EA's Presser: The Sims 3 and Chaotic Theorizing

si hoc legere scis nimium eruditionis habes

EA's press conference for E3 2010 last Monday afternoon brought all the usual splashy graphics and loud noises, but on one of the games they took an interesting tack. Instead of just showing a bunch of supposedly exciting footage of all the stuff you could do with the newest rendition of the Sims franchise, the presenter gave us a couple minutes of pseudo-philosophical discussion about complex and emergent systems.

I had a complex reaction to the presentation: a frisson of geeky delight that someone would go THERE in the middle of the game industry's giddiest Dionysian hype-fest; old Jeff Goldblum quotes running through my head; flashbacks of a message-board conversation several years ago that ranged all over Jean Baudrillard, John Conway's Game of Life and Cervantes' Don Quixote.

I do complexity, both at work and at home. In real life, my ducks don't ever get in a row. Rather, with all the stuff I've got going on, they're usually arranged in a 3D representation of a Julia set fractal, perturbed by an algorithm derived from the song Henry the Eighth played backwards by four 3rd graders on kazoos. As a software developer, this stuff is my bread and butter.

Thinking it through led me to an inescapable conclusion that at the base, his assertions that The Sims 3 is something new and different in that realm don't really hold water. The emergent behavior in the system doesn't come from the computer; it comes from the player.

Trying to explain why this heavy geeking was cool, they brought up an example. He showed some video footage of a Sim-town full of 30 copies of himself. They just set it up and let it run. All sorts of interesting and different things happened. Two of his guys even fell in love with each other. Each of them going their own way had somehow gone and done something so complicated that the designers couldn't even track it. They tried, but they couldn't predict what would happen from a given setup. I'm with him so far; that's the definition of a type of emergent system. Cool.

But then he kind of let the cat out of the bag. They found that while they couldn't by themselves predict what would happen at any given step, if they started another instance from the exact same place and ran the code just a bit faster, they could see what would happen. That sounds cool, until you think about it.

Oh. Whoops. The system played out exactly the same way every time. That may be complex, but it's not emergent. They haven't turned your new Sim into some sort of artificial life form. Each Sim is still just a direct expression of the system, like a chess piece.

Taken in its simplest terms, a Sim has always been a Langton's Ant with fancier rules. This new version hasn't changed that. Start more than one of them running on the same field with different rule sets, and you get what he was seeing.

It didn't shock me that they couldn't predict what the system would become from the starting parameters. We can't describe even the simplest of chaotic systems precisely. Figure and figure all you want, stacking up numbers like cord-wood, and at some point you always end up floating within the tolerance band of the numbers and the phenomenon you're describing. The figures blur into a fractal cloud of probability, like Schrödinger's cat, and your guess literally becomes as good as anyone else's. You're a prophet, not a predictor.

His cherry-picking that research about voluntary muscle movement to imply that a Sim has more free will than we do was pure slight-of-hand, worthy of the best of marketing mountebanks. The human mind is an emergent system in and of itself, with a level of complexity that can't be described even with big charts and an easel. The fact that one part of a human brain works before or after another doesn't have anything to do with how real-world actions work. We don't even know how many different parts are actually part of that process, much less which ones do what thing to what other things. And anyone who says they can tell you what it all means in the end from just one piece in the chain is trying to sell you the Brooklyn Bridge.

That whole discussion is a red herring worthy of Dashiell Hammett anyways. Whether you are on the side of free will or you buy into the notion of deterministic systems, we can all agree that we humans are adept at throwing a monkey-wrench into the works and doing things no one would expect. I submit that even in the lowest-tech game, there has always been an element of surprise, and even room for pure, blind luck.

The first gaming convention I ever attended was in 1987. It was for tabletop gaming (board games, role-playing, that sort of thing). I won my first tournament there—a board game called Talisman. Through a series of lucky circumstances that probably burnt my karma for months after, my Monk fought his way to the center square and held on to the Crown of Command long enough to win the game with only the slimmest margins of life and equipment. If my closest opponent had owned one more life-chit, he would have bopped me on the head like a field mouse and won.

Anyone who knows me could tell you how completely outside my usual nature that was. I'm way too cautious to ever just scream-and-leap like that. I don't work to maximize the possibilities of probability; I work towards certainty. Or, to use a programming metaphor, I want robust code in lieu of elegance. Those are my parameters.

I have no idea how many monkeys would have had to be typing and for how long before they produced a variant of my normal self that would have taken the decisive step that would start the charge to the center. There was no way for me to have known at that time how everyone else's moves would play out and make success possible. Nor could I calculate the nearly Vulcan odds it would take to describe the series of dice rolls, movements, and card plays by myself and my opponents that would be required to duplicate it. To this day, I am mystified. I've played the game hundreds of times since then and never come up with an endgame solution that was even close.

But it's not out of line with the human system. It's our nature to dare greatly at times. The reason quarterbacks will throw a Hail Mary is because, sometimes, the prayer is answered. We do many things that don't bear up under the weight of logic or probability. We make intuitive leaps. We fail to see possibilities until too late. We make wrong choices for stupid and not-so-stupid reasons.

We haven't found a way to really write those into an equation. We can write code that says, "If no one interacts with you for X amount of time, run the Lonely43 engram." We can perturb it by adding secondary clauses that mean something along the lines of "Each time you start a Lonely engram, you get a gradually increasing random chance of Melancholy and then run the (Ben&Jerry's + pajamas - shower) variant for X days." You can even say that after a certain number of cycles of that, you get a randomly increasing chance of the GoneCrazy engram.

Problem is, if you stack enough of that up to even approximate a real human's range of possible reactions, you're back floating around in the tolerance band again. Current modes of logic can only take so much of that before they can't get to the bottom of the decision tree and get a straight answer. Your Sim is stuck staring at the wall where the door used to be.

When discussing complexity in the context of computers, any given equation or system can be classified as either "P" or "NP." "P" in both those terms stands for polynomial, and is referring to the amount of time it would take a computer to calculate or verify the results of the equation or system. P is something a computer can do in a reasonable amount of time. NP is something we can do—often fairly easily—but it would take a computer a very long time to do it, if at all.

It doesn't take a complex problem to run into one of those situations. It doesn't even have to be all that hard. Back in 2000, an English mathematician named Richard Kaye proved that the old time-waster Minesweeper was classified as NP complete. I'm sure if someone did the math on The Sims 3, they would find a set of starting values that would render an NP verdict pretty quickly.

But that just means the computer has problems with it, not that people do. My mom kicks my backside at Minesweeper and she barely passed algebra forty-five years ago. But there's something wired in the meat somewhere that makes it work.

As marketing goes, his comments were a nice break between those vague suits and that one blonde woman who kept swearing as an attempt to make Bulletstorm seem edgy. I'm not saying The Sims 3 isn't going to be cool; I'm saying that unless you add the human element of the player, the game is not nearly as revolutionary and cool. Because the code doesn't make things complicated; we do.

Comments

I don't see where emergence is not allowed by definition in a deterministic system.

The Sims 3 example isn't emergent because of its repeatability. It is not emergent because the behavior they are describing as "emergent" is really just their inability to predict which of their pre-programmed behaviors will occur.

Danjo Olivaw wrote:

I don't see where emergence is not allowed by definition in a deterministic system.

I'd say that's "emergent-like". Kind of like you could make a calculator with a load of pre-stored solutions to common functions, or you could make one that evaluates functions. They're both right, but one is more limited.

One of the problems with modern game development is that it is so time consuming to get up to a 'basic' competitive level of game detail and complexity. Adding systems that can interact by themselves and produce a range of novel, interesting outcomes is one way to make content. Another thing is that emergent systems are more close to simulating real life, which may be desirable to some types of game.

Danjo Olivaw wrote:

I don't see where emergence is not allowed by definition in a deterministic system.

The Sims 3 example isn't emergent because of its repeatability. It is not emergent because the behavior they are describing as "emergent" is really just their inability to predict which of their pre-programmed behaviors will occur.

Indeed. Surely the only way something would be truly non-repeatable is if you made it partially random. I thought emergent just meant behaviour that wasn't explicitly encoded in the system's rules.

I always find the argument that human behaviour is different or 'free' somehow because we do stupid or crazy things a bit odd.

(spelling nazi note: it's sleight of hand, by the way, unless slight is a US variant?)

Whilst I agree with Danjo's definition of emergent (chaotic systems are also deterministic, it's just that infinitely small pertubations on the input can make a large difference to output). The fact that the EA guy put 30 copies of the same guy in probably didn't help. If he had 30 different people doing random walks, then you'd get a different output each time (assuming you have an srand(time(0)); statement somewhere). You just need to inject enough randomness into the system for it to have top level effects. Add lethal weather that murders half of your (differentiated) population.

The clockwork brain thing I've had issues with for a while. I put Roger Penrose's Emperor's New Mind down because I didn't agree with his assumptions. (I am well aware that Penrose is one of the most brilliant mathematicians ever, so it's not very likely that I have actually found a flaw in his argument) Penrose has a theory about Correct Quantum Gravity, the upshot of which is that strong AI is impossible because the brain has fundamentally non-deterministic behaviour (i.e. that probabilistic quantum effects govern the brain's function at a macro level). He used the Chinese Room paradox as a basic point to his thesis. This paradox has a guy in a room with an impossibly large English-to-Chinese phrasebook. He receives English text through one slot in the door, he looks it up in the phrasebook (which is complete for all possible input phrases), copies out the correct Chinese translation, and passes it out the output slot. Penrose's point is that the man does not understand Chinese, but from the outside it appears as if he does. Penrose likens our current AI systems to this room. My problem with this is that it's never been demonstrated to me that our brains aren't also like this room, but with the entire current state of the brain, it's memories etc. and all current sensory inputs as being the English text in.

Anyway, I think you can do better AI than we currently generally get, you just have to spend more time doing it, which is not usually cost-effective for the developers. I've written fuzzy logic-style routines, and whilst you still have to know what your complete outputs are, and they take a lot of tuning, you can get interesting behaviour out of them, especially with random factors added in. You can weight all the current options, and add enough randomness into them so that occasionally the CPU quarterback goes for it and takes the Hail Mary pass. I think a lot of the work being done in games AI is adding these kinds of human imperfections into the mix so that they make more interesting and less predictable opponents.

The real shock to me is that there's no random number generation going on inside The Sims 3? Now I haven't done a whole lot of AI programming, but when I have, there was always some degree of RNG to keep it from being too predictable.

Emergent gameplay, like Danjo & others say, is not mutually exclusive with deterministic though.

No, you guys have got the wrong end of the stick here. "Randomness" my itself does not create emergent properties - it just changes the statistical variation of possible responses. Indeed, randomness is likely to DESTROY emergent properties.

Go look at the Wikipedia for emergence. Emergent properties don't merely have individuals doing new weird things - those are just statistical outliers. Rather, they have new collective behaviors. I don't know the Sims very well, but something like flashmobs or wife-swapping parties or Conway's Game Of Life self-sustaining patterns are a possibility. Even flocking behavior can be thought of as 'emergent'.

Randomness can work to destroy coherence. If the bird's rule is to try to stay about 3 feet from some other bird, then adding a rule like "every few seconds go land for an hour" will only destroy the flock.

Moreover, repeatability does NOT mean that emergence has not taken place. This is something we learn in in science when doing Monte Carlo simulation/integrations: pseudo-random numbers are and should be repeatable. What they should not be is correlated. You can always change your random seed to get a different virtual-experiment. Emergence comes when some large sample of these experiments generates new coherent behavior that was not implicitly or explicitly in the rules of the game as you set it up.

So what is my point?

Um

What was the question?

Quintin_Stone wrote:

The real shock to me is that there's no random number generation going on inside The Sims 3? Now I haven't done a whole lot of AI programming, but when I have, there was always some degree of RNG to keep it from being too predictable.

Emergent gameplay, like Danjo & others say, is not mutually exclusive with deterministic though.

No, it's not that there isn't any random number generation, it's that the seed is probably a static value, so for any given scenario, the outcome is predetermined.

For speed, I would generate a table of random numbers at the start of n length(where n is some suitably large number), then just iterate through it.

I'm pretty sure that the Sims AI programing is a fairly complex Markov chain, where the Sims' next action is determined solely by their current state. If the pseudo-random number generator is seeded as a table, the AI becomes completely deterministic wrt the actions of each sim. Because of that, the result of any given run is predetermined by the seed.

Nathaniel wrote:

I don't know the Sims very well, but something like flashmobs or wife-swapping parties or Conway's Game Of Life self-sustaining patterns are a possibility. Even flocking behavior can be thought of as 'emergent'.

I was thinking more along the lines of a love triangle where each of the three Sim participants takes a different different role: aggressive, passive, and neutral. Over time the configuration is unstable, and the three Sims shift in role. Aggressive to neutral, neutral to passive, and the passive guy falls out of the triangle. A passing Sim finds the neutrality attractive, and aggressively courts, and in doing so attracts the passive Sim, re-completing the love triangle.

Over time the love triangle would shift through the populace of 30 Sim selves, each cycle spitting out a participant and consuming another. If conditions were right, further triangles could emerge and 'compete' for members. Each triangle would be distinct from others, even though over time they would no longer be comprised of the original Sims. They'd just keep sliding along the populace, like a kite in Conway's Game. I'd call that emergent behavior.

I have no idea if Sims 3 is capable of supporting that semi-stable configuration. There's probably simpler examples that are more plausible.

At times I wonder if the idea of emergent behavior is just our minds' inability to truly process reality. Our brains are capable of understanding the underlying properties of the system, and have too much pattern recognition not to see these strange behaviors. Even when we map out the behavior, the human brain doesn't have enough ram to hold the entire process all at the same time, and the discrepancy is filed symbolically as emergent.

I'm not sure. Can't say the term isn't well-defined because I can't consume the entire definition. "and it cannot be reduced to their sum or their difference" strikes me as the likely offending phrase that makes emergent a poor word.

Those wanting to be entertained by something unpredictable and intelligent should procreate or buy a cat.

BishopRS wrote:

Those wanting to be entertained by something unpredictable and intelligent should procreate or buy a cat.

I am scared by the binary nature of the that choice.

cube wrote:
Quintin_Stone wrote:

The real shock to me is that there's no random number generation going on inside The Sims 3? Now I haven't done a whole lot of AI programming, but when I have, there was always some degree of RNG to keep it from being too predictable.

Emergent gameplay, like Danjo & others say, is not mutually exclusive with deterministic though.

No, it's not that there isn't any random number generation, it's that the seed is probably a static value, so for any given scenario, the outcome is predetermined.

Yeah, fair enough.

BishopRS wrote:

Those wanting to be entertained by something unpredictable and intelligent should procreate or buy a cat.

Corrected.

You want cogent arguments about free will, AI, Chinese Rooms and so on, go read Dennett and Hofstadter. Especially Dennett.

But anyway, loved the Known Space reference.

Hans

Came for the Sims, stayed for the lively discussion on deterministic/emergent gameplay.

I love you guys.

Zelos wrote:

sleight of hand

Indeed!

Very interesting article; thank you.

Danjo Olivaw wrote:

I don't see where emergence is not allowed by definition in a deterministic system.

The Sims 3 example isn't emergent because of its repeatability. It is not emergent because the behavior they are describing as "emergent" is really just their inability to predict which of their pre-programmed behaviors will occur.

I'm not sure where you got the idea I said that. What I meant when I wrote this agrees with you on both points.

The main ideas I was trying for:
-- the Sims may be complex, but not emergent due to it's repeatability
-- Free will/ deterministic discussion is a red herring all together because it's irrelevant to the complexity/emergence discussion; both kinds of system can have emergent behavior
-- the level of system complexity required to bog down the computer is nowhere near as much as the human mind can handle

momgamer wrote:

-- the Sims may be complex, but not emergent due to it's repeatability

I'm saying that repeatability is not a disqualifier for emergence.

edit: Although the specific example of two Sims characters falling in love is also not emergent. Sims characters falling in love is supposed to happen.

Okay, now I disagree with you.

By itself, on a very limited scale, maybe. Complete, predictable congruence like the presenter describes, no. A lot depends on which type of emergence we're talking about, though.

We're running into the limits of the information I have. All we've got to chew on is that marketing wonk parroting concepts I'll bet he doesn't grasp deeper than a Newsweek article level of complexity. Without more specific information we're not much better off.

momgamer wrote:

By itself, on a very limited scale, maybe. Complete, predictable congruence like the presenter describes, no. A lot depends on which type of emergence we're talking about, though.

But there is only one type of emergence; that a system displays properties or behaviour that could not be determined from the rules the system follows. Whether a system is stochastic or deterministic is irrelevant

DanB wrote:
momgamer wrote:

By itself, on a very limited scale, maybe. Complete, predictable congruence like the presenter describes, no. A lot depends on which type of emergence we're talking about, though.

But there is only one type of emergence; that a system displays properties or behaviour that could not be determined from the rules the system follows. Whether a system is stochastic or deterministic is irrelevant

The irrelevance of the determinism discussion was one of my points.

But no, there are several types or levels of emergence, depending on how strictly you want to measure it and which realm the system is in (physics, nature, etc). I included a link in the original article that discusses the matter rather thoroughly -- click on the word "emergent" in the fifth paragraph and scroll down to the "Emergent structures in nature" heading.

momgamer wrote:

The irrelevance of the determinism discussion was one of my points.

That's not what this quote says:

Oh. Whoops. The system played out exactly the same way every time. That may be complex, but it's not emergent.

If I'm very much mistaken you appear to be saying that as the system is deterministic (plays out the same way given the same starting conditions) is it complex but not emergent

momgamer wrote:

But no, there are several types or levels of emergence, depending on how strictly you want to measure it and which realm the system is in (physics, nature, etc). I included a link in the original article that discusses the matter rather thoroughly -- click on the word "emergent" in the fifth paragraph and scroll down to the "Emergent structures in nature" heading.

That there are first, second and third order levels of emergence doesn't mean there are different types of emergence. There is one type of emergence different ways you might sub-classify what is happening depending on the system and the micro- or macro-scoptic level you are looking at.