Conversational Characters

Week 6LCC 2700: Intro to Computational MediaSpring 2005Janet H. Murray

Turing

From thinking machines to credible machines

“Can machines think?” is a meaningless and loaded question

The foundation of the field that came to be known as Artificial Intelligence

The “Turing Test” — a human engages in natural conversation with a human and a machine. If the human cannot tell the difference, the machine passes — based on the Imitation Game (a party game in which players try to guess the gender of a

A specific, discrete state machine : context

Alan Turing, Can Machines Think? (1950)

Q. If machines could think, how would we know it?

A. Indistinguishability using the “imitation game”:

The new form of the problem can be described in terms of a game which we call the "imitation game." It is played with three people, a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a room apart from the other two. The object of the game for the interrogator is to determine which of the other two is the man and which is the woman. He knows them by labels X and Y, and at the end of the game he says either "X is A and Y is B" or "X is B and Y is A." The interrogator is allowed to put questions to A and B.

Some notable Objections to the Turing Test(as covered by Turing)

• Theological - thinking is a function of an immortal soul; machines can’t think– Turing: God could grant a computer a soul if he wished

• Mathematical - There are limits to what logic can answer– Turing: Humans are pleased with the fallibility of machines

• Consciousness - Only composition from emotions could = brain– Turing: We have no way of knowing if non-human things

experience emotion• Lady Lovelace - Computers are incapable of originality

– Turing: computers can show things that are not immediately recognizable

Turing Machine

• An abstract machine that represents computation• Model:

– A person executes ordered operations on the contents of an unlimited number of paper sheets. The sheets contain a finite number of symbols. The person stores one of a finite number of states.

• Example– Change the state of the current page to 1 and move

one symbol to the right– If your current state is 4, move to page 456

• A Turing Machine that simulates any other Turing machine is said to be a universal Turing machine

Turing Completeness

• A system equivalent to a universal Turing machine• Unlimited storage can be assumed or ascribed to qualify

a system as Turing complete– Babbage’s Analytical Engine– Programming languages (C, Java, Lisp, etc.)– Formal grammars

From Thinking Machines to Credible Machines

• Move from thinking to believability (Turing)• The responsibility of computation (Weizenbaum)

– Computation as a representational system• Computation as literary expression (Murray’s adoption of

W.)

Experiencing Characters in Traditional Media

Drama embodied actors reciting speeches and exchanging dialog, gesturing, using props in unisequential story

we interpret their inner life based on what they say and do while we are observing them directly

Filmembodied actors seen in moving images, with dialog, voice-over narration; shorter scenes, fewer words; smaller gestures; unisequential or multi-sequential

we interpret their inner life based on what the camera shows us of them and of their world

Experiencing Computational Characters

Computational characters come alive when they executebehaviors in response to our participation in their world

Inscription in both directions bits/clicks

Circular transmission display – input - output

Interpretation in both directions images, words – pointing, typed words

We interpret their inner life based on how they respond to what we do

Traditional vs Computational Character Making

All characters are imagined by creators and readers/audience as having inner, continuous reality, and a range of possible behaviors beyond what is represented in the text

• Writers imagine the character’s deeper self – the psyche, soul, feelings, consciousness, personalities out of which all the behavior we see arises

• In computational media, we can create a character by inscribing and representing the deeper self, and a range of possible actions, and letting the computer dynamically create the behavior : e.g. the Sims

Traditional vs Computational Character Making

In computational media, we can create a character by

inscribing parameters and/or rules, and then watching the computer dynamically generate the behavior

inscribing parameters and then operating the character

inscribing parameters and then interacting with the character

Parameterized Characters

The Sims : Parameters realized in behaviors

Characters in Computational Media

Characters can be created procedurallyby new behavior (rules)by parameters for existing behavior

Characters are made real by participationby scripting the interactor to form expectationsby giving the interactor props and actions to shape

their behaviorsExchanges between character and interactor should be

coherentreadable

Genre expectations shape interaction

• Dramatic genres: mysteries, thrillers, romances, situation comedies, etc.– Create expectations of possible sequences and

outcomes– Include familiar roles (detective, villain, Mr. Right,

insensitive husband, etc. )– Include familiar events and actions (interrogation,

fight, kiss, insult, etc.)– Include props that carry scripts and expectations

(blackmail note, gun, bouquet of flowers, sixpack of beer)

E.M. Forster: Flat vs Round Characters

• Flat characters do not change• Flat characters always respond in the same way to the

same situation• Flat characters make good comic characters

– Predictable, creating anticipation– Inappropriate, because repeat same response in new

situation

Classic Flat Character

Jack Benny, the stingy man

Thief: Your money or your life!(long pause)Thief: Your money or your life!!Jack Benny: I’m thinking….

Interactive flat character

• Less satisfying to play because too predictable• More satisfying to interact with because you can

anticipate and evoke the behavior

Interactive round character

• Hard!• “Selectively” round characters

Conversations with Characters:

• Loebner Prize established 1990– Hard to do it with knowledge base

• The real world is too hard to represent• Fact-based conversation is hard to fudge

– Yet same character (Julia) failed Loebner but passed the test in a social MUD

• Human beings are often unresponsive • Comic characters are often unresponsive• Human beings often converse in formulaic patterns• Dating / flirting insult conversation is very

formulaic

Loebner Prize 2003

?

Humans outperformed all the chatterbots

Humans outperformed all the chatterbots

Rollo Carpenter, Jabberwock Loebner Prize Winner 2003

http://www.abenteuermedien.de/jabberwock/

Conversations with Characters:

• Free text input creates high expectations– Façade– www.interactivestory.net

Then why did Eliza work?

• Clear scripting of the interactor by dramatic situation• Clever key word/response pairs• Very clever DEFAULTS• Character’s lack of responsiveness is dramatically

motivated

Dramatic Agency

• Procedural + Participatory + Dramatic compression• Character’s if/thens and parameters are suited to the

dramatic world• Interactor’s actions are well suited to the dramatic world• Character’s behaviors are evoked smoothly by

Interactor’s actions and satisfy dramatic expectations of the situation

• Interactor is able to do things that effect the character in significant, dramatizable ways

Exercise: Design a Conversational Character

• Character name and default image (sketch or describe)• Opening line (e.g. “Sorry to phone so late. I’m in

trouble.”)• Prompting label for the interactor (e.g. “Mom:”)• 7 key word/response pairs• Responses should feed back new possible

keywords• 7 default statements (when no key word match is found)