#adaptive mapping

The last part of Will Wright's "Will Wright's Design Plunder" presentation at GDC 2001, called "Story Recognition," provides a picture of how what has been given the name "System of generating personalized player’s game reality in The Sims 2" should work. It binds together all the disparate pieces of available information and observations from players.

Here's a record of the presentation:

The presentation of this system begins at minute 58 and goes all the way to the end of the presentation.

This is what Will Wright says:

Now, the last thing I want to talk about here, real briefly, is interactive storytelling.

I've always been really happy in my little sandbox, you know, doing these open-ended games, these simulations. And I've always really avoided the idea of interactive story. But at this point, as you can see, our fans are going to basically drag me into it whether I like it or not. So I've been thinking about this quite a bit lately, how we might enable this.

What's happening right now is the simulation is going off and doing one thing, and the fans are saying, "Oh, to hell with that, I'm going to tell my story over this direction." And frequently, they're having to take these Sims that just want to, you know, pee and eat and sit in front of the TV, and they're grabbing them, "No, stand over here, I need to take this shot." And so they're like these little actors on strike, you know, that you're trying to manipulate to take the screenshots.

So, what we need to do is try to get these things more to converge, bring them back into line. We've got basically this open-ended space, you know, you go in a direction in, but it's very flat dramatically. Now, drama, you know, is linear but has this added dimension that we don't have. And so, how do we get this added dimension of drama in our open-ended space?

I think the path to this, I see, is not teaching the computer to tell a story, but rather teaching it to recognize a story. If, when you're playing these stories out in the Sims, the computer can actually get some sense of the story that's in your head, it could then start helping you a bit.

It might, at first, be just a very simple thematic recognition. It might look at the things you've bought in the Sims and the interactions you've chosen, "Oh, you tend to kiss a lot, and you buy this type of stuff, I think you're doing sort of a romantic story." Or, you buy, you know, heart heads and jars, or shackles in your dungeon, and it might say, "Oh, that looks kind of like horror." And then it looks at what you're doing, it might even drive events to resolve it later.

You know, if it's not quite sure if you're doing a horror or comedy, you might walk into a room, and then there's a chainsaw and a cream pie, and it's looking to see which one you pick up.

Now, there are a number of approaches we can take towards this. One might be language to parsing, in language parsing, this is the way computers try to understand natural language. They'll look at the string of words and from those try to build higher and higher levels of features and eventually parse it into a complete sentence. I suspect there might be something similar possible with story parsing, where you look at an event stream and look for higher and higher levels of structure coming above to build the computer's understanding.

Let's assume that, for instance, we could do this. If the computer could parse this story, then it could actually start changing the presentation of the story in something like The Sims. We could start having the camera angles change, we could have the lighting change, the music. So, if it thinks you're doing horror, the lights start to come down, the spooky music starts, you see lightning in the background, the camera gets really close in, ambient sound effects, and all this stuff.

And the next step would be probably for the computer to actually start driving the events, rather than the events being random as they are now, events that are there to support your story.

Another way to look at this is that we have users always in the game trying to confine these success landscapes. If the computer can detect what goal the user is trying to pursue in the game, then perhaps, so say that a million people are playing this game every day, and after they play it, there's some fitness test.

You can measure how long you've played, maybe you score at the end, you give it a 1 to 10, "Oh, I really liked that experience today." Then it goes back up to the server, and it compares all the results. So, at the top level, we can have the server discovering these rules of story. It's not like we're trying to engineer a story grammar, what we're trying to do is develop a system to where we have enough information for the servers at the top end to discover this. This is very much like SETI at home, where we're doing a vast parallel search, except in this case, we're really using parallel things like SETI at home.

So, I think long term this might really be practical. And of course, you can look at different players, and they're going to want to play the games in different ways. So, a mapping that might work for me won't necessarily work for you. So, at some point, you want to classify the players. It might say, "Oh, that's the type of player that really enjoys the comedy experience." So, they get this mapping, and they'll be evolving a certain mapping for that group of players. These other players, we know, elected to kill and torture their Sims, and so we're going to give them a completely different experience.

So, that's about it on this. One more thing: once you've done this, you know, assuming we could do this, which is a big "if" - I know it's a very difficult target - once you finish this, once you've done the parsing, presentation, the computer is helping you tell the story. In essence, you've made a movie at the end, what you can then share with your friends. Maybe you can even go back and edit the camera angles. Other people with the game, you should be able to send this movie, very much like Quake movies we're doing, which was a very cool idea. If you have the engine, it's a very small amount of data I have to send you to send the movie, or it could be still saved out as a JPEG or an MPEG and given around other people, posted on websites. So, really what I think this is going to end up being more like is The Truman Show, where you're Truman kind of living your life out. You know, there's boundaries, but the boundaries are as far away as we can make them, and the computer is back there, sitting there, looking at what you're doing, saying, "That might be dramatic. What if I had this happen?" And maybe trying things on you all the time. The computer might be sending NPCs into your life, you know, seeing, "Oh, I see this love story developing. I'd better make the jealous ex-girlfriend appear at the door right now," so always, you know, trying to keep the dramatic tension going.

So, that's about all I have to say on that. I just wanted to close with one more remark, and that's that in game design over the years, I found that two things really made the biggest difference that I've seen in design, and it has to do with self-confidence and determination. A lot of times, you'll come up with a really crazy idea, and you'll start telling people about it, and they'll say, "That sucks. That's horrible." And frequently, you know, that's not always the case. I've only had two games where I had that experience where most of the people I told, you know, my idea for the game, they said, "That's a horrible idea," and those two games were Sim City and The Sims. All the other games I would describe, when people say, "Oh, sounds great. Let's do it," and they were all on board. They understood it.

So, basically, you know if you have an idea like that, and you go around telling everybody, and they tell you it sucks, you know, will prevent it pragmatically. It probably does suck. But um, there's really a chance that it doesn't, that you're actually going to jump outside the box. If other people don't understand it, that's probably a good thing, and that's where the determination comes in. At that point, it's entirely up to you to have the determination to fight, to scream, to do whatever it takes to get that idea out there, and maybe, you know, maybe it'll be a success. So, thank you.

Description of the process of determining the theme of the story that the player has in his head and setting camera angles and lighting - a description of what the developers wanted to implement for The Sims 2, one of the goals about told Mike Sellers (Mike Sellers), game designer of game:

Make the environment adaptive based on the player's decisions. This is probably my favorite thing that didn't make it in: we wanted to vary the lighting, camera angles used, and music based on the kinds of objects the players purchased -- so if you bought a bunch of creepy things vs. hearts and flowers, we'd use lighting/sound/view angles corresponding more to a horror movie vs. a romance. Among other things we were able to show that the same animation of two Sims kissing looked and felt very different when set to romantic violin music vs. smoky sexy jazz music (one exec who watched the two sequences said the jazz one had a lot better resolution -- but it was literally the same video of an animation in both cases).

From this answer you can also understand the degree of readiness of this particular part of the system, which is responsible for changing lighting and camera angles, before it was hidden.

But the system itself, the system of story recognition in the player's head is still in the game, and it works. How does it show itself in this case? You can find this out at the link directly after the question you asked - "How does the system behave?":

Add to that these interviews with Will Wright:

EA SimCity 2000 Will Wright Interview

Will Wright's 5th minute answer to "What do the people get out sim games?".

And this interview:

Will Wright on Big Thinkers [March 2000]

Here's what Will Wright says in this interview:

I mean, I think there's certain areas of technology that I would like to explore, and I'm not sure what would come out of that exploration. Like, what artificial intelligence or something like that? Well, actually, I think more importantly would be um, adaptive, uh, intelligence, adaptive computing, where the game can actually measure what I'm doing, look at what I'm enjoying, what I'm not enjoying, and then reprogram itself around my activities. So, I might get this game, and I, the longer I play it, it's basically uh, rewriting itself around what I like to do. And so, you might get the same game, and after a month of play, your game and my game are totally separate. You know, we're doing totally separate activities, totally separate settings. That would be good. Yeah, and you know, so in that case, it's more like having a, you know, it's like when you're a kid and you have, you know, a really good friend that you like to play with, and together you build these shared imaginative worlds and these activities you go and do together. Um, you know, I think the computer could do that to some degree in the future.

This relates with a previous interview and it also ties in with an answer from Mike Sellers to a question about goals in developing The Sims 2.

And it looks like The Sims 2 should be like this, a perfect Will Wright game.

На русском языке (On Russian language):