Lecture 39 — Education: How AI is Changing Learning

Part 1 (20 minutes): Basic explanation with analogies and examples

Goal: Understand at what level AI influences learning processes.

An illustrative analogy is that of a personal learning assistant: imagine every learner had a considerate mentor who observes progress and mistakes, provides appropriate exercises, offers explanations at different difficulty levels, and gives immediate feedback. Modern AI systems can take on parts of this role by analyzing data about learning behavior and generating responsive materials. These functions operate on three levels: (1) Diagnosis: identifying gaps in understanding, (2) Adaptation: selecting and tailoring learning tasks, (3) Feedback: providing responses and guidance for improvement.

Concrete examples make this tangible. Adaptive practice platforms adjust difficulty based on the answers given and systematically repeat what remains uncertain. Automated feedback tools can comment on the structure and clarity of written responses or give pointers for better argumentation. AI can transform instructional material into different formats, for example converting a textbook passage into a summary, multiple-choice questions, or a visualization. Language models can serve as "sparring partners" that help with formulating, translating, or understanding complex texts.

UNESCO highlights in its recommendations that such technologies offer opportunities for personalized learning but also raise questions about accessibility, fairness, and the role of the teacher. Education policy and schools therefore face the task of designing and using technologies in ways that strengthen learning and do not exacerbate inequalities (UNESCO, Recommendation on the Ethics of Artificial Intelligence, 2021).

It is important to emphasize that AI tools currently do not fully replace teachers. They are tools for diagnosis, practice, and feedback; pedagogical relationships, motivation, ethical guidance, and social contextualization remain central responsibilities of teachers and learning environments. The current scientific consensus views AI as a supportive element, not an automatic substitute for human educational relationships (Stanford One Hundred Year Study on AI, 2016).

Part 2 (20 minutes): Deepening and introduction of key technical terms

Goal: Understand technical terms and classify the main opportunities and risks professionally.

Key terms useful for the discussion can be explained in an application-oriented way. "Personalization" refers to the adaptation of learning content, pace, and sequences to individual learning levels. "Learning analytics" means the systematic analysis of digital traces (e.g., response times, error profiles) to detect problems early. "Formative feedback" is feedback during the learning process aimed at guiding learning; AI can provide this automatically at scale. "Automated assessment" refers to procedures that evaluate tasks mechanically, which is technically demanding and not yet error-free, especially for open-text tasks.

These methods open up opportunities that are systematically named in several reports. First, they enable individualized support because learners can receive feedback more frequently and more targeted. Second, teachers can be relieved of routine tasks through automation and focus more on didactic planning and personal support. Third, barrier-free and multimedia access can be created to support inclusive education. These potentials are noted as central advantages in public debate and policy papers (Stanford AI100, 2016; UNESCO, 2021).

At the same time there are tangible risks that must be addressed politically and pedagogically. First, the availability and quality of data are crucial: incomplete or biased data foundations lead to faulty decisions or unfair recommendations. The European Commission's high-level expert group on AI and the OECD emphasize transparency, non-discrimination, and accountability as guiding principles for the use of AI, particularly in areas with integrative effects such as education (European Commission, Ethics guidelines for trustworthy AI, 2019; OECD, OECD Principles on Artificial Intelligence, 2019).

Second, there is the danger of so-called "overtrust" or overestimating automated results. Learners or teachers might accept feedback from automated systems without verification, even though these systems still make mistakes or miss relevant nuances, especially in complex, creative, or culturally bound tasks. Third, new inequalities can arise: those who have access to high-quality AI tools benefit more; this can amplify existing educational differences if policy and institutions do not ensure equitable distribution. Fourth, data protection and control over personal learning data are critical points because learning analytics collect extensive information about behavior and performance.

Which competencies therefore become more important? International analyses, for example on the labor market and training trends, show a shift toward skills that go beyond pure factual knowledge. These include metacognitive skills (self-regulation, learning to learn), basic digital skills and data literacy (understanding how data are generated and interpreted), critical thinking in dealing with outputs of automated systems, as well as communication and collaboration skills. Reports such as the World Economic Forum's "The Future of Jobs" emphasize analytical skills and social competencies as key in a labor market changed by automation, which has a direct impact on educational goals (World Economic Forum, The Future of Jobs Report, 2020).

In conclusion for this deepening: policy guidelines and technical standards are important to limit risks. The OECD and the European Commission formulate principles such as robustness, explainability, data protection, and human oversight as prerequisites for the trustworthy use of AI. Such principles are not only legal or technical requirements but directly influence the design of teaching-learning scenarios and the training of teachers (OECD, 2019; European Commission, 2019).

Part 3 (10 minutes): Applications, limits, and thought exercises

Goal: Learn concrete application scenarios, reflect on limits, and stimulate personal reflection.

Practical applications already exist today: adaptive practice platforms in mathematics and language instruction, automatic plagiarism and writing-feedback tools, AI-supported translators and text-to-speech tools for accessibility, as well as simulation environments in which students can practice complex decisions in a safe setting. Such applications are documented in pilot projects and the broader market; their benefit is particularly evident when they are closely linked with pedagogical goals and teacher support.

At the same time clear limits apply. The scientific evidence base for long-term learning outcomes from AI interventions is still limited; there are not yet enough independent, published long-term studies demonstrating sustainable effects and unintended side effects. In addition, the quality of automated assessments for creative, nuanced tasks is not yet equivalent to human evaluation. At an organizational level, data protection, infrastructure, and teacher training are practical hurdles.

In closing, three short thought exercises are intended to make both the opportunities and the risks tangible. Each task is meant as a prompt for discussion and can be reflected on individually or in groups.

First exercise: Imagine an adaptive learning system that automatically suggests the next exercises. Which criteria should guide the selection of exercises so that the system strengthens both performance and motivation, and which control mechanisms would you implement for quality assurance? In your answer, you should consider data protection, fairness, and instructional oversight.

Second exercise: A university is considering introducing AI-supported automated assessment for submissions to reduce grading times. What risks do you see for the validity of exams and for students' feedback behavior? How could you combine automated assessment with human review to reduce these risks?

Third exercise: Think of competencies that are currently underrepresented in schools and universities but seem important in an AI-pervasive everyday life (e.g., data literacy, dealing with uncertainty, prompt and system understanding). What would a curriculum look like that systematically promotes these competencies without cutting classic subject content without replacement?

In summary: the current scientific and policy consensus views AI in education as an opportunity for individualization and scaling of feedback, coupled with significant requirements for governance, data protection, teacher training, and equitable access. Long-term effects are plausible in some areas but not comprehensively empirically established everywhere; therefore accompanying evaluations, transparent standards, and a strong role for pedagogy as a normative authority are central (UNESCO, 2021; Stanford AI100, 2016; OECD/European Commission, 2019; WEF, 2020).