A couple years ago, I had a conversation with a level designer who told me that "Someday, everything will be multiplayer, so there won't be any need for AI."
To be fair, it was bit more heated than a "conversation," as I tend to have a hard time with people who disregard the importance of AI ... particularly coming from someone enjoying a design role on a single-player game.
The statement is obviously absurd if you really think about it. There are any number of ways to attack that line of reasoning.
One counter-argument is that multiplayer really *ISN'T* taking over the world. It's come into its own over the last five years or so (which is undeniably a very good thing), but there's no reason to think we can take that curve and extrapolate it to infinity.
Following this line of argument, I could argue some combination of the following:
A) It's not justified by market data. Particularly in the much larger console market, growth of multiplayer gaming on consoles hasn't experienced anything close to the explosion that many expected.
B) Many games and game genres just don't work as multiplayer games ... for example, tycoon/sim games, most mission-based/story-driven FPS games, golf games, adventure/puzzle games (think Ico or Shadow of the Colossus) or some of the turn-based strategy games which are too slow to be playable in multiplayer.
and/or
C) There are major issues with multiplayer that are difficult or impossible to solve with technology -- most notably the atrocious behavior of many online gamers and the vast gap in skill levels between players which automatically discriminates against everyone who isn't part of the elite 10-20% of top-ranked players.
But I really don't want to follow that line of reasoning because someone will inevitably accuse me of criticizing multiplayer, which I'm not.
And then the thread starts to devolve into a contest between multiplayer and single-player, which is really pointless.
I love both apples and oranges. I believe they are both tasty and healthy fruits packed with essential vitamins and nutrients, and both have a bright future ahead of them.
---
The second line of argument is to say, OK, let's assume that that level designer was right, and multiplayer really IS somehow going to completely take over all forms of gaming (which, again, is silly, but let's accept it for the sake of argument).
Even if that were true, AI would STILL have an enormous role to play in the future of gaming. Here are the roles AI would still be able to play in an exclusively multiplayer world:
1. Generic opponents/human analogues -- for example, the "bots" of Quake, Unreal, or any other FPS that provides them. Even in a multiplayer world, players still want to have generic opponents that can fill empty slots multiplayer games. Inexperienced players also want to be able to practice against papier-mache bad guys before going online to get crushed by real players. As another example, in many strategy games, if a human player disconnects before he's defeated, you want to be able to replace him with an AI player that can keep on fighting so that the game isn't suddenly thrown out of balance.
2. Minions -- i.e. units under direct player control. All of the armies and villagers and other units a player controls in a strategy game or a God game such as Black & White require extensive use of AI, and this applies equally to single-player and multiplayer. The players in a football or basketball game still require AI if the human player is the coach.
3. Teammates -- these are units under AI control which fight alongside the player but which he does not need to micromanage. For example, "lancemates" in a MechWarrior game, "wingmen" in Wing Commander, or "squadmates" in a tactical shooter game.
4. NPCs -- the non-player characters a player needs to interact with. In an MMO such as World of WarCraft, this includes all of the merchants, quest givers, guards, and generic conversation givers scattered throughout the world. These may be relegated to simple conversation trees and bartering interfaces, but there's still an enormous amount of room for AI ... particularly if you imagine some of the more interesting behaviors in an RPG such as the upcoming TES:Oblivion (and its much-vaunted "Radiant AI" system), which are also possible in an online context.
5. Commanders -- This is a very uncommon role for AI but it's interesting enough that I think it merits attention. When I worked at Microsoft, I spoke with some folks at Microsoft Research who were working on a Wing-Commander-esque MMO called "Allegiance." In Allegiance, players could either play as individual combatants or as commanders who issued tactical commands to the other players. Unfortunately, a few years after launch, they found that no one wanted to play as a commander since all the other players ignored their orders ... so they were looking to build an AI system to replace it!
So in this case, you have sort of the opposite of the "minion" role, where the AI is planning the overall strategy and issuing orders to the human players in a multiplayer game.
---
In conclusion: I don't think there's any chance that single-player gaming is on the way out anytime soon ... and no matter where gaming takes us, game AI will always have plenty of room to grow.
---
Final thought:
Your question also touches on the Big Question of game AI, though, which is: How do we take the vision of the kinds of AI we all know we can make and carry those into the reality of the market?
The answer is game design: iterative improvements on the same tried-and-true patterns of game design aren't going to get us there. AI opens new doors for game design. Not all of the possibilities are entertaining, not all of them are feasible, and not all of them are marketable, but I would argue that some of them are all three of those things, and that's where (in my opinion) game AI has been trying to get to from the start.
But I think I'm already way over my quota ... So I'll stop now and promise to talk more about that in a future post.
Posted by Paul T at March 7, 2006 08:24 PMWell, I work for a company whose main goal is to make massively multiplayer games (SOE), yet I'm working on a single player game. That's at least one data point for the theory that there will always be both types of games. :)
I think it is pretty valid (and has more supporting evidence than just the above) to say that story-driven single player games will always be *a* focus in the market, if perhaps not the main one. There will always be a demand for both, not just because of our AI, but because of the nature of players.
But it's still entertaining to postulate what the future of game AI holds - as Paul points out, the design possibilities afforded by new techniques in AI seem the most interesting. I think there will be more multiplayer games that put some distance between what you the player decide and what your character does - you'll see MMO's where you control a character or set of characters that actually have more agency than typical MMOs today. You provide you character(s) with a much simpler set of inputs than the number of actual actions they can perform. You make them do cool things, and it's easy to do (and has limited interface requirements to make them do it, ie. works well on a console).
The problem, though, is that it seems like we're the ones who really have to convince designers to take risks in their designs with the AI - they're not technical enough to push those boundaries, and it can be tough to give up control of the design of a system that greatly impacts the game to someone that isn't a "designer". It really does seem like we have to sell them the idea. Then we have to convince the producers. Then we have to convince the marketers. And then of course we have to convince the actual consumers.
Damn. Sounds like a lot of work. :)
> Will AI be relegated to the roles that humans don’t want
> to play (e.g. shopkeepers) and “ambient life” (e.g. rats,
> rabbits, roaches, and crows), or can characters with an
> inner-life (as Bruce puts it) complement the human
> characters such that there will always be an equal
> demand for both?
I think the complementary scenario is the most interesting. Not an easy one to bring to pass, as it requires (as Borut notes) changes at most if not all levels of the game development organization. AI-heavy design in a multiplayer world is a fairly new thing; perhaps as different from previous design tropes as MMOs are from single-player games.
> Most of us (on this blog) focus on creating believably
> human behavior. We want AI that can navigate, take
> cover, flank, and use vehicles as well as the human
> players.
I kind of wanted to take issue with this. Are more effective flanking, strafing, and navigating seen as roughly synonymous with "believably human behavior"? I can see this in some games, but I'm not sure how much better QuakeBot AI has to be. It's significant that The Sims, mentioned above and probably the most AI-heavy game out there, focuses almost entirely on domestic behaviors, not FPS-style flanking or the like.
Posted by Mike Sellers at March 12, 2006 08:32 PM(Apologies for the lack of quotes above; I didn't realize html codes for italics wouldn't come through. To clarify, the first and third paragraphs are quotes from the original post above.)
Posted by Mike Sellers at March 13, 2006 04:47 AMI went ahead and added '>' indents into your original comment, Mike, to make it more clear what you were quoting.
I think you raise two great points, both of which I totally agree with.
One is that we have to avoid being anthropocentric when thinking about game AI, and avoid thinking about it in terms of strictly _human_ intelligence. Working on Metroid games, I work on dozens of dramatically different creatures and life forms that each require their own totally customized AI systems, and part of the challenge of each one is trying to figure out how each particular type of creature would think. If we limit ourselves to human behavior and certain aspects of human intelligence, we're also forfeiting a lot of possibilities for AI and constraining our options unnecessarily.
(I'm reminded of when chess grandmaster Garry Kasparov was defeated by IBM's Deep Blue, and remarked that he felt as if he were playing against an "alien intelligence" ... and I think that in many cases, giving the impression of a specifically alien kind of intelligence should be our goal.)
The other point ties in to the Big Question I mentioned earlier, which is that game design needs to evolve, and at least some of the evolutionary paths we can take will require us to bring AI and game design a lot closer together. Which is part of what I am hoping this blog will help us accomplish.
Posted by Paul T at March 14, 2006 09:35 AMgood point mike. I meant to say that most of us focus on believably human *combat* behaviors.
overall though, I think we're making the same point. combat behavior is fairly well evolved at this point, and there's a huge population of gamers showing interest in more domestic human behaviors, and social behaviors. for this reason, we should look at modeling the subtleties that bring characters to *life* which is what bruce's article is about.
We are saying the same thing then. :)
I can't say much beyond that except that I'm pedaling as fast as I can. I'm always interested to find out about others' efforts in this area though.
Posted by Mike Sellers at March 16, 2006 05:32 PMMike, I just noticed your link for Online Alchemy; sounds like you're working on some interesting stuff! The Dynemotion system is interesting but hard to figure out from the details provided -- let me know if you guys will be posting or publishing any more info about this anytime soon (or if you're up for lunch sometime; I'm in Austin as well).
Posted by Paul T at March 16, 2006 07:43 PMThanks Paul. We're not announcing any specifics yet, though I did talk somewhat about our AI at the AAAS meeting in St. Louis recently.
Send me your contact info, or let me know if you're going to be at GDC.
Posted by Mike Sellers at March 17, 2006 10:44 AMwhile our thread is pretty light, this discussion was continued on Terra Nova with some heated debate:
http://terranova.blogs.com/terra_nova/2006/03/failings_of_mec.html#more
The question of whether game AI characters can be *more* believeable than player characters is similar to something I was thinking about the other day. I was reading through this talk:
http://ai.eecs.umich.edu/people/laird/game-seminar/Liden.ppt
and on slides 102 - 110 he touches on things that AIs can do that aren't necessarily realistic but are fun to play against. They're all things that humans might choose to do sometimes, but he argues that we should make our AIs do them, if not all the time, then a lot of the time.
So, this made me wonder: is it theoretically possible to create an AI that's *more* fun to play against than a human?
I totally understand that there's a certain je ne sais quoi about playing against a real live human being, but is there some sort of crossover point where the AI does *so* many more fun things than a human, that it outweighs the je ne sais quoi? Perhaps this crossover point, if it exists, is twice as close if AIs are able to be both more expressive and more fun.
I guess this would be sort of like "holodeck addiction" from Star Trek: a fantasy world where things are so much "better" than the real world that people would rather just exist in the simulation.
What do you guys think? Is the thrill of playing against a live human more powerful than anything that AI can throw against it?
Posted by Euan Forrester at March 28, 2006 03:13 PM> What do you guys think? Is the thrill of
> playing against a live human more powerful
> than anything that AI can throw against it?
I think that with enough time and resources, we're capable of creating AIs that are far more skilled and inventive at gameplay than a human player and far more interesting and responsive in many ways.
At the same time, part of the gaming audience insists on playing with human players, and I can think of two main reasons:
A) They want the thrill of competing against an actual human player and won't ever be satisfied defeating an AI opponent. Part of the thrill for these players comes from knowing it's a human opponent, and a hard-core Halo player won't get any satisfaction from teabagging an AI even if it otherwise acts exactly like a human player.
B) They like to socialize with actual human players, and they want genuine social interaction that an AI can't and shouldn't try to imitate.
There's no question that it's possible to create an AI that's "*more* fun to play against than a human," as you put it. I've experienced that many times in a lot of different games, and I've seen AIs do a lot of things human players couldn't.
There are many different kinds of fun, but at least a handful of them -- particularly the two I listed above -- can't be replicated with AI, and that shouldn't be one of our goals.
Posted by Paul T at April 6, 2006 03:39 PM