Exploring What People Need to Know to be AI Literate: Tailoring for a Diversity of AI Roles and Responsibilities
A CHI ‘25 research paper that uses service design’s value flow models, built for 41 AI applications across 14 domains, to identify the roles whose responsibilities AI complicates beyond the usual “user” and “developer,” then compares those responsibilities against the competencies AI literacy research already recommends. Along the way it splits AI users into distinct roles, separates an AI system’s strengths from its benefits, and demonstrates the value flow model as a way to find every stakeholder a technology touches.
License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0/ (stated on the paper’s first page).
Key points
- Value flow models: a service-design analysis that maps the value exchanged among all stakeholders in a service, each usually captured as a role, combining contextual design’s information-flow models with stakeholder maps. The central difficulty is choosing the right granularity; the aim is a “good enough” model that’s useful to reason with, not a precisely accurate one. See Service Design.
- Sixteen roles in five clusters: organizational (executive managers, legal teams, marketing/PR), regulatory (policymakers, legal professionals), media, builders/integrators (AI developers and data scientists, product managers, HCI/designers, AI integrators, data labelers), and users. Eleven of the roles had responsibilities AI was found to complicate.
- Users are several roles, not one: consumers (who usually choose the service), workers (who don’t choose their employer’s AI but can bear liability for mistakes it contributes to), citizens (who must use government AI to reach a public service, such as a city business-advice chatbot that told owners to take illegal actions), students, and impacted stakeholders (patients, welfare recipients) who never use the system but are affected by others’ use of it and need to know when AI is involved. See Human-AI Interaction.
- Medication-cabinet case: a nurse was convicted after dispensing the wrong drug from an automated cabinet whose search used type-ahead, in a list where drugs carry both brand and generic names; manufacturers then changed the software to require up to five typed letters before searching. See Predictive Input.
- Strengths aren’t benefits: what AI can do differs from what a particular role gains from it. Decision support that diagnoses textbook cases adds little for clinicians who already handle them, while a model with moderate accuracy can still deliver value where risk is low. Benefits differ by role (revenue for an executive, less tedious work for a worker, a better-informed choice for a consumer). See Human-AI Interaction.
- HCI/designers as an AI-complicated role: designers struggle to understand AI’s capabilities, to envision AI services that are actually buildable, and to design interfaces that reduce unintended AI harms.
- Gaps in AI literacy competencies: existing frameworks cover recognizing AI opportunities, building AI, and collaborating, and partly cover identifying risks, but barely address identifying benefits, estimating costs, weighing benefits against risks, or monitoring and refining deployed systems.