These notes are a summary of concepts presented in “How might people interact with agents.”
Donald A. Norman. 1994. How might people interact with agents. Commun. ACM 37, 7 (July 1994), 68–71. https://doi.org/10.1145/176789.176796
- User Control and Trust
- Elements of control
- Systems must ensure users are in command of their operations
- Privacy
- Integral to trust
- Maintaining confidentiality is a subset of trust
- Providing accurate expectations
- Minimize false hopes by aligning system capabilities with user expectations
- Conceptual model transparency
- Reveal underlying operations to help users understand system actions
- Elements of control
- Managing Human-Agent Interaction
- Removing complexity
- Simplify the interactions while making actions traceable when needed
- Feedback customization
- Allow users to adjust the form and level of feedback
- Contextual understanding
- Provide updates that are comprehensible and reassuring
- Removing complexity
- Safeguards and Constraints
- Runaway computation prevention
- Integrated mechanisms to keep systems within safe operational bounds
- Undo capabilities
- Ensure actions can be traced and reversed when required
- Explicit permissions
- Avoid unwanted or expensive operations without user consent
- Runaway computation prevention
- Autonomy and Technical Challenges of Agents
- Agent independence
- Agents are often created independently, leading to potential conflicts
- Synchronization issues
- Address contradictory actions and asynchronous coordination
- Distributed environments
- Manage complexities in networks, processors, and spatiotemporal relations
- Agent independence
- Building Social Acceptability
- Trust through transparency
- Provide consistent updates on agent actions and system state
- User reassurance
- Ensure users can verify actions and maintain privacy
- Reliability and acceptance
- As system reliability improves, user trust and comfort will grow
- Trust through transparency
- Designing Effective Communication
- System image
- Present capabilities and actions through an appropriate conceptual model
- Feedback channels
- Use graphics, sound, and verbal messages to convey updates
- Agent advice and information
- Clearly communicate recommendations and outcomes
- System image
- Temporal and Spatial Considerations
- Temporal coordination
- Address asynchronous operations across distributed environments
- Impact on other activities
- Ensure agents act in harmony with unrelated systems and processes
- Temporal coordination