h2g2 The Hitchhiker's Guide to the Galaxy: Earth Edition

I'm interested in your interest in figuring out how to use fictional characters across multiple media to build relationships between people and technology.

Please tell me about it - it sounds like fun. I'm getting into Natural Language Processing on computers, and seeing how 'character' can apparently emerge from programs that use words to communicate with users. Not quite the same thing as your job description, but gently related. Did you help with Starship Titanic?

I did work on Starship Titanic, as Producer in the end, you certainly figured out where I started getting interested in this subject didn't you?

Our parsing engine made me realise how direct engagement with a character could help pull players into the game world. I believe it can also add an engaging facet and interface to gameplay.

The specifics of what I want to play with is based on the premise that virtual characters can pull people into a more personal relationship with content.

I don't think it is necessarily about developing better and better parsers, although at some stage it is vital that we enable a dialogue between the character and the player. I feel there is a lot of relatively lo-tech techniques which can be used to build this relationship.

Does this help?

Yes, absolutely - I just wondered if I was barking up the wrong tree, or just barking.

I'm pretty convinced that the future of computer-human interaction (insofar as it involves creating a machine that can pass the Turing Test) will lie in finding ways to take advantage of the susceptability of humans to perceive subjective 'character' traits in computer-generated text. Humans are pretty poor at seeing the underlying level of complexity of things, because they are so sure that what they see is what's really going on. We can easily be fooled into believing we can 'see' all sorts of things. Take a flock of birds, for example. Seemingly incredibly complex behaviour - you'd think it would be impossible to see what's going on, who's leading the pack, who initiates the sudden turns of the flock, etc. The thing is, flocking behaviour can be modelled on computers using about 50 lines of code and very simple procedures, and the emerging behaviour is good enough to fool us.

The same is probably true of many things - humans just push their perceptions onto what is in front of them making them see something that's more complex than it really is. We are easily fooled. Televisions, as another example, fool our brains into thinking we're looking at 3D images that we view through a window, but it's just a cunning illusion of shading and blurring!

Have you heard the story of the guy who was up late on night in the college computer lab, and happened to log onto Eliza, an AI system that someone had left logged in which was designed to be a rudimentary artificial psychiatrist? It had a few cunning rules to deflect questions and comments back to its interlocutor to keep the 'conversation' going, and operated through standard text window, like online chat. Of course, the guy on the other terminal didn't know this, and just thought it was someone else in another room on the faculty. He got pretty angry with this 'person' who seemed to be trying to taunt him. The perfect axample of someone not knowing it was a computer he was talking to to start with, and so visualising its apparent character in terms of human traits that he could understand. He was fooled! (And very angry when he found out!)

I always wonder if there are any very clever and new computer programs being tested out in chat forums anywhere, and nobody knows it!

I heard of some prototype narrative generating software being tested in one fairly well known vrml-based chat world.

I'm expecting, in about 20 years' time, to be able to ring up any large company or help desk, and after talking with a human on the other end and then hanging up, be left wondering whether it was really a computer program.
It'll sure beat the "Press the zero button now", and "Press 4 to make me explode" things we have at the moment.

Voigt-Kampf tests here we come...