.

I agree here. Overall the video was excellent. The shortcoming stemmed from just presenting us with the problem, as opposed to the problem and how people are working to change it.
I would have liked to see someone presenting research saying "We're doing *this* to move AI forward." Or even compare some really well done AI to poorly done AI (both created in the same time frame).

The only thing that really got to me was the voice of the narrator. Didn't feel like it fit the overall feeling of the video. Other than those minor gripes, I enjoyed it! Keep it up guys, I'll be looking forward to more :)

No mention of L4D's AI Director?
Other than that, I enjoyed it.

Frizzle:

randommaster:

CaptainCrunch:

randommaster:
Nice video, but it would have been cool to hear about some breakthrough instead of how stuck we are in AI development.

It would have been nice to report such a breakthrough. Sadly, most AI research doesn't go toward developing better games, and the development that does go toward game AI is grossly underfunded and under appreciated.

I tend to think of AI in games as it relates to the movie industry, and the saddest truth of the world of entertainment (in all forms) is that the blockbuster system dictates "more, faster" rather than "better." Game AI is a lot like practical effects like makeup and puppetry - they are only implemented when the content demands it, and the quality can swing wildly from "amazingly well done" to "slapped together in 20 minutes."

I know that advances in AI are slow and I really appreciate all the work that goes into making these videos. I just thought that this only covered the current state of AI, which isn't going anywhere fast. I would have had a more documentary- or report-like feel if there was some discussion about the major advances in AI.

Or you could, you know, completely revolutionize the field of artificial intelligence so that you haave some cool stuff for the video. That seems totally plausible.

I agree here. Overall the video was excellent. The shortcoming stemmed from just presenting us with the problem, as opposed to the problem and how people are working to change it.
I would have liked to see someone presenting research saying "We're doing *this* to move AI forward." Or even compare some really well done AI to poorly done AI (both created in the same time frame).

The only thing that really got to me was the voice of the narrator. Didn't feel like it fit the overall feeling of the video. Other than those minor gripes, I enjoyed it! Keep it up guys, I'll be looking forward to more :)

I almost responded to this with a post about Samus, which would have been really confusing and embarrassing.

Other than that, it would have been nice to compare something like MGS2's last boss fight, where it would read your controller's input and react to that, and the AI that goes into fighting game, for example.

My bro was telling me about a bit of research that was done into making a an A.I. which could edit its programming to adapt to players. Unfortunately the AI kept "choosing" to delete itself so nothing ever came of it. Interesting idea though...

Jimmy_shredshot:
My bro was telling me about a bit of research that was done into making a an A.I. which could edit its programming to adapt to players. Unfortunately the AI kept "choosing" to delete itself so nothing ever came of it. Interesting idea though...

Perhaps it was trying to commit suicide?

On topic, never thought I'd see the day when there came a serious show on The Escapist. That said, great show.

Headless Zombie:

Jimmy_shredshot:
My bro was telling me about a bit of research that was done into making a an A.I. which could edit its programming to adapt to players. Unfortunately the AI kept "choosing" to delete itself so nothing ever came of it. Interesting idea though...

Perhaps it was trying to commit suicide?

or sabotaging itself from the future?!

Why keep on fighting when your soul purpose is to lose?

I always thought that the enemy group AI in Metal Gear Solid 2 was an almost hideously important branch of game-AI development, which went absolutely nowhere. I play games now where the soldiers don't feel as intelligent as they did in that game.

They walked prescribed routes...unless they heard a noise, or saw something suspicious. They checked in on one another every few minutes; if one fell, and was reported absent, others went to check on him. If they couldn't find him, they conducted a search. If they found him unconscious, they woke him up; if he saw you knock him out, they conducted an alerted search; if he didn't, they took a suspicious look around. If you shot out their radios, they were cut off from the network of communication; they ran for help, rather than engage. They covered around objects, and attacked from multiple angles and sides in unison. If you took enough of them out, they would stop actively advancing and start lobbing grenades from behind cover.

I'm not saying the AI was anythign astounding (especially not when compared to actual military squad procedures), but I feel like it was such a quantum leap forward in that department...and...just...went nowhere.

Gears of War monsters shoot at you until you shoot them to death. And then it's over.

teknoarcanist:
I always thought that the enemy group AI in Metal Gear Solid 2 was an almost hideously important branch of game-AI development, which went absolutely nowhere. I play games now where the soldiers don't feel as intelligent as they did in that game.

They walked prescribed routes...unless they heard a noise, or saw something suspicious. They checked in on one another every few minutes; if one fell, and was reported absent, others went to check on him. If they couldn't find him, they conducted a search. If they found him unconscious, they woke him up; if he saw you knock him out, they conducted an alerted search; if he didn't, they took a suspicious look around. If you shot out their radios, they were cut off from the network of communication; they ran for help, rather than engage. They covered around objects, and attacked from multiple angles and sides in unison. If you took enough of them out, they would stop actively advancing and start lobbing grenades from behind cover.

I'm not saying the AI was anythign astounding (especially not when compared to actual military squad procedures), but I feel like it was such a quantum leap forward in that department...and...just...went nowhere.

Gears of War monsters shoot at you until you shoot them to death. And then it's over.

Great point. It reminds me of other moments in gaming where I felt like congratulating the enemy on having more though than just approaching the player like a zombie with ranged weapon in hand:
- Half-Life: The first time the soldiers spotted me, they talked to each other and seemed like they tried to flank me. Even more amazing at that time was when I tossed a grenade, they ran from it. This was a huge step up from my experiences with Doom and Duke Nukem 3D.
- No One Lives Forever 2: The enemies use of objects in a room and the way they would search for the player's character. I thoroughly enjoyed watching the enemies actually appear to live in the place they were, rather than just waiting for the player.
- Unreal Tournament: Tweaking custom bots to mimic play styles was a lot of fun for me.
- XCOM UFO Defense: Stands out because I could swear that aliens would rather lay in wait to ambush my soldiers than charge through a door after I would ambush them.

Good AI, even if only slightly improved on from previous games, makes for a great game. Many games I consider classic in some way stand out not because of their graphics but the challenge presented by the enemy did not appear to be scripted and repetitive.

this actually had very little to do with A.I. The central thesis seems to be that AI has been crowded out by graphics in competition for processor cycles. This is a silly idea, since the biggest obstacle to A.I. in games isn't hardware but programmer attention and ability.

Why didn't you interview an actual A.I. researcher? Game A.I., as it stands now, is only tangentially related to traditional A.I. research. On top of that, you spend a lot of time discussing things like interfaces and procedurally generated content, which has almost NOTHING to do with A.I. whatsoever. I was expecting to see something on the potential applications of existing A.I. technology in gaming; i have to say i'm disappointed.

Actually I've wondered how to grow a story for a character as opposed to writing one in a game. Take AI elements from different game genres. Take a game like Fallout or Oblivion, create factions with a system of enemies coded in, give the faction AI a goal and reactions to player action. Put in an observation system so you don't have telepathic groups.

The world shapes itself based on the events put in motion by the factions and the player. A cause and consequence system with the AI moving forward and there being no real set ending for the game. In one game, Faction A might be able to get an upperhand. In another A and B are taken out by C. Obviously the player would be able to affect events by having his own reactions to what is going on.

Each faction could have its own moral code. Perhaps Faction A doesn't kill women or children and takes over a village after X number of defenders die, graciously accepting surrender. Meanwhile Faction B loots villages, only kills a few, and then leaves to be able to come back and fleece them again. Then Faction C has no qualms about putting anybody to the sword that raises a hand against them. And so on.

Then the player can make a choice based off his own ethos or play the factions off each other. Possible even do enough acts to become a force unto himself and gather followers.

Instead of a quest system, a chronicle system that details your bigger exploits as well as some of your more daring side adventures. There would be special things in the game that also affect the outcome without being directly tied to a faction. Perhaps an enemy such as a Rogue Bot or Giant Dragon keeps one area in constant dispute because none of the factions can beat it.

Just one of my thoughts on a story that grows to a climax or some sort of endings without it being in stone from the beginning or simply one of 2 options of morality.

I noticed the crap AI when was playing RA3 today. Great game but when built about a defence with towers and what not the ai showed how stupid it really was. On the hardest difficulty level, they started sending troops one at a time towards my towers but they didn't attack them, they were just trying to fulfill some unknown goal or move somewhere. You'd think AI would have advanced at least enough so they don't sacrifice units for no reason.

CaptainCrunch:

randommaster:
Nice video, but it would have been cool to hear about some breakthrough instead of how stuck we are in AI development.

It would have been nice to report such a breakthrough. Sadly, most AI research doesn't go toward developing better games, and the development that does go toward game AI is grossly underfunded and underappreciated.

I tend to think of AI in games as it relates to the movie industry, and the saddest truth of the world of entertainment (in all forms) is that the blockbuster system dictates "more, faster" rather than "better." Game AI is a lot like practical effects like makeup and puppetry - they are only implemented when the content demands it, and the quality can swing wildly from "amazingly well done" to "slapped together in 20 minutes."

I fully agree with the statement above. I was under an NDA for the game I was working on at the time of the interview and couldn't elaborate on what we specifically were doing on the AI side. Currently in games 80% of AI is perception, meaning that regardless of backend complexity it's what the player sees that is taken as AI.

Uncharted is a good example of poor or simplistic AI that is accepted because there's enough quality animation variance to be perceived as good. In Ghost Recon we had 3 tiers of AI that the player would never know was active. It was Team->Platoon->Company and other military games follow suit. The problem in a shooter is that the player never sees that top level actions unlike that of an RTS where the troop movement is evident. One team would engage the player and another team would flank. Often the player perception was that we cheated and spawned in AI behind the player.

The key in my opinion is believable AI from the players perspective. L4D for instance has great zombies because they run and look like zombies not because there's some great zombie AI running them. Halflife 2 had great AI with all the ambient characters and the facial animation and response system. Like I mentioned in the interview, the AI is dependent on the embedded metadata in a level to have a toolkit of actions. The metadata can be static (nav mesh, cover or object data... etc) or dynamic (destructible environments, player created content... etc) but it has to be perceived as believable and not get stuck in crappy state loops. We've all seen it in games where AI X gets stuck running into a wall or just repeats the same action loop... and it breaks the whole immersion. As gamers we let a lot of little things slide if the core experience is fun. Unfortunately I'm still under and NDA and can't go into some of the cool shiznit that we're doing.

There's others side of learning algorithms and concepts of neural networks and other data based on the players actions (again dynamic metadata) that I feel people overly complicate. Simple things like having a history of where enemies were killed and having subsequent enemies avoid that area (shoot a guy and the dude next to him runs to the same spot .. wax and repeat). A lot of work goes into making a game and often AI is left to the end by the virtue that the layer of scripting and data for the AI to use in a level is often reliant on the level being in a final state. I think we'll see a number of improvements though AI is not an easy thing to get metrics on since it's the player's perception rather than the elegance of the code behind it.

Sorry for the long post but if you made it this far what current games do you think have the best AI and why?

thanks,
Patrick Sebring

Yes, R. Michael Young, game graphics have been such a burden on the CPU all this time... Oh wait, no, actually games have been offloading this task onto a seperate processor(s) since the mid-ninties. And people wonder why industry doesn't take (university) games academia seriously?

As far as I'm concerned, AI innovation in games has been slow because:
1. You could invest huge amounts of time improving the AI, and plenty idiots would hardly notice the difference (ie, some of the console tards on this very forum who thought Far Cry 2 had impressive AI). This isn't a problem for graphical improvements (the ooh shiny factor).
2. The rise of high speed internet/multiplayer gaming has meant that most players looking for (relatively) intelligent competition can just play against real opponents, reducing demand for better AI (a bit).

With some games where AI is really important, like civ or galciv, you'd think the designers had no clue about the game mechanics at all when they made the AI.

Really basic stuff like specializing cities/planets and protecting transports with combat units is usually missing.

You always end up with drooling opponents having massive discounts before it gets challenging.

"People are seeing that there are other things that sell games besides graphics"
It's taken people how long to figure this out? Seriously? I swear to god I raged when I heard that.

End users have been custom scripting AI for games with less than spectacular graphics (Warhammer 40k Dawn of War, DotA), so it's not that we aren't "capable", it's just that too many people are buying games solely because of the amount of bloom and global illumination rather than the story or actual design elements.

If your main goal is to look at pretty pictures with some level of interactivity, get Nintendo to make some QTE games for you, or just go watch a freaking movie with a controller in your hand.

Otherwise, tell them (with your wallets) that games with deep story lines and intelligent enemies are what will sell more and hopefully we'll see more.

josh797:
way too overly dramatic. good concept, but seriously, there was nothing informative about this video. the information included was minimal, and it seemed the video was just a litany of obstacles rather than an expose on AI in games.

I agree in full. Not a word a lie.

cobra_ky:
this actually had very little to do with A.I. The central thesis seems to be that AI has been crowded out by graphics in competition for processor cycles. This is a silly idea, since the biggest obstacle to A.I. in games isn't hardware but programmer attention and ability.

Why didn't you interview an actual A.I. researcher? Game A.I., as it stands now, is only tangentially related to traditional A.I. research. On top of that, you spend a lot of time discussing things like interfaces and procedurally generated content, which has almost NOTHING to do with A.I. whatsoever. I was expecting to see something on the potential applications of existing A.I. technology in gaming; i have to say i'm disappointed.

Better explained there. I really can't build on these points, everything of value has been said. Jeez, it's usually me spewing all the hate-bile on this website, but now it seems I have a few apprentices. Remember boys and girls, appreciate nothing, even if it's given for free! Could this site please stick to games critiquing, and stop the pseudo-intellectual look at games development? Want that? Go to www.gamesutra.com

UtopiaV1:

cobra_ky:
this actually had very little to do with A.I. The central thesis seems to be that AI has been crowded out by graphics in competition for processor cycles. This is a silly idea, since the biggest obstacle to A.I. in games isn't hardware but programmer attention and ability.

Why didn't you interview an actual A.I. researcher? Game A.I., as it stands now, is only tangentially related to traditional A.I. research. On top of that, you spend a lot of time discussing things like interfaces and procedurally generated content, which has almost NOTHING to do with A.I. whatsoever. I was expecting to see something on the potential applications of existing A.I. technology in gaming; i have to say i'm disappointed.

Better explained there. I really can't build on these points, everything of value has been said. Jeez, it's usually me spewing all the hate-bile on this website, but now it seems I have a few apprentices. Remember boys and girls, appreciate nothing, even if it's given for free! Could this site please stick to games critiquing, and stop the pseudo-intellectual look at games development? Want that? Go to www.gamesutra.com

actually the intellectualism is part of the reason i come here. i wouldn't be interested if i wasn't learning anything. honestly i was trying to be constructive but obviously my disappointment got in the way of that.

 Pages 1 2 NEXT

Reply to Thread

Log in or Register to Comment
Have an account? Login below:
With Facebook:Login With Facebook
or
Username:  
Password:  
  
Not registered? To sign up for an account with The Escapist:
Register With Facebook
Register With Facebook
or
Register for a free account here