Exploring the mental space of autonomous intentional agents

Area of Science:

• Cognitive Psychology
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Inferring intentions from motion is a fundamental human psychological ability.
• Understanding how humans categorize intentional agents and actions is crucial for fields like AI and robotics.

Purpose of the Study:

• To empirically investigate the psychologically natural categories of intentional agents and actions.
• To identify key parameters that influence human perception of agent behavior and intent.

Main Methods:

• Utilized computer-generated autonomous agents with programmable "personalities" in a 2D environment.
• Employed multidimensional scaling to analyze participant judgments of agent behavior similarity.
• Manipulated agent reaction parameters to nearby agents at a critical distance.

Main Results:

• A primary dimension in subjective mental space related to agent "personality" was identified.
• A specific parameter governing agent reactions to proximity significantly influenced perceived behavior.
• This parameter was found to modulate a more fundamental dimension of perceived hostility or friendliness.

Conclusions:

• Human perception of agent intentionality is significantly shaped by specific behavioral parameters, particularly proximity reactions.
• These findings contribute to understanding the psychological basis of social perception and intent attribution.
• The study provides insights for designing more perceptually intuitive and predictable artificial agents.