I need the program info again – but in any case, it’s the AI behind the psychological profiling in Silent Hill, how it was boiled down from large amounts of psychological items to something that mostly works.

Gwaredd Mountain (Climax Group)

AiGameDev.com summary, and slides.

My Thoughts

I really enjoyed seeing how the metrics are gained in the game (even more so how it’s on a Wii) and how a psychological test done through the game is.

Talk Contents

This is more a talk on the opposite of knowing what the player thinks of the AI, but what the AI thinks of the player.

Silent Hill is a well known franchise, and better known for psychology rather then gore.

Question raised by the designers; “What if we can get inside the players head?” – use their fears. Hard, as it turned out. Human beings find this hard too.

Don’t know anything about psychology – we’re game programmers. Reading lots of books and blogs and anything else. Asked local experts – “Okay, you’re silly but we’ll help you in the right direction”.

Also got a psychologist and put some of the programmers into therapy.

Need a model of a human being and plug it in to go. But there is no convenient model by psychologists. The DISC Personality model, Myers-Briggs “dichotomies”, (Some) NLP Meta Programs, all different models with different dimensions using the dimension as the model.

For Silent Hill Personality Profile, got The Big Five (or rather four, since Neuroticism is hard), plus some morality ones. Got the model, but how do you build it up?

Very difficult with the interface to the wii being a controller, how to model it accurately? Well, ignore that to start with and use a real questionnaire, inspired by the therapy – based on Myers-Briggs. Also track what they look at and what they do is also tracked. Influences the model.

Self-fulfilling prophecies also are used – a jacket test where it doesn’t matter what jacket a key is, it is how they look for the key on them.

What is changed? Changed some of the detail not the core part. Characters can change depending on the profile and arc. The monsters change, angular to rounded. The adverts also change; along with the extra things – like what you get when you phone in-game numbers.

Does it work?

Are the profiles accurate? You get your psychological report at the end of the game. It is difficult to say; it appeared from testing to be mainly accurate but people could be reading into it like horoscopes.

The starting psychology test also affects the most things; so the in-game part perhaps doesn’t affect it as much but it is still not if it was working in a game or not; in games we have randomness involved in how the scene was setup (ie; might pick the first and so forth). The initial tests seem accurate however.

Is it more scary? People being scared by monsters would be the same regardless. Players however seem to think they are being watched and at least get a different experience with different psychological tests, thus replays.

It was also marketed and suspense involved there so players get involved, even if they don’t understand the limited things behind the game.

So is it more scary? Not as much to with psychology or personality; but just how you react to stimuli (primal reaction). Perhaps could have done physiological things to test what scares people but very difficult to do it at run time – a camera is needed.

Applicability to other games? Not all games have an in-game psychologist antagonist. Would be interesting to see more implementations of it, and more complicated analysis of meaning behind actions – pushing forward the movement stick for instance. Need to work on modelling the human computer interaction.

The challenges with doing it is content production; getting the amount of content in a length of 8 hours but has at least 3 times the content. Had custom cutscenes for content swapping. Testing was ad-hoc with some checking after playing, and had custom tools to inspect data.

Others; Left 4 Dead is most famous using this kind of profiling. Estimate “emotional intensity”, damage taken, things dying and so forth, which changes the population density.

Tomb Raider; Metrics like causes of death, total deaths, completion time, and used a neural network to do different types; Veterans, Solvers, Pacifists, Runners. Wouldn’t use neural networks personally to analyse it however.

Take Away

A unique and interesting take on “experience management”. Started with a game idea, and built the technology for the game, so build specifically for the game.

Questions

Q. What about giving them a new experience without any particular player action?

Unlike films players tend to interact in different ways. Some different metric then the psychological is certainly possible.

Q. The players unable to be honest with the game, were they typically happy with the game?

What the players infer from seeing it get an experience from the wrong input, makes them believe the illusion in some cases, but other cases players don’t go along with it.

Q. Did you learn anything odd about the psychological tests on co-workers?

With my co-workers you learn new things, but surprised? no.

Q. Using the Big 5 personality traits, why not neurotocism since it might be important in fear?

Very difficult to come up with a meaningful test in a game; couldn’t work out how to do it.