Lifelike Agents

Uploaded from authorPOINT
Views:
 
     
 

Presentation Description

No description available.

Comments

Presentation Transcript

Lifelike Web Agents: 

Lifelike Web Agents Erik Harris Eva Jettmar Sangeeta Jesrani Slony Taranjot Kaur

Key Questions: 

Key Questions What happens if an animated agent is completely lifelike (looks like a person)? Does it matter whether the agent is labeled 'human' or 'computer-generated'?

Key Questions: 

Key Questions Who are users more willing to 'forgive' if the agent is deficient… - socially (nonverbally flat) - technically (slow frame rate; jagged) ?

Importance of the Questions: 

Importance of the Questions Psychological Issues - 3 explanations: A. Users will rather forgive a computer agent (expectancy effect). B. Users will rather forgive a human agent (specialist effect). C. There will be no difference (The Media Equation).

Importance of the Questions: 

Importance of the Questions Design Issues: Lifelike Agents (Ananova, Video Rewrite). How much does it hurt when agents are less than perfect? What is more important to users: Technical or social quality of a lifelike agent?

Context: 

Context Auction website Auction Products are described by lifelike agents in short clips. Agent is labeled 'human' or 'computer generated' (in fact always human).

Questionnaire: 

Questionnaire Attitudinal Measures: - Quality of website - Credibility of agent - Personality of agent - Liking of agent - Interest in using agent for other web tasks

Questionnaire: 

Questionnaire Behavioral Measure: Bidding Amount

Procedure: Lab Experiment: 

Procedure: Lab Experiment Participant seated in front of computer with auction review website up. Participant sees products and watches agent descriptions. Participant bids on product, then goes on to the next product (10 products total).

Procedure: Lab Experiment: 

Procedure: Lab Experiment Participant completes questionnaire. 3 days later: Memory test (web-based)

Demo: 

Demo

Manipulations: 

Manipulations Factor 1: 'Human' - 'Computer' Agent Factor 2: Nonverbally social - flat Factor 3: Full-motion video - jagged video (4 fps)

Results: 

Results

Results: 

Results No differences between 'human' and 'computer-generated' agent.

Quality of Website: 

Quality of Website

Agent Credibility: 

Agent Credibility

Liking of Agent: 

Liking of Agent

Involvement with Agent: 

Involvement with Agent

Interest in Agent for tasks: 

Interest in Agent for tasks

Bidding Amount: 

Bidding Amount

Psychological Implications: 

Psychological Implications No difference between 'human' and 'computer-generated' agent. The Media Equation applies to lifelike agents.

Design Implications: 

Design Implications Appropriate social behavior of agent is extremely important. Technical perfection is not important. Social cues come through even when technical quality is not very good.

Open Questions: 

Open Questions Lifelike agents vs. animated agents. Lifelike agents and personality. Appropriateness of nonverbal cues.

Contact:: 

Contact: Eva Jettmar true@stanford.edu