Research Article
Adult Learners' Perspectives of AI Applications in Supporting Andragogy
Synthesis: Kim, Lin, Yu and Detrick (2026) interview 20 adult learners to understand how AI applications can support andragogy — adult learning theory's emphasis on involvement, experience, problem-centeredness, and relevance. Through scenario storyboards and paper prototypes, they find that learners regard AI as a collaborative learning agent — a reflective partner they learn from and with — rather than an answer-giving oracle. Three design principles emerge for AI systems that support andragogy: human-in-the-loop (shared mental models and human-AI co-creation), emotional design (calibrating AI reliance and empathetic communication), and adaptability (continuous adaptation and interoperability).
Core Finding
Adult learners perceive AI as supporting every facet of andragogy — involvement, experience, problem-centered learning, and relevance — and view it as a collaborative learning agent that helps them solve complex problems, share knowledge, deepen understanding, and hone higher-order thinking. For design, they do not want perfect, zero-error AI; they want systems that are human-in-the-loop, emotionally attuned, and continuously adaptive. Because the study centers on perceptions rather than outcome measures, it contributes design-direction evidence rather than direct learning-gain claims.
How the study worked
Sixty adult learners from three universities in the eastern United States were screened on online learning experience, education level, gender, discipline, and the kinds of AI they had used; twenty were then interviewed in depth. The interviews were anchored in eleven AI-application scenarios — presented as storyboards and paper prototypes developed with 24 graduate students — so that learners reacted to concrete tools rather than to "AI" in the abstract. Each scenario was deliberately tied to one component of Knowles' andragogy, and the AI type was left open (intelligent tutoring systems, exploratory learning environments, AI-powered dashboards, automatic writing evaluation, and chatbots).
| Andragogical principle | Component | Scenario prototype | AI type exercised |
|---|---|---|---|
| Involvement | Intentionally motivated to learn | AI tutor offering immediate personalized Feedback, progress tracking, celebration of achievement | Intelligent Tutoring |
| Involvement | Intentionally motivated to learn | AI-powered teachable agent that the learner must teach | teachable agents |
| Autonomy | Autonomous learning | Monitoring tools giving adaptive assessment and data-driven insight into progress | Learning Analytics |
| Involvement | Self-assessment | Systems that analyze work, generate rubrics, and track ongoing performance | Automated Assessment |
| Experience | Transformative learning | Empathetic chatbot holding emotionally grounded conversations about different perspectives | Affective Computing, Conversational AI |
| Experience | Experiential learning | AI simulations (VR/AR) for risk-free practice with complex scenarios | Intelligent Tutoring |
| Problem-centered | Contextual learning | AI case library that generates and critiques realistic customized case studies | Problem-Based Learning |
| Problem-centered | Critical thinking | Interactive dashboard for exploring and analyzing information | Learning Analytics |
| Problem-centered | Critical thinking | Higher-order question generator that prompts introspection and multiple perspectives | Generative AI |
| Relevance | Goal orientation | AI planner that breaks large goals into actionable steps and schedules them | Self-Directed Learning |
| Relevance | Personal/professional relevance | AI career coach offering career-path exploration and tailored advice | Pedagogical Agent |
The scenario set is itself a design contribution: it shows how each andragogical principle can be operationalized as a concrete AI affordance, and it gives designers a checklist that maps tools to adult-learning aims rather than to generic "engagement."
AI as a collaborative learning agent
Across the four andragogical principles, a consistent picture emerges: adult learners do not treat AI as a source of final answers but as an interactive partner that prompts deeper inquiry. Participants repeatedly framed the ideal AI as a thought partner that asks reflective questions, challenges assumptions, and verifies rather than supplies. One put it succinctly: "it's better when AI makes me question my own answer" (P3). This orientation reframes the AI-in-education conversation away from automation and toward human-AI collaboration, where learners retain control over judgment and self-direction.
Involvement, autonomy, and self-assessment
AI supports involvement by sustaining Motivation — sparking questions, tapping interests, offering continuous Feedback, and acknowledging effort ("seeing my progress in real time keeps me motivated," P5). It strengthens autonomy by offering choices, creating flexible learning spaces, and promoting metacognitive awareness via AI coaches and interactive dashboards. For self-assessment, predictive and diagnostic analytics let learners review performance and set realistic goals while retaining control over the reflection process ("AI shows me where I stand without judging me," P11). Throughout, learners value the pairing of human-like conversational agents with data-driven dashboards.
Experience: transformative and experiential learning
AI supports experience through transformative learning — conversational agents as AI facilitators guiding structured role-play and debate that challenge assumptions, and immersive tools (AI-powered simulations, exploratory learning environments) that bridge theory and practice. Experiential learning emerges through realistic workplace simulations and virtual trips (AR/VR/XR) that let learners apply knowledge in safe, risk-free contexts — aligning with Kolb's experiential and Mezirow's transformative learning theories.
Problem-centered learning and critical thinking
For problem-centered learning, AI activates prior knowledge (question-generation tools: "AI helps me remember what I've learned before starting something new," P20) and provides data-driven, fact-checked insights. Crucially, learners want AI to strengthen — not replace — critical thinking: hone questioning skills, verify information, and organize ideas via concept maps ("seeing my ideas mapped out helps me see what's missing," P19). This echoes the knowledge base's recurring theme of AI as a probe for inquiry rather than a content delivery mechanism.
Relevance: goal orientation and career connection
AI supports relevance through goal orientation — real-time progress monitoring, personalized recommendations, and breaking larger goals into manageable steps ("AI keeps me on track when I lose focus," P16) — and through personal/professional relevance, where AI coaches and recommender systems act as career companions and networking hubs. This connects AI support to lifelong learning and higher education career trajectories.
Three design principles for andragogical AI
Human-in-the-loop (HITL): Learners want systems that build shared mental models with them — human-interpretable decision-making so they can understand and anticipate system behavior while the AI continuously learns their expectations, preferences, and goals. They also demand human-AI co-creation: "It's not the system that decides what to offer for my learning journey. I'd like to take part in deciding…" (P20), letting learners influence generative parameters and model choices as co-creators rather than passive users. This is a direct call to honor learner agency.
Emotional design: Aligned with affective computing, learners want AI to calibrate an appropriate level of reliance — being explicit about its capabilities, limitations, and sources so they can decide when to Trust or negate its output — and to practice empathetic communication that recognizes, understands, and responds to emotional states, cultivating positive emotions (excitement, joy, satisfaction) while managing frustration and anxiety. This emotional attunement supports Well-Being and self-regulation in learning.
Adaptability: AI should keep learning from new interactions after deployment rather than remaining frozen on static datasets, graduating from closed Problem Solving to varied real-world problems. It should also be interoperable, moving and exchanging data seamlessly across platforms and tools — even enabling multi-agent collaboration.
Relevance to the knowledge base
This paper is the companion to the knowledge base's Designing AI systems to support a productive-failure-based learning: insights from adult learners on AI applications entry (same author group), and together they form a cohesive picture of how adult learners want AI to scaffold rather than substitute for learning. Here the lens is andragogy specifically: it operationalizes Knowles' principles (involvement, experience, problem-centered, relevance) into concrete AI affordances and design guidance. It strongly anchors the Adult Learners, Human-in-the-Loop, and Affective Computing concepts, and connects AI design to autonomy and Learning Theories. Its finding that learners resist over-prescriptive, answer-giving AI reinforces the knowledge base-wide theme of preserving learner agency and judgment in human-AI collaboration.
What this means for practice
- Designers. Build AI that asks reflective questions, challenges assumptions, and verifies rather than supplies answers: participants repeatedly rejected the answer-giving oracle framing ("it's better when AI makes me question my own answer," P3).
- Designers. Make reliance calibratable by having the system state its capabilities, limitations, and sources so learners can decide when to Trust or reject an output.
- Designers. Design for emotional attunement as well as correctness — empathetic communication that recognizes frustration and anxiety while sustaining Motivation and self-direction.
- Designers. Leave learners in control of the learning path: participants wanted to shape generative parameters and model choices as co-creators rather than accept a system-decided journey (P20).
- Instructors. Map each AI tool to a specific andragogical aim using the study's eleven-scenario set — a teachable agent for involvement, a dashboard for autonomy, a case library for problem-centered learning — instead of adopting tools for generic engagement.
Limitations
- Twenty adult learners were interviewed from a screened pool of 60 degree-seeking students at three East Coast U.S. universities, so non-degree-holding adults and learners in other cultural and socioeconomic contexts are unrepresented.
- The data are perceptions collected with storyboards and paper prototypes built by 24 graduate students, not measured learning outcomes, and no authentic deployed AI system was studied over time.
- Participants reacted to eleven scenarios deliberately designed to map one-to-one onto andragogical principles, so the close fit between AI affordances and andragogy is partly a property of the instrument.
- Ethical dimensions — data Privacy, algorithmic bias, and Trust — were not probed in the interviews, and the authors frame the three design principles as design direction rather than validated effects.
Citation
Kim, J., Lin, X., Yu, S., & Detrick, R. (2026). Adult learners' perspectives of AI applications in supporting andragogy. Educational Technology Research & Development.