This year's AIIDE (Artificial Intelligence and Interactive Digital Entertainment) conference seems to have been quite the show. We're fortunate that around the web some very fine people have gone to the trouble to document what happened in detail and with keen insight ( Andrew Stern: [1], Michael Mateus: [2], Robin Hunicke: [3]). Others? ...
A sound-bite summary is unfair, but we have to start somewhere. Not knowing where to begin, I'll turn to Robin:
But after all my travels, one thing is perfectly, crystal clear: Game AI is in a rut...If anything, todays admittedly complex game AI is “just the price of entry” for modern titles. When it fails, ratings suffer - but when it succeeds… do players really notice? Managing the set pieces of gameply (playing “traffic cop") won’t take titles beyond the 8’s, and they won’t push gamelay in ways that broaden the market. For that, we need to invest in new approaches, incremental change… and tools that harness (and expand) the power of human animators, artists and designers.
In the end its hard to see how Game AI will matter much without the game design to amplify its value. Or as the old wry line went: why bother with exquisitely mannered and deeply intellectual Quake bots - they're dead or you're dead in 10 secs. Who'd know. So there has to be a bootstrap - between game design *and* the AI design. Where would you start?
I think that where you start depends mainly on whether or not you really believe that a game featuring AI characters is equivalent to an interactive story. Personally, I'm sure that this idea is viable, so I start where most storytellers start - with the characters and their goals, problems, and motivations. You're right not to hone bots that only live for a few seconds into fully-fledged characters. Those are cast as extras - cannon fodder. Look for interesting characters instead.
Look for a decent villain, for instance - those are always good to have! Who's the man behind the curtain? The one who sends those bots the player's way? That could be an AI character worth developing as exquisitely mannered and deeply intellectual! Make him complex! Let the player discover what motivates him! Make the boss fight emotional! That alone doesn't ensure you first place in the "Make The Player Cry" contest yet, but it would be a start. Definitely it would be a step towards product differentiation - how many games can you name that introduced memorable villains, compared to how many movies? (No, franchised villains don't count - I said introduced.)
Game developers going that way will really have to learn how characters and their stories work, instead of assuming that they already know that part. I firmly believe that interactive stories can only be developed on the basis of a solid knowledge about storytelling in general.
For those starting out with the idea that AI characters can somehow be developed outside of story structures, I don't have any advice. They'll have to find something else that they get their structure from, but I don't know what that could be. Anybody else?
Posted by: Dirk Scheuring | Jun 09, 2005 at 12:13
Dirk, it sounds like you are talking more about putting good storytelling into games than about AI. While I agree with you, I don't think we need a new breed of AI so that we can have interesting villains. For that, we need to invest more in professional writers as designers, like Obsidian has. Star Wars: Knights of the Old Republic introduced several very interesting and memorable characters, including the villain. Depending on the choices you make earlier in the game, the boss fight has the potential to be very emotional, and KotOR currently holds first prize, IMO, in the "Make The Player Cry" contest. I got close, anyhow.
But none of that requires good AI. That just requires a talented writer, and the ability to script the villain's fight moves around his boss-fight dialogue. KotOR is one of my all time favorite games (and I'm playing KotOR2 now specifically for the character writing), but as much as I love it, it didn't introduce any non-combat AI.
I think the first frontier for stunningly new AI is MMOGs. In single player games as they stand now, current-generation combat AI is generally enough to provide the sort of gameplay players are interested in (i.e., there's little demand for "exquisitely mannered and keenly intellectual Quake bots", as Nathan put it), and character-driven stories would be better off hiring a good writer than trying to create AI that would write the story for them. But MMOGs have more opportunities for innovation, and their social nature makes them the perfect match for social AI.
In the end its hard to see how Game AI will matter much without the game design to amplify its value.
Well, yes. Which is why we can't just have game AI handed to us by non game designers. AI created by game designers who understand good gameplay and who want to expand game AI beyond Quake bots and expand gameplay beyond killing said Quake bots, has the potential to change the types of games we make.
I agree that we can't just create cutting edge AI and expect others to build games around it. We have to build the AI and a game that utilizes that AI to its best potential at the same time. Similar to Unreal and the Unreal engine: one wouldn't have been much without the other originally, but now the engine is used for all sorts of other games. I think we can get to the point were new sorts of (non combat) AI are used in many different sorts of games, including single player games, but the first few games to use it will have to be built in concert with the AI itself, by game designers who understand both good gameplay and good AI.
Posted by: Samantha LeCraft | Jun 09, 2005 at 12:57
Interesting to constrast Robin's evaluation of game AI "in a rut" versus the rosier outlook taken by the NYTimes article.
I'd suggest game AI is not in a rut, per se - instead it's standing on a slope, part way up a massive mountain, looking up at the distant peak, gone as far as it can go on foot, and now has to break out the mountain climbing gear and prepare for some very strenous, painful rock climbing.
Also, a comment on Robin's point about few women at AIIDE — that's very true, but let's not forget two of the primary voices that have had such an influence on framing, guiding and motivating this work for many researchers, particularly for interactive drama - Brenda Laurel and Janet Murray. And Marie-Laure Ryan is a newer voice with lots of new excellent analysis as well.
Posted by: andrew stern | Jun 09, 2005 at 13:02
Great opener, Nathan. And I think Robin is spot-on as usual.
In the end its hard to see how Game AI will matter much without the game design to amplify its value. ... there has to be a bootstrap - between game design *and* the AI design.
Completely agreed. On its own, AI doesn't get you anywhere. It's similar to really cool graphics. You go "wow" for a moment, but then (unless you're a complete AI geek) say, "okay, so what?" If the AI doesn't enable any interesting and meaningful decisions -- that is, gameplay -- then it's at best window dressing.
Even more than that, while game-like apps like Facade are impressive, getting anything like that over the many hurdles leading toward making an actual product is difficult to say the least. AI alone doesn't sell to publishers or to the primary gaming audience (and while I'm glad to see Bing speaking at a conference like this, I'd be happier to see EA put its money where Bing's mouth is as far as AI and similar innovation goes -- Will's maverick project aside).
Where would you start?
Where I started three years ago was with trying to re-conceive both how AI can contribute to gameplay and how gameplay changes in the face of new AI (beyond pathfinding and combat). For me this involved re-thinking a lot of pretty basic assumptions about how we (narrowly, IMO) conceive gameplay and interactions between players and AIs. And frankly, I don't think shoehorning gameplay plus AI into 'interactive story' is the way to go, any more than shoehorning gameplay plus graphics into 'interactive movies' worked well. Games are a new form of interaction and entertainment, and AI helps differentiate us from other existing forms: AI is something games can do that movies and stories cannot (and have no need for).
The route we've taken at Online Alchemy is one that we're calling 'artificial psychology' -- AI characters with their own motivations, personalities, nuanced emotions, emergent relationships, memories, etc -- and all as relevant to the game as a character's thoughts and feelings are to a traditional story. The underlying idea is that you don't want to have to drive them around like an author does in a story; you wind them up and let them go. This is neither authorial nor directorial; it's something different. The gameplay comes from the situations (not quite as random as it may sound from this) in which the player(s) are faced with their own goals and decisions.
I strongly believe that we can add significant depth to existing game experiences (especially MMOGs) and create new kinds of gameplay. Whether we will or not depends on whether such efforts can garner the attention of funders (investors, publishers, etc.), and whether they believe there's an audience for anything beyond click-and-shoot or kill-monster-get-gold.
Posted by: Mike Sellers | Jun 09, 2005 at 13:03
I'm glad you mention 'artificial psychology', Mike. It seems that the questions of how to make AI mimic (yes, mimic) human behavior -- at least well enough to suspend disbelief -- is one of the most important sets of issues to be considered by those working in and with AI, especially in games.
So much is determined by what effect you're trying to have on players' psychology and the flow of the game. It seems we've gotten pretty good at making AI that can play a game (at least a fairly mechanical one) convincingly and effectively without resorting to what most people would consider cheating.
However, what if we want an AI that gets players to believe there is a human commander behind the scenes of the enemy's army (or even yours)? A human makes human errors and can be "psyched out," a tactic very important to the history of war and conspicuously absent from almost all (or all) video games dealing with war.
Or how about politics? Admittedly, politics is an insanely complicated and subtle beast to tackle. Yet, perhaps a game AI would only need an elementary understanding of politics
to create an interesting political landscape for players to play with. After all, most players won't want to be outsmarted! They just want entities that can be manipulated and who appear to be working in the interests of their nation's/character's interests.
As with most things in game development, it's all about cheating. It's about finding the smallest working solution. But to know what that smallest solution is we need to learn from the study of psychology, system dynamics, player studies/ludology, and more.
Never forget that a child is willing (and predisposed) to see a human-like being in a stuffed animal and grown adults often treat their cars like a child of their own flesh and blood. People want to see humanity in the things with which we interact. As game developers, we need only to design our games to inspire such imagination in our games' players and then compliment them with the corresponding AI.
As as we find what the complimentary AI might be, perhaps then we will know "what's next" for game AI. No doubt it will be varied and exciting. =)
Posted by: Darren Torpey | Jun 10, 2005 at 08:36
I was noodling on storytelling a while back, and coming from that direction, after several iterations, came to a similar place that the AI folks seem to be reaching, although I wasn't totally innocent of what people like Chris Crawford and Will Wright have been saying about the subject.
I really think that AI is part of the proper vehicle for telling story, expanding mobs from simple xp bags to mobs becoming story objects as well, so that any NPC has the potential to become a true character, for the right player character. I also think that we need to consider closely elements of games such as game physics and user interface as dramatic elements, as well as functional game elements. These are, in many ways, the setting of any in-game story.
I also think that, while professional writers could help storytelling in games, they also may have blinders regarding the interactive nature of games, which makes games relatively unsuited to traditional narrative storytelling. In fact, I'd expect the ultimate story objects to generate scripts via chatterbot modules, which could both be seeded with plot elements and employ heuristic elements to incorporate player input. The storytelling role in that case would be more directorial than authorial. Ideally, one would set the stage, and allow the player to improvise with the AI actors.
If my general line is of interest, I have a series of posts on f13( http://forums.f13.net/index.php?topic=3456.0 ) where I was developing my thoughts about this. It's all a bit rough and slightly salty, which is what you get when you do your thinking on a public board.
Posted by: Evangolis | Jun 10, 2005 at 17:13
I feel storytelling can add purpose to a game, while AI is just a vehicle for it. Personally, that's my point of separation in MMORPGs between "Gametic" and "Virtual Worldy". The former provides directives to players while the latter expects them to do so themselves (for the most part, in both cases). The best way to provide directives is through a narrative environment, leaving it to the players to avoid it if they want (to grind).
Unfortunately, even the voluntary narrative-based MMORPGs like CoH, WoW, and EQ2 really involve little more than the same old tried-and-true Kill X/Collect Y. The quality of writing is almost unrelated, whether it's a single-step quest or a twenty-step long story arc. Ultimately, how a player Grinds is almost the same to how they Quest. The lowest common denominator is XP-per-kill, the favored method to regulate it is awarding it at the end of a fight, and because players want predictability they can master to most efficiently advance to new abilities/thing, the mobs should not be too smart.
Ergo (in my opinion), players won't really appreciate, and may even get annoyed by, smart AI in the XP-per-kill games that currently dominate attention.
Having said that, I think there's an acceptable range. For example, most mobs in WoW are pretty stupid, which sort of advertises PvP. Meanwhile, the mobs in CoH are both numerous (seems 3-to-1 against players at all times) and seem to know their abilities (which are the same players can get), as becomes more obvious at later levels. They're still designed to be killed, but at least there seemed a robust attempt to make fighting interesting.
Posted by: Darniaq | Jun 11, 2005 at 12:28
Just FYI, the presentations from (almmost all of) the AIIDE invited speakers (Doug Church, Chris Crawford, Damian Isla, Jon Schaeffer, Will Wright) are now available on-line at the AIIDE web site.
-M
Posted by: Michael Young | Jun 16, 2005 at 12:29