Human-Computer Interaction for AI Systems Design: Reflections on an Online Course on Human-AI Interaction for Professionals

Per Ola Kristensson, Emily PattersonAAAI 2025 · DOI

Abstract

Designing effective Human-AI systems requires more than just technical machine learning expertise or standard user interface design. This paper reflects on a professional education course at the University of Cambridge that addresses the pedagogical gap in Human-AI system design. By introducing a "Triple Diamond" framework rooted in design engineering, the course provides a systematic approach to building AI systems that are effective, efficient, and safe.

Introduction

While resources for machine learning (the "engine") and Human-Computer Interaction (the "interface") are plentiful, there is a significant gap in teaching the holistic design of Human-AI systems. These systems are uniquely challenging because they are complex, tightly coupled, and characterized by emergent interactivity—properties that only manifest during actual user interaction. You cannot design interactivity directly; you can only design the conditions for it. This paper describes a curriculum designed to move professionals beyond "design fixation" (jumping straight to a specific AI solution) toward a holistic, systems-thinking approach.

The Triple Diamond Model

The core of the pedagogical approach is the Triple Diamond Model, which adapts traditional design engineering methods to the AI lifecycle.

The Triple Diamond Model: a systematic framework for designing human-centered AI systems.

The model consists of three distinct phases:

  1. Diamond 1: Developing a Well-Defined Design Task. This phase focuses on creating solution-neutral problem statements. By defining human needs and system boundaries before selecting a technology, designers avoid the trap of "AI for AI's sake."
  2. Diamond 2: Elaborating Function Models. Using techniques like FAST (Function Analysis System Technique) diagrams, students decompose high-level goals into sub-functions. This allows for a granular determination of which tasks should be automated and to what level.
  3. Diamond 3: Designing for Emergent Interactivity. This phase addresses the "function carriers"—the AI algorithms and interfaces. It focuses on mixed-initiative interaction, explainable AI (XAI), and shared control metaphors to manage the inherent uncertainty of AI behavior.

Key Insights and Results

The course emphasizes several critical concepts for modern AI design:

  • Solution-Neutrality: Focusing on utility and core human needs over specific technical solutions ensures that the resulting system provides actual value.
  • Ironies of Automation: The curriculum highlights that increasing automation can paradoxically increase the demands on human attention and skill. Understanding this helps designers build better fail-safes.
  • Risk Management: Rather than aiming for "zero risk," the course teaches students to arrive at an acceptable level of risk through systematic tools like FMEA (Failure Mode and Effects Analysis) and fault trees.

The course has seen significant success since its launch in July 2023, with over 200 professional students. Evaluation metrics show high satisfaction, with over 90% of participants rating the experience as "good" or "excellent." Qualitative feedback highlights the value of the "methodical approach" in helping diverse teams—including UX designers, engineers, and managers—communicate through a shared language of functions and system boundaries.

Conclusion and Future Work

While successful, the course faces challenges in balancing intensity for working professionals and bridging the gap between pure design and technical systems engineering. Future iterations may include optional "learning pathways" for flexible pacing. The authors also see a need for more sophisticated modeling of interactivity, moving toward data-driven simulations of user intent to help designers predict system behavior before deployment.

Summary updated

BibTeX
@inproceedings{kristensson2025human,
  author = {Per Ola Kristensson and Emily Patterson},
  title = {{Human-Computer Interaction for AI Systems Design: Reflections on an Online Course on Human-AI Interaction for Professionals}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2025},
  doi = {10.1609/aaai.v39i28.35173},
}
Figure