The principle of intent - technological perspectives m. b. cowley pg dip-stat ba dphil

The relevance of intent to human-androidstrategic interaction and artificial consciousness

Michelle CowleyUniversity of

Southampton

RO-MAN 2006– The 15th IEEE International Symposium on Robot and Human Interactive Communication, 6-8 September 2006, University of Hertfordshire.

Overview

Introduction to the premises of the paper Androids and their human counterparts The principle of intent in human reasoning Strategic intent as a psychological benchmark in

android science Conclusion

Introduction: Illustrative example

Consider Stanley Kubrick’s 2001: A Space Odyssey

HAL (Heuristic Algorithm) vs Frank in a game of chess;

But unknown to Frank… Strategic plan: To kill the astronauts Strategic intent is fundamental to human

consciousness

Androids and their human counterparts

Is the primary problem in android science one of external design?

The Uncanny Valley (e.g., MacDorman et al., 2005)

Androids fall victim to this uncanny valley because they are perceived to be human-like but lacking something.

The ability to reason with intent may facilitate perception of the android as human enough to bypass the uncanny valley– Hunchback of Notredame.

The principle of intent in psychological theories of reasoning

The word ‘intent’ tends to be synonymous with other nouns such as plan, purpose , objective, aim, target etc

To socially interact effectively, an agent should be aware of its own intentions, to be able to conceal those intentions, and anticipate the intent of its human counterparts in the ‘mind reading’ sense explicated by the Turing Test (Turing, 1950)

For example, planning an action and reasoning with evidence towards a desired or undesired result (e.g., Harman, 1985).

Intentional states are assumed to consist of a representative context in a psychological mode

That is, intent may be represented internally as cognitive models

Mental Model Theory (Johnson-Laird, 1983; 2005)

The principle of intent in psychological theories of reasoning

Intentional Models Consider what people may represent in the proceedings of

the following: A man smoked a cigar and was killed because an explosive

was hidden inside it. The police find that the cigars in the man’s cigar box have been skilfully rewrapped with explosive hidden inside them. Several strands of long hair are found underneath the cigars. The man’s wife Martha has been accused of the murder [cited in Byrne, 2005].

Premises If Martha’s hair is in the box then she is the murderer… Martha’s hair is in the box … Conclusion Martha is the murderer

Strategic intent as a psychological benchmark in android science

Cognitive Science expOpponent Interaction5 novices and 5 masters3 standard board positions3 mins think-aloudVerbalised move sequences

(Chess program Fritz 8)

Adapted from Cowley & Byrne (2004, and in submission)

Expected and intended outcomes

StartingPosition

Node

Terminal Node

Possible Falsification

- f5 gxf5

Bc3

Qg2 Rxa3 bxa3

Nxc3

Adapted from Cowley & Byrne (2004, and in submission)

-

Masters attempted to seek falsifications to their plans whereas novices tended to only intend confirmations of their plans.This result has been replicated (Cowley & Byrne, 2005, and in submission)

Human vs. Android chess players:A classicist experimental

approachAndroids and humans play chess Imagine a chess player instructed to interact with an android

to ‘practice interacting with an android’ Unknown to the chess player the practice is an ‘intentionality

awareness test’ (The Sally-Molly dog problem; Wellman, &Estes, (1986)

2 conditions: Intentionally aware vs Mind-blind Android Present an ‘einstellung’ chess position and it is the human’s

turn to move The human player’s strategic choice may be affected by

his/her assumption of the android’s ability to represent an opponent’s intent

Conclusion

A case for reasoning in the strategic context of chess as a key psychological benchmark with which to evaluate human-like agents

Interactive research with androids who appear to have human-like strategic intent (optimal or biased/myopic) could narrow the gap between the theoretical sophistication of internally contained sapient artificial agents, and robotic agents who possess a human-like body.

Acknowledgments

The University of Southampton’s School of Social Sciences New Researcher’s Grant Scheme.

Comments and encouragement from Prof. Karl MacDorman