Meg Kurdziolek writes that modern AI interaction design must learn from the history of Human-Computer Interaction (HCI) to avoid repeating past mistakes. Using observations from robotics and early Xerox PARC studies, she argues that even highly intelligent systems fail if they do not account for "situated action"—the unpredictable, improvised nature of human behavior—and the sensory gap between machine perception and real-world context. To move toward successful Agent Interaction Models (AIX), designers must prioritize human competence over autonomy by creating interfaces that support continuous steering, collaborative sensemaking among multiple stakeholders, and transparent communication of uncertainty.
- The "man vs. machine" study by Lucy Suchman highlighted how rigid automation fails to account for the moment-to-moment improvisation required in real work.
- LLMs create a "stochastic parrot" effect where conversational fluency masks a lack of underlying physical or situational understanding.
- Effective Agentic UX requires an "intervention handshake," allowing users to provide lightweight, continuous steering without resetting the system state.
- Trust is often treated as a 1:1 interaction in current chatbot interfaces, but real-world high-stakes AI operates within complex multi-stakeholder networks (e.g., doctors, patients, and auditors).
- The "man vs. machine" study by Lucy Suchman highlighted how rigid automation fails to account for the moment-to-moment improvisation required in real work.
- LLMs create a "stochastic parrot" effect where conversational fluency masks a lack of underlying physical or situational understanding.
- Effective Agentic UX requires an "intervention handshake," allowing users to provide lightweight, continuous steering without resetting the system state.
- Trust is often treated as a 1:1 interaction in current chatbot interfaces, but real-world high-stakes AI operates within complex multi-stakeholder networks (e.g., doctors, patients, and auditors).
