Live examples are cases for initial testing ground of main thesis: the inferential axis, the encoding axis, and the representational axis

Case 1 — Visual disinformation

Model: Generative AI   +   Intervention: Visual design   ➔   Theme: Visual disinformation

A generative model produces images or visual content that is plausible without being true. The design intervention concerns how such content is composed, styled, and circulated — and therefore how visual plausibility is manufactured. The case asks what happens when visual design, traditionally a practice of clarification, becomes complicit in a logic of generation that optimizes for credibility rather than correspondence to fact.

Case 2 — Adaptive machine

Model: Machine Learning   +   Intervention: Human–machine interaction   ➔   Theme: Adaptive machine

A system that continuously learns from interaction with a user, adjusting its behaviour over time. The design intervention is the feedback loop itself: what signals the machine collects, how quickly it adapts, and how that adaptation is made legible to the person interacting with it.

Case 3 — Counterspeech

Model: Generative AI + Intervention: Discourse and narrative design → Theme: Sexist hate speech

A hybrid system combines rule-based detection of hate speech patterns with generative models that produce contextualised counterspeech responses. The design intervention concerns how a counter-narrative is shaped: tone, register, timing, and the decision of whether to address the target of hate speech, the perpetrator, or the bystander audience. The case asks whether an automated response can carry the social and rhetorical weight that effective counterspeech requires, and how design can make that response feel like a genuine act of language rather than a classifier output.

Case 4 - Legal judgment

Model: Predictive AI   +   Intervention: Decision making   ➔   Theme: Legal judgment

A predictive model is used to support or inform a legal decision. The design intervention concerns how probabilistic output is translated into a decision-support tool for a domain — law — that is itself a paradigm of explicit, rule-based, deontic logic. The case is a direct confrontation between two logics: statistical inference and normative reasoning.

Case 5 — Process optimization

Model: AI models   +   Intervention: Functional decomposition   ➔   Theme: Process optimization

AI models are used to optimize a workflow or process, with functional decomposition as the design method: breaking a complex process into elementary functions that can each be modelled, measured, and improved. This is the most classically logical of the seven cases.

Case 6 — Empathic dialogue

Model: Machine Learning   +   Intervention: Emotions   ➔   Theme: Empathic dialogue

A model is trained to recognize and respond to emotional signals in dialogue. The design intervention concerns how an inherently ambiguous and continuous phenomenon — affect — is rendered into something a system can act on, and how the resulting responses are perceived as genuinely empathic rather than merely simulated.

Case 7 — Clinical diagnostics

Model: Generative AI   +   Intervention: Decision making   ➔   Theme: Clinical diagnostics

Generative AI supports diagnostic decision-making in a clinical context, where responsibility for the outcome remains squarely human. The design intervention is about how a generated hypothesis is framed, justified, and handed over for clinical judgement — not how it is produced.

Case 8 — Stereotypes

Model: Rule-based & Generative AI   +   Intervention: Social psychology   ➔   Theme: Stereotypes

A hybrid system — explicit rules combined with generation — intervenes on the terrain of stereotypes and social cognition through a conversational agent. This is the most developed case in the collection, with an initial project plan already sketched: it moves from a question-and-answer model toward a dialogue model that no longer requires complete prior knowledge, training instead for uncertain conversation.