AI-ADU: From Ideas to Practice | Using AI Responsibly in Adult Education
AI can make adult learning more personal, accessible and engaging – but only if it is introduced with care. Every promising tool brings risks of bias, exclusion and over-reliance. This article, based on Module 2 of the AI-ADU training programme, looks at how adult educators can move from ideas to practice, what real pilots teach us, and how to keep AI responsible, inclusive and human-centred.
Before adopting AI, educators should ask three questions:
- What problem are we trying to solve?
- Does AI improve learning outcomes or teaching efficiency?
- Could the same objective be achieved without AI?
Where the answers point to AI, it can be useful for personalised learning, assessment support, reducing repetitive administrative work, and accessibility and inclusion for learners with diverse needs.
A roadmap for piloting AI
The safest way to bring AI into adult learning is step by step. Module 2 suggests a five-stage roadmap:
- Discover possibilities – explore how AI can support teaching, learning, accessibility and engagement.
- Test in context – try AI tools in small, low-risk learning activities before wider implementation.
- Adapt to learners – use AI to personalise learning and respond to diverse needs and abilities.
- Share and reflect – exchange experiences, challenges and ideas with colleagues and learners.
- Lead innovation – promote responsible, inclusive and pedagogy-first uses of AI in education.
Throughout, the principle is the same: start with small and manageable activities, identify where learners and educators need support most, and combine AI tools with human guidance and collaboration.
Six case studies, six bias dilemmas
Real-world pilots show what AI can do for adult learners – and where it can leave people behind. Each of these cases comes with a “bias dilemma”.
| Case | What it does | The bias dilemma |
|---|---|---|
| Berlitz · Azure AI Speech | Assesses pronunciation in real time and adapts speaking practice to learners’ level | Early versions struggled with non-standard accents, penalising learners from diverse linguistic backgrounds |
| Be My AI | Gives visually impaired adults real-time visual descriptions via a mobile app | Struggles with complex or low-light scenes and needs stable connectivity, affecting learners in under-resourced areas |
| Pounce (Georgia State University) | A 24/7 chatbot supporting adult learners through enrollment and coursework, reducing dropout | Could misidentify at-risk learners from incomplete data; learners with limited tech literacy struggled with the interface |
| Carnegie Learning MATHia | Adaptive, self-paced maths instruction with real-time feedback and instructor dashboards | Relies on learner motivation and does not address maths anxiety without human encouragement |
| Squirrel AI | Adaptive tutoring for professional exams, diagnosing gaps and generating practice exams | Early models favoured predictable study habits, marginalising adults with irregular schedules such as working parents |
| Project Euphonia | Speech recognition trained on impaired speech, with real-time transcription and custom voices | Early models favoured predictable impairments, potentially excluding adults with severe or unique conditions |
The common lesson: these tools work best for learners who are already advantaged. Pairing AI with human input – peer practice, role play, mentoring, group tutoring, braille, audio guides or sign language – is what makes them inclusive.
The ethical challenges
Beyond individual tools, AI in education raises four broad ethical challenges:
- Algorithmic bias and inequity – AI systems can reproduce biases related to gender, race and socioeconomic status. This may lead to unfair assessments and unequal learning recommendations, with under-represented learners particularly at risk.
- Lack of transparency – many AI systems operate as “black boxes”. Learners and educators often cannot understand how decisions are made, which reduces trust and makes accountability difficult.
- Privacy and data security – AI platforms collect large amounts of personal and educational data, with risks of data breaches, misuse and unethical surveillance.
- Accountability and autonomy – excessive reliance on AI may reduce human agency, and there are concerns that AI could replace rather than support teachers.
There are pedagogical risks too. AI can automate poor teaching practices, spread misinformation and place too much focus on measurable skills, while critical thinking and ethical reflection are neglected. Plagiarism, academic integrity and over-reliance on technology are real concerns for learners and teachers alike.
Regulation has not fully caught up. Existing frameworks such as GDPR and the EU AI Act do not fully address educational risks, and a 2022 survey found that only 4 of 23 Council of Europe member states had specific AI-in-education policies.
Reducing the risks
The risks are real, but they can be managed. Module 2 highlights five ways to mitigate them:
- Bias audits before deployment, using diverse and representative training data.
- Explainable AI that gives learners and educators understandable explanations of how decisions are made.
- Human-in-the-loop systems, where teachers supervise AI-based decisions and can intervene and override automated outputs.
- Robust data governance, with strong data protection, encryption and user control over personal data.
- AI literacy, so that teachers are trained to use AI ethically and evaluate tools critically.
“Human in the loop” deserves a closer look. The term can suggest that AI is the main actor and the teacher only steps in to correct errors. Module 2 argues for the opposite: education is not an engineering process but a relational and social act, grounded in language, curiosity and care. Pedagogy must be the foundation, not a fallback for when automation fails.
Good practices for everyday use
- Never enter personal data, non-public information or internal documents into generative AI tools.
- Review all AI output before use; you remain responsible for the result.
- Be transparent: declare substantial use of generative AI as “AI-assisted”.
- Teach learners about AI critically, including its limitations such as deepfakes and bias.
- Rethink assessment, especially homework, since no tool can reliably detect AI use.
- Minimise the environmental impact by avoiding unnecessary, computationally heavy tasks such as non-essential images or videos.
Building inclusive AI practice
AI should enhance human teaching, not replace educators. Adult learners have different needs, motivations and digital skills, and some will need extra support to use AI tools at all. Human feedback, empathy and flexibility remain essential. Inclusive AI means balancing innovation with accessibility and fairness – starting small, checking who might be left out, and keeping the educator firmly in charge.
This article is based on Module 2 of the training programme of the AI-ADU project “Building paths to the future” (2024-1-LT01-KA220-ADU-000255165).
Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the National Agency. Neither the European Union nor National Agency can be held responsible for them.