Human-AI Interaction |
Summary
This course offers a systematic introduction to the emerging field of Human-AI interaction, focusing on the wide-ranging opportunities and challenges posed by AI at the interface and beyond. AI has transformed interaction through perception, natural language conversational interaction and commonsense reasoning, as well as with highly autonomous operation, and simulated affective and social intelligence. Yet these advances bring significant new challenges, and design practices have often lagged behind. The course introduces core AI interaction paradigms, in the context of real-world applications, and examines their affordances and limitations, particularly in relation to predictability, uncertainty, safety, interpretability, value alignment, and bias.
Objectives
Introduce key Human-AI interaction paradigms; direct manipulation, ubiquitous computing/mixed reality, and agent-based systems; Recognise common usability challenges introduced by AI, such as inscrutability, unpredictability, non-determinism, and difficulties in preventing and recovering from errors.
- Evaluate the pros and cons of anthropomorphism and treating AI as social actors in interaction design; Gain hands-on experience designing perceptual, predictive, and conversational interfaces.
- Analyse algorithmic decision-making systems, focusing on interpretability, explanation, human-in-the-loop processes, and the risks of automation bias; Examine information asymmetry and the principal-agent problem in autonomous AI systems; Explore interpretable and explainable AI methods, including counterfactual approaches such as LIME and SHAP, understanding their capabilities and limitations.
- Critically assess unethical applications of Human-AI systems, including behavioural micro-targeting, deceptive design patterns, and the misuse of deepfakes.
Contents
- Foundations and Paradigms –interaction paradigms from traditional graphical user interfaces to intelligent, adaptive systems; Usability challenges in AI Systems, including opacity, unpredictability, non-determinism, and difficulty in recovering from errors;
- Perceptual and Agent-Based Interfaces: interaction modes that leverage human-like capabilities, such as vision, speech, and natural language understanding; user engagement with conversational agents and multimodal systems;
- Predictive and Affective Computing: systems that sense, infer, and respond to user states; how AI can personalize experiences based on predicted preferences or affective states; risks of misinterpretation and overfitting;
- Algorithmic Decision-Making and Human-in-the-Loop Design: decision-support and automation systems; assessment of how AI-driven decisions can be made transparent and interpretable, how to balance human oversight, and how to mitigate issues like automation bias and over-reliance on algorithmic outputs;
- Ethics and Manipulation in Human-AI Interaction: how AI can be used to persuade (or manipulate) users, including through behavioural targeting, nudging, and the deployment of dark patterns; dangers of synthetic media and deception;
- Anthropomorphism and Social Perception of AI: psychological and social dynamics of human interaction with seemingly intelligent agents; long-term implications of AI systems on individuals and society.
Requirements
Interaction Design recommended, not required; AI Governance recommended, not required
[an error occurred while processing this directive]