Posted by Michael on April 25th, 2006, in Development

Penny Baillie-de Byl summarizes the reasons why we do things as a force of nature. Nature (which almost sounds like “god” in her exposition) wants us to do certain things (namely survive as individuals and survive as species). To make sure that we do these things, nature rewards us when we do them. The rewards come in the release of chemicals in our bodies that make us feel good. A long time ago, and for a very long time, hunting and gathering were necessary for survival. This is why nature rewards us when we hunt or gather.
While hunting and gathering has disappeared as an activity for most people, the natural reward system is still in place. This is why we experience pleasure when we play games.

Anyway, disregarding whether this is a satisficatory explanation of human behaviour, it struck me that it might be a plausible model for the behaviour of autonomous characters.
I’m always interested in finding ways to concentrate on the ear when trying to paint an eye (Kees Vollemans). So rather than starting from the individual’s motivations, drives, preferences and moods to make a character act autonomously, we could play nature and build a reward system for NPCs. Something like a list of actions that we want the NPC to perform. And for each action that it performs, it gets a number of pleasure points. That way, a characters only needs one motivation: to gain pleasure points.

To differentiate personalities, this list of actions, and the points awarded for performing them, could be slightly different for each character. The list should also probably be dynamic. So we can change it from above if we want a certain thing to happen. Or to prevent a character from continuously doing the same thing to increase its pleasure points. Perhaps the number of points that the NPC gets for performing a certain action is relative to the amount of time that has elapsed since it last performed the action. That way each time an action is performed, the number of points decreases. But over time, the award increases.

It’s scary to reduce human behaviour to such mechanistic systems. But that is not the point. We don’t want to create autonomous organisms. We don’t want to recreate human individuals. We want to create characters whose behaviour has a dramatic and emotional effect on the player in the context of a narrative environment.

No Comments »

Posted by Michael on April 25th, 2006, in Books

With a title like that, it was hard to resist this book in the context of Drama Princess.
The enthusiasm tempered somewhat, after reading the first quarter of the book, because the target audience is clearly programmers, and not designers. Also, the focus is on NPCs as opponents to the player rather than elements in a narrative.

The book goes on to summarize the diverse techniques that games have borrowed from A.I. to build NPCs without becoming overly ambitious anywhere. It really is not much more than a very comprehensive first step towards programming NPCs the traditional way. The book virtually exclusively focusses on characters in games that behave like the player, either as opponents or partners. It also exclusively deals with classic A.I. concepts and not with any from the more “behaviorist” side of the A.I. spectrum, which seems more suitable for our goals.

That last chapter of the book deals with something closer to home: Creating Believable Non-Player Characters in which finally the distinction is made between provable intelligence and the perception of intelligence, or believability. Which is ultimately the only thing that matters, in my opinion.

Disappointingly but not unexpectedly, the writer immediately jumps to the common conlusion:

To truly create an NPC with believable autonomy, interaction, and presence we must focus on the development op the NPC’s mind because it is the essence of behavior that needs to be captured.

Obviously, we disagree. Not necessarily with her conclusion but with the supposed obviousness of it. Since all we care about is the perception of the player, surely expressing is more important than capturing. And why would the mind be the essence of one’s behaviour anyway?

Next she quotes Loyall’s requirements for believability in an artificial being (also quoted by Michael Mateas):

• personality
• emotion
• self-motivation
• change
• social relationships
• the illusion of life

Overall, this is a very interesting book for programmers who want to create NPCs for contemporary games. It covers all the tried and true systems and you can sort of pick what you need. It lacks optimisation routines and cheap tricks. So, despite of the exercises, it still remains very theoretical.

For designers or programmers unsatisfied with NPCs in contemporary games, this book doesn’t have much to offer, apart from giving one a sense of what’s out there. There’s no new ideas in the book or inspiring points of view.

1 Comment »

Posted by Michael on April 23rd, 2006, in Books

This is a collection of essays by many game developers. I have just started reading it but I have already discovered that our preference for “faking” the intelligence of our autonomous characters is not a new approach. It seems very common practice in games. At least in theory. In practice, it does seem that developers, being engineers, I guess, easily fall for the seduction of scientific AI. And since what they are making is just a game, the risk exists that they end up with “fake AI”, rather than “fake intelligence”.
Anyway, so far this book has been a very inspiring read, if only because of the many small practical tips.

1 Comment »

Posted by Michael on April 22nd, 2006, in Games

I have a problem with The Sims. The game is very depressing to me. If you’re not a dedicated gamer, your Sims quickly become sad or suicidal. So instead of the attractive doll house that the game seems to be, it turns into a hamster cage full of pathetic creatures that probably eat each other too.

Another reason why it is depressing is related to the AI model in The Sims. It is the typical model of the scientist or the engineer: break everything apart into its constituent parts and then put it back together with artificial replacements for these parts. For a game, this is perfect: you end up with a whole series of gauges that you all have to keep full. Obviously these gauges continuously drop and filling one of them often empties another. Hours of entertainment! If you’re into that sort of thing.
But the effects of this form of interaction on our feelings for the characters in the story is detrimental. We want to feel empathy for these very versatile creatures but they never give you a break. They end up whining and nagging so much that your own sympathy gauge drops very rapidly. And you start looking for ways to kill them instead…

Chris Crawford once paraphrased Hitchock with regards to The Sims: while the Master of Suspense considers drama to be “just like life, but only with the boring bits cut out”, The Sims is also “just like life, but only with the dramatic bits cut out”. He has a point. Even in the sequel Sims 2, despite of its heavy focus on having friends, falling in love, having children, breaking up, and even dying, etcetera, The Sims doesn’t move you. It doesn’t even move you as little as a banal soap opera. It plainly does not move you. With all their variations in appearance, all their animations, all their clever AI, The Sims do not succeed in doing that one thing. This makes me think that the solution of Drama Princess will not be found in quantity. Perhaps quite the opposite.

5 Comments »

Posted by Michael on April 22nd, 2006, in Development

When modeling a character in 3D, we have often felt the inclination to model it in a realistic way. Even though we know that a certain level of stylisation is much more desirable for our games. The same applies for natural effects in real-time. When trying to show rain, the urge to draw every raindrop is great, even though in the most realistic of media (film), raindrops can hardly be seen.

A similar thing seems to happen when creating autonomous characters. It is very easy to fall into the trap of realism: to analyze human behaviour, break it apart and put it back together.
This reminds me of an imaginary machine I invented as a child to prove or disprove the existance of the supernatural. It was basically a xerox machine for humans, inspired by teleportation. The idea was to analyze the human to be transported down to the level of atoms and then recreate him or her exactly with “off the shelf” atoms. If this copy had a soul, the existance of the supernatural would be disproven. 🙂

A similar thing happens with AI that attempts to be naturalistic: it breaks apart human behaviour and then puts it back together. And in the end result something seems to be missing: the soul is gone! With Drama Princess, we should attempt to create the illusion of the existance of a soul, an not the soul-less copy of a human organism.

2 Comments »

Posted by Michael on April 22nd, 2006, in Books

I chose this book to begin our research because it seemed like an easy and general introduction to some concepts in artificial intelligence. And indeed it was.
The focus of the book was very heavily on academic artificial intelligence and its potential and not so much on the pseudo-AI that we are more interested in.

No Comments »

Posted by Michael on April 22nd, 2006, in Games

Yorda is one of the autonomous characters in games that we find most inspiring for Drama Princess. The reason for this is that, although she does not possess very elaborate artificial intelligence, she has a major dramatic impact on the experience of the player. She feels very much alive and you feel affection for her.
In terms of technology, the character of “frail princess” that she plays, maybe be a way of cheating oneself out of developing a complex A.I. system. So the question remains as to whether this approach can be applied to characters with different personalities.

Still, being able to hold hands with a virtual character is simply one of the most moving things I have ever done in a game.

No Comments »

Posted by Michael on April 22nd, 2006, in Websites

GameAI.com is a huge collection of articles, containing, amongst many others, AI in Games: A Personal View by Richard Evans. The latter obviously talks about his design of the creature in Black&White. “In order for the player to see his creature as a person,” he says, “the creatures had to be: psychologically plausible, malleable and loveable.” And about the latter: “empathetic attachment is intrinsically reciprocal: (…) if you want the player to get attached to his creature, you must first ensure the creature is empathetically attached to you!” A very clever man, Richard Evans, I wonder what he is up to these days (since obviously Lionhead doesn’t need him anymore now).

No Comments »