« Rape in...yeah, you know... | Main | Organizing the Virtual Organization »

May 08, 2007

Comments

1.

When I think of typical npc behavior, I think of the typical VW npc--little more than one of those old soda-vending machines that can talk.

When I expand my thoughts on it, I think of the npcs in the Sims. Those are a quantum leap above the standard MMO npc--they have goals, wants, and desires. But, at the same time, they will repeatedly, say, buy the same product over and over (naturally that's also a game mechanic, but it's not particularly immersive--how many run-down cars does one pixel-collection need?).

As my thoughts grow, I consider the npcs in Bioware’s games. As far as gameplay, goes, they are generally masterpieces of their time, drawing the player emotionally into the story and fulfilling their function perfectly. But those npcs are also limited in their overall behavior—when not interacting with the player, they do nothing. Again, they ruin the immersion.

I’ve not seen better npcs anywhere than those two examples, and I think it’s fair to say that, commercially, they represent a certain culmination so far in AI development. Or AP (artificial psychology) in this case. They mimic emotions and emotional responses, creating an emotional reaction in the player, yet even with that, they are far from what we’d really like to see in VWs.

I would never go so far as to say research in anything is wasted, but in a VW in which players yearn and beg for immersion and things to do, isn’t using npcs for everything inefficient? Why can’t players receive a notification that they have quests to give out (along with the appropriate information)? Why can’t other players take on the roles of npcs as bandits, say, or even as the guardians of a dungeon or the local police, etc.? Properly handled in the mechanics, no one would ever need to complain about lack of content again.

One of the traps of creating an MMO (specifically, as opposed to a general VW, which may or may not fall into this pattern) is that it attempts to replicate a single-player game experience on a broader stage. This is what’s familiar and so this is what’s done. We’ve all seen this time and time again—the boss npc has to respawn for someone else to come along and be the hero. An MMO is all about everyone having the exact same experience.

Perhaps we should move away from that paradigm, as seductive as it is apparently, and concentrate more on a real VW that has elements of MMOs in it—that is, starting with the idea of a VW, then figuring out how to make it fun and not just a glorified chatroom. Then we can get away from the idea that npcs (whose only real function is to recreate the exact same experience for countless players) are even necessary.

2.

What people think they want and what they really want sometimes do not match terribly well.

Research shows that when NPCs are more believable, the players are more 'attached' to the world. They 'miss' the NPCs more. They get more pleasure out of interacting with the NPC. If you want to create a more addictive/seductive world, it looks like better AI for NPCs is a good way to go.

I asked some WoW players if they could change anything in the game what they would change. They said they wanted bigger swords and faster horses. Well, I hope Blizzard has better plans than bigger swords and faster horses!

3.

NOTE: I'm a "newbie" in this field and most of what I say here is just random thoughts of others combined with my own conclusions. By no means this is to be considered reliable, is just the opinion of one individual lacking in proper scientific reference.

There are many reasons why the presence of artificial individuals wouldn't work on virtual worlds, at least in the current state of the art. One reason is that the inclusion of these agents doesn't guarantee that a virtual culture will emerge, and this is required for the system to work as a whole, or at least it's necessary to the virtual world designer to have certain control over this.

We know culture is determinant on the individual psychology, we could embed this on the artificial individual and expect the whole virtual society to work according. But since this a chaotic system, even a small discrepancy in the starting configuration could lead to completely unexpected results.

Another reason is that most virtual worlds are themed to cultures and societies that we really don't know much about. Fictional races, medieval societies, extraterrestrial worlds, futuristic societies, etc. Even the accurate stories of J.R.R. Tolkien are not enough to study the culture of elves and gnomes, we might try but we'll end filling the blanks with our own culture, and I don't believe this would end up with believable artificial individuals (once they're running on the target context).

Yet another reason is the player character, the avatars or whatever incarnation of the human presence in the virtual world. Most of the times we impersonate a fully grown adult in the virtual world but, if there really is a virtual history and society running on these worlds, we became virtual outcasts from the go. We don't know anything about the history, the culture, the politics of this world, and yet we're fully grown adults wandering this world without even knowing how we got there. Growing and learning could be simulated but would players accept this?

Nevertheless there is one main reason why virtual worlds should have artificial individuals: because it would be too damn cool. The range of possibilities would widen enormously. People will actually care for what the damn npc has to say, rather than scroll to the objectives and rewards. The promise "you can become a hero" would not be a shallow one, because heroes aren't born from fighting monsters, heroes are born when they choose who to fight and what for.

4.

I think the more you get into AI the more you will be tempted (rightly) to get into larger world, economy, ecology and social issues modeling. If NPCs have behaviors based on their environment for example, will the modification of the environment by players (or other NPCs) cause an impact?

Or for example, would a "Bring me 10 merloc feet" quest giver at some point say, "You know, I've got plenty now. I don't need any more thanks." Would a glut of players bringing raw metals to an NPC blacksmith cause an economic ripple in a supply and demand economy? Would NPCs migrate to other locations or towns when there were economic downturns or other threats to their safety in a given location?

It could certainly create a slower MMO with less grind, but on the other hand it could also create frustration if tracking down or "working with" NPCs was hard. Knowledge would become a much more valuable commodity between players since NPCs no longer are so predictable.

5.

I've pretty much been an advocate for believable AI since I realized what that meant. While there are many good reasons not to have good AI (you really have to bypass the Uncanny Valley to have good AI), when you have specific goals in mind, then AI is good.

For example, if your goal is not to build community (AI gets in the way of player-player interaction by making player-NPC interaction overly meaningful), but rather to build immersion, then AI is good.

6.

In terms of breaking immersion the static nature of MMOG's is a bigger problem than rudimentary NPC AI. Consider the typical "Free the Slaves" mission that comes along in every MMOG. It doesn't matter how many you free or how many overseers you beat up, you can be assured that the next time you run past there will be another set of slaves and overseers set up carrying out the same set of rote behaviors over and over again.

It struck me recently what an odd world we live in. Real slavery is a horrible thing. Yet the typical game treats it as entertainment. It trivializes the whole topic.

Finally, I don't think immersion is required for mainstream success. Look at WoW. The "magic circle" or whatever is something that appeals to a minority of the potential player base, and it's not important at all to the majority. The point of MMOG's after all is to interact with other people--what's the point then of lavishing care on NPC AI?

7.

Hmm... I welcome artifical life/pyschology in NPCs, but are less sure if AI is going to be sufficient to make believable adult humanoids. I think designers should shape their world to the "psychological" capabilities of their engine. (Frame NPCs as robots,a liens, animals etc.)

Advanced AI for dynamically organizing and evolving groups of agents (as teams and culture) has a lot of potential IMO.

8.

Would I enjoy a world full of "smart" NPCs?

Honestly, I just don't know.

It seems like a desirable thing to me, but how much of that is just my innate techie lust for coolness regardless of actual utility? Would such a world actually be fun as a game? Or would it be interesting primarily as an AI-explorer's sandbox?

Maybe the question I need to ask is: Assuming we can get NPCs past the "uncanny valley," what kinds of new gameplay could be enabled by NPCs with an order of magnitude (or more) improvement in plausibility of behavior?

To help that conversation along, I thought of a brief list of "intelligent" features I'd like to see added to today's NPCs. In rough order of increasing difficulty:


  • Emotions - lizard-brain reactions to environmental stimuli

  • Perception - awareness of environmental features

  • Communication - ability to transmit state information or knowledge to other NPCs

  • Memory - awareness of previous encounters as positive or negative toward the NPC's plans

  • World-knowledge - understanding of thing-behavior and people-behavior in the world (as in Doug Lenat's Cyc project)

  • Planning - ability to propose actions to achieve low-level, immediate goals and high-level, long-term purpose

  • Self-reasoning - ability to effectively describe internal states and knowledge

  • Do these capabilities suggest fun things that could be done in a gameworld that simply aren't possible with today's NPCs?

    --Bart

    9.

    hey yusuf ~

    sometimes, also, people want what they're familiar with, and the limit of imagination is actually a limit of technology. HeadCase is working on this, and i hope we see something seriously different in six months.

    lilith knows its time.

    ultimately, the trick is to make sure that there is an emotional user interface that allows people to interact with the characters in a meaningful way. it might be narrative, or it might be comercial, but ultimately it has to be emotional.

    10.

    I'll add: "NPCs are the game!".

    When you have a game world with AI/AP, players don't spend all their time killing monsters. They spend their time interacting with NPCs, although some combat may still exist, just as running and jumping (the heart of a more ancient species, the platformer) still exists vestigially in many CRPGs/MMORPGs.

    11.

    I've gone the rounds with Mike over the "artificial psychology" question before, what it amounts to is that he and I are approaching the same problem from different directions. I lean towards "artificial life" and "social simulation" approaches, where the dynamic elements of the world are subject to manipulation at the population level, he's coming from the bottom up with individual NPC's that display more motivation and emergent effects.

    Ultimately, I think his approach is better suited to single-player games than MMO's, but I could be wrong. And most of my objections to their use in MMO's are based on scalability and computational limits, which means they are ultimately temporary.

    --Dave

    12.

    Dave,there is no reason you can't do both. You can have bottom-up emergent behaviour guided by a top-down constraint-like system. Seems to me that you would need a bottom up system in order to deliver a variety of "personalities" (whether that is on the individual level or on the cultural level).

    (Not sure what you mean by artificial life, isn't that usually tied to bottom-up emergent behaviour?)

    Ola.

    13.

    It's a difference of focus, artificial life systems tend to be based on very simple behaviors on the individual agents, and the complexity emerges from their interactions (and social simulation is very similar). From what has been explained to me, the approach Online Alchemy is using has much more subtle and complicated behaviours in the agents. More realistic at the level of individual encounters, but not as scalable.

    I'm still pursuing the concept of the "Big World", really large areas with large numbers of agents (hundreds of thousands). The personality based approach requires far too much CPU to be practical on that scale for quite some time.

    --Dave

    14.

    I see your point, but I also believe that one can do both within the current computational limits. Think in terms of probabilistic reasoning and differentiated sampling densities in different segments of your agent population?

    15.

    Now's our chance to elevate the computer game from a simple passtime into an art. Games are held back now because most games utterly fail to invoke any emotional response. Instead of "ohmygods, i missed and shot one of the kids," we have "ooops, i clicked the wrong button, darn." If the characters seemed more real, we might change that.

    16.

    Would it sell? I suspect no, not to the current audience of Dungeons and Dragons fans. But there's a whole world out there. A world where most don't even buy video games. Yet.

    17.

    At the heart of any question about making a feature better (or dumping it) is they question of why we have it at all.

    Any/all AI in any game is, to a great extent, going to drill down to a root goal of some kind of puzzle solving. I want to beat the game. My wife plays math/word games in books. The intelligence in those has been programmed by the writers, and she is pitting her skills against them. Just like an NPC who re-rezzes again, if a guest in our house enjoyed a couple of those puzzles, they could go out and buy the same book in a clean-state. Solving puzzles is fun, and figuring out the best (easiest, flashiest) way to beat the crap out of a huge Iron Troll is, in many ways, a type of puzzle.

    Figuring out how to beat the crap out of another player, even one whose character has the same attributes as the Iron Troll, is less about puzzle solving and more about monkey pee.

    By which, of course, I mean status, counting coup, showing off, feeling good about yourself in the presence of others/peers, whaling on your kid brother, etc. etc.

    Even if the experience were almost the same, my reasons for engaging in PvP encounters (of any kind, not just combat) vs. PvE are going to be vastly different. Let's take economics, for example. In games where you can buy crap from a merchant NPC, you really don't care about the future state of your relationship with that character, except as it will affect future prices. Right? Some games have built in a good/bad reputation system, or a "if you steal from this merchant, he won't sell to you," and that's fine. But in PvP economics, my trade with you, Mr. Player, will affect other aspects of our relationship. If I stiff you or give you a good bargain, that's going to have consequences outside that particular set of behaviors.

    So, to me, the question boils down to one of intent: which activities do you want to be handled purely (or mostly) as puzzles, and which should be social and/or experiential ones?

    The dissonance between these explains, in many cases, the trouble over RMT. Some players see the combined experiences of the entire game as experiential, and thus to be explored in social ways, or, at least as entertainment/art. Others see various sections of the same game as the first 30 pages of a 100 page Suduko book, and "Why the hell should I do the first 30 pages, if I'm already good at intermediate level puzzles?" They want to skip to the part they have interest in. For them, the "puzzle" aspects of the quests, leveling, farming, grinding at the earlier levels are not as interesting. So they "solve" those by a quicker route; RMT -- they flip past the early parts of the puzzle book.

    WoW was designed so that people could play much of the early game either in groups or alone; and that's what causes this tension, I think. It makes the game flexible, and that's cool... but it also proposes a challenge. When the exact same puzzle can be solved by either one high-level character, or five lower level characters, you have two very different types of game play at work in the exact same space. You have one that rewards social play (the group mode), and one that rewards puzzle work (single player).

    Again... all would be fine if these types of players left each other alone. But we know this isn't the case. And folks playing the puzzle game (leveling, either by solo work or through RMT) then get into the spaces of the social players.

    I say this simply to point out that two types of play, and two methods of enjoying the same platform, can result in both positives and negatives. It's not just about AI... it's about game design. Duh.

    If quests, beyond a certain level, had to be given by characters, and involved PvP encounters, then higher level quests would require either fairly even PvP encounters (equal levels), or a balanced group doing the will of or going up against a higher level player. In this case, there would be much less tension between puzzle-based and social-based activities, because the puzzles would be created in a social setting: e.g., the character giving you the puzzle is a player, and there are social consequences of repeatedly screwing up his stupid request for 100 bats eyes -- he'll stop giving you quests and send you on your merry way.

    18.

    What is the ultimate reason one wishes to add more humanlike NPCs? To make the world more accurate--to increase it's realism and make the players'/visitors' reactions deeper?

    If that is the case, then what we really are looking for are two things--of which character (that is, NPC AP/AI) is one. The other is story.
    A wonderfully emotive NPC with great graphics and that responds well to each human individually, but that has a woeful backstory/quest, etc. to tell is not going to be very fulfilling.

    It would be a grave mistake to pump up the tech level of a simulation without keeping pace with the artistic side (in this case, the writing). Otherwise, you've gone from a talking soda-vending machine to a talking, emotional mannequin (aka a fashion model?). A step up, maybe, but not something that's necessarily very involving to a human looking for good gameplay.

    19.
    Now's our chance to elevate the computer game from a simple passtime into an art. Games are held back now because most games utterly fail to invoke any emotional response. Instead of "ohmygods, i missed and shot one of the kids," we have "ooops, i clicked the wrong button, darn." If the characters seemed more real, we might change that.

    Why would AI invoke an emotional response? After all, what good is AI if its used to model an insurance salesman with whom you have thrilling discussions about escrow and mortgage payments?

    What evokes an emotional response is story. And it's almost impossible to get a story into an MMOG. Single player games on the other hand are entering a golden age right now, especially the shooters. I think Half-Life 2, for example, features writing and acting that are on par with anything coming out of mainstream Hollywood.

    20.

    @Andy: I think it's interesting that your discussion of AI and games doesn't touch on the role of narrative and story at all. You talk entirely in terms of puzzle solving, likening playing an MMO to doing Sudoku. This is probably an appropriate metaphor for most MMOs. But what if an MMO could be more like a novel? When you read a book, you're investing time and emotion into caring about characters, their interactions, and their story. What if we could have NPCs that were as engaging as the characters in a book, but unlike a book, you can actually play as a character as well? Is that workable at a massively multiplayer level, conceptually if not quite yet technically? To me, that's what the idea of artificial psychology suggests...the ability to create characters that are inherently engaging and relatable, not more puzzles.

    21.

    Not just AI in characters, but AI in the world state engine. Feedback if you "push" here (raid a stronghold that is a large source or consumer of goods, for example) causes a "twitch" here (prices of particular raw materials goes down on world NPC vendors).

    Predetor/Prey relationships. If there are no predators, the population of its prey can blossom causing all kinds of issues for players (wow, no more lions, and now the deer have eaten all the bushes!)

    The orcs and goblins have been plotting against each other and want to take a strategic plot of land/artifact/pizza recipe and now that the orcs have been hit hard enough to be crippled, the goblins are starting to get uppity! Stomp the goblins and return the orc/goblin conflict to some sort of normalcy and predictability - or keep stomping orcs 'cause you're rooting for Team Goblin.

    NPCs/Important monsters knowing who their allies are and sending pleas for help via carrier pidgeon, or magic scroll delivery service when it becomes obvious (based on progression through their dungeon/castle/island) that a raid force is moving against them.

    The idea of giving your avatar an "offline" life is interesting. You could have a set of sliders that you can set describing how likely your character is to be found in different venues in a city. Tendency towards avoiding large gatherings, etc. Then the server engine could throw a time-line for that character at each dawning and trigger the movement from place to place during the day based on those likelihoods.

    Combine that with a set of personality sliders (irreverence, piety, violence, lawlessness) and that would help define their activities in those different areas.

    Imagine logging in to find yourself being released from the local jail 'cause your character got drunk in the bar and then went and caused a ruckus in the city square? No penalty (besides travel time from the hoosegow) to the player, but they were adding "flavor" to the environment while you were gone.

    22.

    Whether it'll make a particular game better right now or not, virtual worlds are an absolute playground for AI development and training. They're full of real human speech, embodied avatars in lifelike graphical environments, and spatially contextualized social interaction in large but finite (or at least fully quantifiable) possibility spaces where patterns keep welling up like ocean waves. They're gigantic data dumps of the way we behave and the things we say and how we respond when blank happens. And they resemble the real world more than the web does.

    If I were an AI baby bot looking to learn today, I'd ask mom to drop me off in a virtual world on her way to the ol' semantic web factory. The road to AI runs directly through virtual worlds. Don't relent, Mike.

    I remember when I first showed Second Life to my roommate Christian who now works on the stuff, he was coming out of a robotics and AI background and was like, "Oh, so it's an entire world of programmable robots, and I don't have to lug a piece of metal outside. Sweet!" Virtual worlds are (well, can be, you do have to move towards more open, scriptable environments if you want to see the AI ecosystem of experiments take off) absolute accelerants in this area. I was listening to this Where 2.0 podcast recently about the MIT Reality Mining project where they're tracking the life patterns of subjects using their cell phones and analyzing their patterns to train AI to predict human behavior, and it's another example of the kind of thing that could be done so much more efficiently in a synthetic world. Again, less metal to lug outside and even more data about what people do.

    And the applications will apply outside of virtual worlds too, so, you know, bonus.

    23.

    About choosing between Artificial Society and Artificial Psychology, perhaps an analogy could be drawn to landscape rendering distance: the greater the psychological distance, the less detail an NPC's rules would need for versimilitude.

    When no one's looking, a village could revert entirely to following simple rules. Given the nature of complexity, I think that it would be difficult for someone coming along later to tell the difference between complexity that emerged from simple rules and complexity that emerged from sophisticated rules. As someone approaches, the villagers could manifest increasingly individual behavior based on how close someone would have to be to tell the difference. Then, maximum Artificial Psychology would only be reached by a particular character when actually conversing with a player.

    As a sometime player of MMORPG's, I find this idea stimulating. I imagine that with psychological rendering distance, a world could be allowed to generate an emergent narrative without the "tree falling in a forest" problem. Every plot-affecting decision by an NPC would take place near a player, or more likely while talking to one, since they wouldn't have enough individual intelligence to make those decisions otherwise.

    Also, it would change the nature of a character's record. Right now, I would say that there are two kinds of stories in the games that I play.

    One kind of story is the meta narrative that players tell each other about levels gained and play sessions enjoyed. The other is the quantitative record that (for example) the Station Players character sheets show of mobs killed and discoveries made.

    With an emergent narrative arising from artificial psychology, the game engine would have the potential to automatically track a player character's participation in that narrative, like a free-form quest journal. I don't know how useful or expensive that would be; I just think that it's an interesting idea.

    24.

    @Karen: As a writer, I'm hugely interested in story. And when I GM a pen-and-paper game, I am also hugely interested in my role as creator/narrator (in one sense) and scene-setter in another. I definitely want my players to feel as if they are participants in a story, and moving the narrative along in (hopefully) ways I never envisioned.

    Problem is... even NPCs in pen-and-paper games are played by a live person; a storyteller. I have been interested in participating in the narrative of all kinds of games, sure. And NPCs play as much a part in those as any other aspect of the game. I don't ever feel, though, in their presence and in interactions with them that I am *creating* anything like a story; I am simply following it. Which may be puzzle-like, at points, or simply enjoying the plot as it unfolds.

    The only times I have felt a true part of any creative storytelling, is when I have an audience, even if it be of only one other person engaged in the most limited RP behavior. What is "story" without a listener?

    So, I think, my answer to your question is that, sure... better AI, however it manifests, will improve a story from my perspective as an audience member (better is better no matter what game mechanic or feature we're discussing). But truly interesting stories require that I be able to change something in some way that affects another participant... that results in an audience reaction of some point. And, frankly, I'm not sure I'd ever get to the point where I really care -- from a storytelling perspective -- what an NPC "thinks" about my actions. From a puzzle-solving perspective, sure... I want to get past the troll. But will I stop and have a chat with him around the campfire just to increase the RP aspect of the moment for myself? Nope.

    In the absence of other humans, "in character" and "storytelling" behaviors are, I think, somewhat null. .. unless they advance the game. In which case, thereyou go solving puzzles again ;-)

    I stay in character around other players on an RP server not because it helps me level or find better loot or finish quests, but because I am playing a "meta game" wherein I appreciate good RP/storytelling behavior, and hope that others appreciate it in me. I have always felt that these things bring orders-of-magnitude more pleasure (and learning) to games.

    I'm not sure, though, how one would roleplay with a golem.

    25.

    "Why would AI invoke an emotional response?" Because it could make artificial characters seem real enough so that we care what happens to them.

    "What evokes an emotional response is story" The purpose of story is to reveal character.


    26.
    "Why would AI invoke an emotional response?" Because it could make artificial characters seem real enough so that we care what happens to them.

    Why? I meet plenty of real life people every day--the guy behind the counter at the sandwich shop, the guy behind the counter at the gas station, etc. etc. I don't really have any real emotional attachment to them. Why would I care about the guy behind the counter at the tavern who checks me in for the night?

    Another question: how much AI does a character in a book or a movie display? I'd say it's pretty much zero, and yet the emotional reaction I have to some books and movies is much, much greater than any game I've ever played.

    I think that MMOG's face some real obstacles in terms of establishing a narrative, (the static nature of game worlds, the accompanying inability of players to actually change anything, the wide disparity in player motivations across a population, etc.), especially compared to their single player cousins. I don't see AI as really addressing any of those issues.

    27.

    Why? I meet plenty of real life people every day--the guy behind the counter at the sandwich shop, the guy behind the counter at the gas station, etc. etc. I don't really have any real emotional attachment to them. Why would I care about the guy behind the counter at the tavern who checks me in for the night?"

    There must be someone somewhere in the world that you care about, i hope.

    28.

    Another question: how much AI does a character in a book or a movie display?

    I noticed that Luke Skywalker could fly and fence and shoot, even hold conversations. He even appeared to have an occasional thought or feeling.

    29.

    I think that MMOG's face some real obstacles in terms of establishing a narrative.

    Yes, that's why i want to go beyond the static, menu-driven worlds where everytime i login, the only thing different from my last login is my character.

    30.

    lewy said, I think that MMOG's face some real obstacles in terms of establishing a narrative, (the static nature of game worlds, the accompanying inability of players to actually change anything, the wide disparity in player motivations across a population, etc.), especially compared to their single player cousins. I don't see AI as really addressing any of those issues.

    While I agree with you on the obstacles to establishing a narrative, I have to say that I absolutely believe that better AI/AP addresses at least the first two -- the static nature of the world, and the inability of players to affect that world. I think the element of Artificial Psychology that you're missing here is that it allows NPCs to have both short-term and long-term memory, and emotional associations with the memories and the actors in those memories.

    For instance, take the standard MMORPG quest "Free X from the slavers," that was mentioned here on TN a few days ago. In MMOGs as they stand now, the player goes to the slavers' camp, fights the slavers, and frees the captive. The captive thanks the player profusely: "Thank you so much, mighty paladin! You saved my life! Take this as a symbol of my gratitude!" The captive runs off, and the player, now rewarded, goes on with the next pre-scripted quest. If the player were to hang around the slavers' camp, he would see all the slain slavers respawn, and the captive mysteriously reappear inside his cage, ready for the next player to come along and "save" him. If the player were to again kill the slavers and free the captive, the captive would have no memory of having been saved before.

    Now, consider a similar scenario utilizing NPCs with AP. Two groups of NPCs -- lets call them the Caps and the Gues -- have been fighting each other for several months, perhaps over a land dispute, or business competition, or an insult. As the conflict escalates, the Gues kidnaps one of the Caps. None of this is scripted by the designer. All of the NPCs involved have long-term memory, so they each remember what started the dispute, and AP allows them choose their own goals, based on their needs and desires.

    The Caps and the Gues could continue their feud without any outside assistance, but since this is a MMOG and not a closed AI simulation, let's consider how the actions of one player could affect this situation. The player at hand has an established relationship with the Caps, having done business with them before. The Caps think well of the player, and so when the player stops by the Caps' stronghold to do some business, they tell him about the kidnapping. If the player chooses to help the Caps, they will like him even more, and perhaps help him the business problem he's encountered. (The player is also free to choose not to help the Caps -- this is not a quest in the classic MMORPG sense -- but choosing not to help will also have an affect on the Caps' opinion of the player.)

    The player chooses to go to the Gues' stronghold, and through persuasion or force manages to free the kidnapped Cap. The Cap thanks the player, and returns home. If the player were to visit the Caps' stronghold again, he would find the NPC he rescued there. That NPC would have a memory of the kidnapping, and of the player's actions, and would respond to the player based on that relationship. If another player were to go to the Gues' stronghold, he would not find the kidnapped Cap there.

    The game world has evolved, and the player has had a direct hand in that evolution. That evolution would not be possible, on the massively multiplayer scale, without AP.

    As to your third point, the disparity in player motivations across the population of the game, I think NPCs having individual emotional opinions of each player will change how this manifests in the game world, but we may need further innovation and/or different design methodologies to completely address it.

    31.

    Lots of really good discussion here! A few comments:

    Bart, I largely agree with your list of emotions, perception, communication, memory, etc. I believe these make for a stark change in the kinds of worlds we can create -- much more worldy, and fewer set pieces with vending machines standing around in various places. As Mike Rozak points out, this opens the door to psycho-social aspects of gameplay we just don't see in current MMOGs.

    Andy Havens: Any/all AI in any game is, to a great extent, going to drill down to a root goal of some kind of puzzle solving. [Later] In the absence of other humans, "in character" and "storytelling" behaviors are, I think, somewhat null. .. unless they advance the game.

    Yours is an "achiever" POV that's prevalent in current gamers (and also correlates with male gamers). You might as well have said that all gameplay eventually devolves to shooting something. For many people, once you can be sociable with NPCs, and your sociability matters to your character in the world, solving puzzles becomes at best a means to an end, not the end itself -- and not the only means. As Karen Tanenbaum says, artificial psychology enables much more relational NPCs and gameplay that goes with them, not just more of the same kinds of play we have now.

    You raise the question of reputation with NPCs, which is a key part of any social landscape. Enabling this beyond the "+5 with this faction" level adds a lot to the social aspects of any game. This is, IMO, one of the key benefits of intelligent social NPCs.

    Lewy asks: Why would AI invoke an emotional response?

    There's good research that shows that the more believable an NPC is, including its emotional responses, the more emotional resonance people feel with it -- they attend to it more, learn from it faster, etc. (no reference handy). Of course if you create an AI with poor emotional responses, you'll weaken the emotional connection, but I see no reason to do that. :)

    Others have mentioned the Uncanny Valley, which is a concern. But it's not clear where that Valley lies nor how important or difficult it is to bridge it when talking about behavior rather than animation. Early indications are that we're much more likely to read human-ness into even vaguely human behavior in ways that we don't with form or animation.

    Story is an important element in emotional connection too, but I believe what we consider to be story changes dramatically (no pun intended) in the presence of many actors. Consider the experience of 'story' in a LARP vs. a single-player RPG. The latter is more classically story-driven, but the former is much more experiential and feels like a story experience, even if there's no classical three-act structure.

    I believe intelligent (AP) NPCs are going to enable new kinds of story -- emotionally and socially relevant -- that are otherwise impossible.

    32.

    As I said at the top, MMOs currently specialize in creating the exact same experience for each person. Better AI and different expectations could drastically increase "realism" and fun. Probably all of us here have the shared background of a high-school experience, but how many of those were exactly the same? None.
    But in a game we expect each of our experiences with any given environmental feature to be the same as someone else's. A more expansive AI/AP could alter this expectation to be one of "you got apples, I got oranges--we both got fruit and that's all that matters." Instead of the "we both have to get apples or else" expectation.

    Likewise, such an advanced AI/AP could allow for better storytelling because, finally, the story would be told through each individual player-character's interactions with the environment and with NPCs, allowing for each player to have their own story instead of repeating everyone else's.

    33.

    Regarding your last comment, Lucas: that's exactly my point (see my post on 'story' in Richard's "First Principles" item).

    34.

    In the world, there are smaller and larger pools, fractally, and smaller and larger quests, fractally, -- only one POTUS at a time, but plenty of CEOs and world experts and Don Juans -- so that many, many people can find a quest level that suits them, a pool in which they can distinguish themselves – perhaps in one among many pools in which they play.

    And the pools themselves differ by the types and intensities of the challenges and rewards they offer -- while the intensities of the challenges of any type tend to correlate with the extent of the rewards -- so that humans can "level up" in a manner that Maslow among others has detailed.

    And then there are simulacra of some of the more challenging / rewarding pools at various levels which are made available for vicarious quests -- the world series, the soaps, the televised papal Mass.

    *

    How much of all this is transferable to MMOGs with suitable AP NPCs, Mike?

    35.

    Deep questions as always, Charles.

    I'm a big fan of wheels-within-wheels, arcs within arcs in storytelling, and intuitively I think there's a lot to what you're pointing to in a multi-viewpoint storied environment populated by both people and social, goal-oriented NPCs.

    There are a lot of questions and hazards too: if you're playing in a deep pool, say for the election to the presidency, in the real world you don't log off in frustration when you lose; but in a virtual world are there enough ties to keep you coming back even after a large, public winner-take-all loss? I don't know.

    And what are the relative costs of providing "content" that ultimately only a few people access? Typically this sort of thing is anathema in MMOGs; you need to design content that as many people as possible can use and re-use. But if that cost becomes flatter due to better social design, maybe this sort of fractal gameplay becomes tenable.

    Difficult to say; what you're asking is still over the horizon. But in a good direction.

    36.
    The player chooses to go to the Gues' stronghold, and through persuasion or force manages to free the kidnapped Cap. The Cap thanks the player, and returns home. If the player were to visit the Caps' stronghold again, he would find the NPC he rescued there. That NPC would have a memory of the kidnapping, and of the player's actions, and would respond to the player based on that relationship. If another player were to go to the Gues' stronghold, he would not find the kidnapped Cap there.

    Sure, but there the salient issue is a dynamic game world. From the player's perspective, would it make any difference if that cap/gue conflict was scripted by a developer or generated dynamically via an AI algorithm?

    37.

    From the player's perspective, it's an issue of scale. If we were talking about a single player game with a small-ish game world and a definite ending (Knights of the Old Republic, for example), every encounter like this could be hand-scripted by a team of designers. But for a virtual world, with hundreds of thousands of players (or more), a large game world, and no pre-determined ending, hand scripting enough non-repeating encounters to satisfy the player base for more than a few hours of game time is simply not feasible. So we are left with two options: NPC encounters that are hand-scripted but which reset/respawn, so that thousands of people can experience the same encounter; or AI/AP that dynamically creates encounters with little or no input from a designer.

    38.

    Another thing to consider is the introduction of evolutionary systems in these environments. The static, nondeterministic qualities of current NPC designs are antiquated, and quite frankly, boring.

    Using evolutionary containers, and probability inference, we can create NPC's that evolve in real-time based on the available environment variables, and survivability of the organism (virtual).

    We, the players, would then become another stimulus, or influence to the life of the virtual organism. We could alter their food source, disrupt reproduction, or introduce new learnable concepts, such as farming, mining, etc. We could even teach them to hunt, and kill.

    A post above mentions spawning and having enough NPC's for the community to interact with. Respawning NPC's the engine of the level treadmill. In an evolutionary world, NPC need to reproduce, and thus mutate. Their reproduction rates could be altered of course, but I kinda the idea that if you kill in this world, you really destroy a life, a virtual life, but a life none the less. Its going to be a great time when we get good AI in games.

    39.
    From the player's perspective, it's an issue of scale.

    Actually, I would say that from the developer's standpoint it's an issue of scale. They're the ones whose fingers are going to be saved from having to script thousands upon thousands of quests. (Although that problem should be ameliorated to some extent by the fact that player bases are splintered across numerous servers that really only serve a few thousand people). The point is that the player shouldn't be able to figure out if the quest they're on was generated by an algorithm or a human being, if the algorithm is sophisticated enough.

    40.

    The resources spent on better NPCs are mostly wasted. We would be better off channeling human intelligence in interesting ways, rather than trying so hard to make NPCs that, at best, fool players for a little while before the illusion wears. Usually the illusion is broken due to some weird non-human behavior. Once the illusion is broken, the game's replayability suffers enormously (assuming it relies on NPCs for its playability).

    I feel that NPCs are like the 3.5 floppy drive -- they are included in these worlds because that has been the standard thing to do. But why are we trying so hard to make human-like AI when we have so many humans available to fill those roles? Games like EVE Online are pushing in the right direction by having players fill many of the roles that are played by poor AI in other virtual worlds.

    What is the goal with NPCs? It seems that those who develop them want NPCs to eventually reach human intelligence. They want them to be exactly as intelligent as humans, so that they can fool them in all situations. If that is the goal, then it seems like a poor use of resources considering the fact that we have so many intelligent, Turing test-passing humans available to make the virtual world dynamic and interesting.

    In my opinion, many NPC-reliant games are essentially primitive designs with 21st century graphics technology. They keep the player's attention only so long as he does not see an NPC get stuck behind a bar stool or walk daintily off a cliff. Even virtual worlds such as WoW can be seen as being more or less co-op singleplayer games that, in their design, have not embraced the dynamic human-human interactions that make virtual worlds interesting.

    41.

    "So we are left with two options: NPC encounters that are hand-scripted but which reset/respawn, so that thousands of people can experience the same encounter; or AI/AP that dynamically creates encounters with little or no input from a designer."

    Or, we can design games built around human interactions, encounters and experiences.

    This reminds me of the early days of the Web, when I had to battle with print designers used to absolute control over every aspect of the visual product - placement, color, font, etc., etc.

    Successful design on the Web involved a certain surrender of control and an embrace of a partnership with the "end-user" (a quaint and inappropriate phrase for a participatory medium).

    More accurately, it required surrender of certain things that turns out not to be essential to the experience, and a new focus on ways to direct, encourage, facilitate and afford, rather than dictate.

    Today, the best online design is as much social architecture as visual art.

    MMOs are still being designed largely as if the other human players were an annoyance, rather than a feature to take advantage of, and designers still obsess over providing the same, authoritarian experience to each player - the one the designer wants players to have, as opposed to the ones players can enjoy in mutual creation. Who says everyone must have the same experience to have a satisfactory one?

    AL is useful to populate worlds with clearly non-human activity and the appearance of life, movement, and variety, from animals to robots and the continuum inbetween.

    Spending resources on imitating people when MMOs are full of the real thing is futile and not cost-effective, IMO. And I say that with both the greatest of affection for Mike and a great personal and professional interest in his work and his company's products.

    There are plenty of places where next-gen AI comes in handy, from single-player games to education to training to research simulation. I just don't think an MMO is one of those places.

    It would be like trying to populate Wikipedia with intelligent agents rather than utilizing the wisdom of crowds. What's the point?

    42.

    Samantha said: So we are left with two options: NPC encounters that are hand-scripted but which reset/respawn, so that thousands of people can experience the same encounter; or AI/AP that dynamically creates encounters with little or no input from a designer.

    Galiel replied: Or, we can design games built around human interactions, encounters and experiences. ...

    Inherent in Samatha's view quoted above is not a false dichotomy of "NPCs or humans!" On the contrary, we've found that we can enable gameplay built around human social experiences only via the use of NPCs embodying a certain amount of AP. They form a scaffolding for a level of human interactions, stories, and community that is difficult or perhaps impossible to engender (note: not design, create, or force - ultimately this has to come from the players, using what the world enables) otherwise.

    It may seem paradoxical, but from the very start, the reason we focused on whether we could create believable NPCs in an MMO was so that we could foster more human social and community interactions; these are, after all, the lifeblood of any virtual world (game or not).

    MMOs are still being designed largely as if the other human players were an annoyance, rather than a feature to take advantage of ...

    I think you know this, but to make the point, humans aren't a feature either. They are an integral but uncontrolled, unpredictable part of the game/world. To provide them with (or enable for them) enjoyable experiences that go beyond the repetitive "kill monster, get gold," chat and cybersex in the social realm, and mentally empty button-pushing in things like "crafting," we need different underlying structures from those found in MMOs today. One of the key structures is NPCs that are more believable than cardboard-thin vending machines.

    I know I haven't said much about exactly what the benefit of these NPCs is, how this scaffolding adds to the social gameplay landscape; I hope you'll forgive me if I don't say too much for now.

    ... designers still obsess over providing the same, authoritarian experience to each player - the one the designer wants players to have, as opposed to the ones players can enjoy in mutual creation. Who says everyone must have the same experience to have a satisfactory one?

    Everyone having pretty much the same gameplay experience makes for a more balanceable, predictable game. It means you're sure no one type of player experience (e.g., "paladins" over "rogues") is much better, easier, etc., than another. It also means that you're much less likely to have players become dissatisfied and start spoiling the world you've put together.

    I agree that removing the authoritarian desire in designers to say "this" is the player's experience" is where we need to go; MMOs are already on that road. But doing so in a way that fosters greater sociability, deeper attachment to the world, and more human experiences requires, perhaps paradoxically, the use of non-humans --believable NPCs -- in certain areas.

    Could you create a strongly social, community-oriented world without these? Probably, at least for a time. Some, a relative few, would find the experience intensely enjoyable. But I believe that in a world without NPCs, the breadth of those who find it enjoyable, the depth of connection with the world, and longevity of the connection which the player's choose to keep will be strongly curtailed (as seen in the current crop of social-network sites and early primarily social worlds, as well as in the few MMOs without NPCs).

    43.

    What makes MMOs different than single-player games? I mean, at the very core, if not the presence of other participants to interact with?

    Why are NPCs not considered essential to other forms of non-computer-mediatedm large-scale social entertainment and games?

    What justifies the assumption that the only way groups of humans can tell and share stories (or do it in a sustainable way) is with a bot around?

    Why do you assume that "balanced, more predictable" are desirable attributes in gameplay - they may be from a control standpoint, but are they, in fact, what makes for an enjoyable experience from a player's point of view in a social environment?

    What is the justification for claiming that the way out of fedex-quest and kill-and-loot grinds are better NPCs with better AP?

    In a baseball game, nearly every player has a different experience over the course of nine innings; nearly every member of the audience a different perspective. The fact that only one or two may catch a fly ball, while one or two others get a drunken fan's beer spilled on them, doesn't seem to harm the overall experience - nor make MLB an unprofitable or unsustainable form of entertainment. It might be raining, it might be hot. It might be a pitcher's duel, it might be a blowout. Fans reading about it after the fact have another experience altogether. Rotisserie league folks have yet another. And so on. Yet all of these may be enjoyable, despite being widely unbalanced, completely uncontrollable in most cases and certainly not predictable.

    The assumptions inherent in most designer's approach to MMOs reminds me of the early days of television, when a fixed camera filmed radio announcers doing their shtick. The "better NPC" conversation is like talking about improving the microphone.

    That's not to say that we should throw out the baby with the bathwater. Some TV shows have a voice over narrator, adopted from the radio days. But that doesn't mean that every television show is developed with the assumption that a narrator is necessary.

    I'd like to hear some of the justifications for claims that people can only play together in a virtual world with the help of NPCs.

    How is that any difference than saying that Wikipedia needs better intelligent agents, because people just aren't it?

    I don't suggest I have all or even many answers, but at least I ask the questions; I see all too many unconscious assumptions holding this medium back, causing us to make more of the same only with different bitmaps.

    44.

    I'm going to have to beg off answering some of your questions above that go directly to what we're doing and why. I can say that, like others, I've spent a lot of time questioning basic MMO assumptions (do players need to be represented by a character in the world? it turns out there are really interesting worlds you can create if you say "no"), and have come to these conclusions. Others, obviously, might come to different ones.

    You asked: Why do you assume that "balanced, more predictable" are desirable attributes in gameplay - they may be from a control standpoint, but are they, in fact, what makes for an enjoyable experience from a player's point of view in a social environment?

    This is one that a lot of people have looked at and tried different solutions for. Players desire gameplay that is balanced, because otherwise they become sure that someone else is "getting more" (of whatever) than they are. They'll complain and if you don't listen, will do whatever they can to ruin the world (no, they don't just leave, which would be better all around). Balanced gameplay is easiest to create if it's also predictable, as then it's also testable. There's also a huge portion of players who prefer a predictable experience, as then they know what they're going to get out of it.

    However, players also like some degree of unpredictability if it is meaningful from their POV and remains within expectable, explainable parameters. Their experience need not be the same as everyone else's nor even "balanced" at a first order if it is sufficiently meaningful and enjoyable.

    There may well be multiple ways to get to that sort of gameplay; I'm sure time will tell. We've chosen to look at this from the POV of building more meaningful social relationships and how this is enabled, supported, or thwarted by various in-world elements -- including the NPCs (if any).

    To use your analogy, I'd say that what we're not looking at making a better microphone, but that we're attempting something more along the lines of multi-viewpoint ensemble storytelling rather than linear one-camera one-plot-arc stories. Will it make difference? We'll see.

    45.

    Mike, I appreciate the constraints you are under in terms of discussing specifics - I was rather hoping, as I know you were, to broaden this discussion to a wider one involving the deeper question that you posed at the end of your blog post.

    As for gameplay assumptions, I find it fun, and useful, to question even the most basic ones, and make sure we bring our unconscious, most fundamental ones to the forefront, so we can examine them and learn new things.

    For example, concerns about someone getting more than someone else are only an issue in an environment of scarcity coupled with an assumption of utility.

    People rarely complain about not having as many colors as the next player, or not having as many letters to type with. Nonetheless, we can have a lot of fun playing with virtual graffiti, or having multi-threaded conversations without end. Certainly political wonks and baseball fans have never run out of colorful words to use.

    MMOs have lots of human brains playing; to what extent are we using that free resource to enrich our game designs?

    If you are going to create smarter NPCs, I suggest (half-jokingly) employing them to fulfill the damn fedex quests their dumber NPC buddies assign; and give me some clever NPC newbies to slay the damn NPC bunnies and rats and loot their vanishing corpses - and savvy NPC buyers to deal with the annoying NPC vendors. Then, perhaps, game designers will start to think outside of their RPG boxes, and there will be some money left in the research budget to hire a group psychologist, an anthropologist, an ethicist and a political organizer (not to mention a magician, a jazz musician, a stand-up comedian and a poet).

    It's as if cruise lines were focused on creating smart robots to populate their cruises, emulating passengers, so that each ship would have a small number of humans whose experiences could be controlled, balanced and equivalent by interaction with those robots. What's the point? It all seems to me to stem from a control fantasy that plagues the MMO design community.

    MMO environments offer an opportunity to tap into the wisdom of crowds. We've learned that in those kind of situations, careful design, involving a great deal of shared power, softsecurity and a gentle touch, can make all the difference between a Wikipedia and a mob.

    I don't see many developers thinking deeply about that kind of surrender, and how to make it work constructively and sustainably.

    Incidentally, ARGs are a place to look for useful lessons in MMO design. So is Wikipedia.

    46.

    Good posts, Galiel. I believe tapping into 'the wisdom of crowds' is the future of MMOGs.

    47.

    Great discussion and sorry I'm coming to it so late. Because my vw work is not so much in the mmo world anymore, it's interesting to see the AI question framed exclusively in that sector, where it can assume near religious levels of passion.

    Alot (but not all) of our work is in what could be called generating finite simulations within persistent worlds. We generally create large worlds within which users have specific purposes for relatively finite periods of time. There's lots of measurement, assessment, re-running, etc. And in these areas, there's simply no doubt that AI at various levels is necessary. Crowd behavior is a classic, low-hanging-fruit kind of case. Interactions for cultural familiarization, doctor-patient scenarios, etc. all gain scalability when you can decide who is wetware and who is not.

    Most of the conversation also deals with NPC's as objects of interaction, while I'm also thinking that they may have at least as much value as extensions of self...returning back to agent-based presence. As online worlds increase their interactions with other systems that becomes an interesting leverage point.

    I am leaving aside issues of practical implementation (although we're actively doing that) -- the challenges of memory/alignment, and the overarching problem of interaction in a voice-dominated environment...there are many serious challenges. But I long ago answered Mike's question in my own mind with a definitive "yes" -- necessary. At least in the worlds I work in.

    48.

    Great to see you posting here, Robert. You're right about crowd behaviors and agent-based presence; these are areas that are ripe for virtual worlds (including MMOs) but which have been all but ignored thus far. All it's going to take is one solid example of how these revitalize the virtual worlds' experience and everyone will be slapping their foreheads with a big "d'oh!" :-)

    Oh, and congrats on the release of OLIVE.

    The comments to this entry are closed.