New framework sets tougher standard for responsible AI use in education

A new study published in Systems warns that the careless use of artificial intelligence (AI) and digital technology could undermine learning, weaken independent thinking and expose students to ethical risks unless educators and course designers are trained to use them competently.

The study, titled A Theoretical Framework and Evaluation Instrument for the Competence to Use Artificial Intelligence and Digital Technology in Educational Processes, introduces a new competence model for the responsible educational use of AI and digital technology, known as AIEDTEC competence, and validates a short questionnaire designed to evaluate courses that teach this competence.

New model defines what competent AI use in education requires

The researchers define AIEDTEC competence as the ability to acquire and understand technological and pedagogical knowledge in a way that allows individuals to apply, analyze, evaluate and create educational processes using AI and digital technology.

The model is built on four levels:

1. Systems thinking: This is the mindset needed to understand AI-supported education as a complex system, not as a set of separate tools. In a classroom or digital learning platform, learners, teachers, content, software, data, feedback systems and AI models interact. The effect of any one technology depends on how it fits into that wider environment. This systems perspective is a key departure from narrower digital literacy models. The authors argue that learning environments involving AI must be evaluated as interconnected systems. A tool that looks technically advanced may still fail if it does not support cognition, motivation, feedback, ethics or instructional goals.
2. Relevant knowledge areas: The researchers adapt the well-known technological pedagogical content knowledge model, often used in teacher education, but narrow it toward the design and evaluation of AI and digital technology in education. They emphasize pedagogical knowledge (includes didactic principles, instructional design, human information processing and research methods), technological knowledge (covers computational processes, algorithms, data structures, machine learning and software implementation), technological pedagogical knowledge (connects these fields, showing how AI and digital tools can be used to support learning), and technological pedagogical content knowledge (applies that understanding to specific subjects or educational domains).
3. The third level uses Bloom's taxonomy to define the cognitive processes needed for competence. The model does not stop at remembering facts about AI or digital tools. It requires understanding, applying, analyzing, evaluating and creating. In practice, this means a competent user should be able not only to describe an AI tool, but also to judge whether it supports a learning goal, identify risks, design better learning processes and evaluate outcomes.
4. Development over time: The model proposes four phases. Learners first acquire basic pedagogical and technological knowledge, then apply and analyze that knowledge, then integrate pedagogy and technology, and finally evaluate and create AI-supported learning environments. The authors argue that these phases build on one another and cannot simply be skipped.

The model implies that AI education should not begin and end with showing students how to use a chatbot. It should build toward the ability to design, evaluate and improve AI-supported learning systems with attention to evidence, ethics and practical use.

Framework shifts focus from AI literacy to responsible educational design

How is AIEDTEC competence different from broader AI literacy? General AI literacy often focuses on understanding what AI is, how it works and how it affects society. The proposed model is more specific. It asks what competence is needed to use AI and digital technology in educational processes in ways that are effective, safe and beneficial.

Education is not a neutral setting for technology deployment. AI tools influence how learners think, practice, receive feedback and interact with knowledge. If such tools are poorly designed or used without reflection, they may reduce rather than improve learning.

The authors also differentiate their model from teacher-focused frameworks. While teachers remain central to classroom technology use, AI-supported education increasingly involves multiple actors, including course designers, software developers, learning scientists, data specialists and institutional decision-makers. The framework is therefore intended for a broader group involved in creating and evaluating learning environments.

The model also has practical value for curriculum design. It provides a structure for courses and study programs that train people to work at the intersection of education and digital technology. The authors report that the framework has already been used as the theoretical basis for a master's program on AI and digital technology in learning and instruction at Goethe University Frankfurt.

The model's focus on systems thinking also responds to one of the biggest weaknesses in current AI education debates. Much of the public conversation focuses on individual tools: whether students should use ChatGPT, whether teachers should permit AI-generated drafts, or whether software can automate grading. The framework shifts the question toward design: how AI, pedagogy, learners, content and assessment fit together.

This approach could help institutions avoid both extremes in the AI debate. It rejects uncritical enthusiasm that treats AI as an automatic solution. It also avoids blanket resistance that treats AI use as inherently harmful. Instead, it argues for competence: the ability to make informed, evidence-based and ethical decisions about when and how technology should be used.

Ethics is built into the higher stages of the framework. Learners who reach the advanced phases should be able to evaluate the limits of AI-supported learning environments, discuss risks, and create responsible applications. This includes attention to fairness, trustworthiness, privacy and the educational consequences of automation.

Evaluation tool shows strong early results

In addition to the theoretical framework, the study introduces a questionnaire to evaluate courses that aim to teach AIEDTEC competence. The instrument is not designed to measure each learner's full competence directly. Instead, it assesses whether courses are effective learning opportunities for building that competence.

The questionnaire is based on the Kirkpatrick model for training evaluation, focusing on three dimensions: reaction, learning and behavior. Reaction refers to how learners perceive the course, including whether learning goals were clear, whether the course was enjoyable, and whether AI or digital tools used in the course were useful and trustworthy. Learning refers to perceived knowledge gain. Behavior refers to whether learners engage with the content beyond the course and expect to use it practically.

The researchers tested the questionnaire with 240 university students across 18 psychology and teacher-training courses dealing with AI and technology in educational processes. The results showed good to very good psychometric quality. The overall scale produced strong reliability, and the factor analysis supported the intended three-dimensional structure.

The questionnaire is short, which is important for practical use in real courses. Long evaluation instruments can disrupt teaching or reduce participation. A brief tool makes it easier for universities, training programs and professional development providers to evaluate whether their courses are helping learners build the foundations needed for responsible AI use in education.

The study's empirical results showed that the questionnaire can be used either as a single overall score or across three separate dimensions. That gives institutions flexibility. A course designer may want a broad evaluation of whether a program is working, while a faculty development team may want to know whether a course is strong in learner reaction but weaker in encouraging practical behavior beyond the classroom.

The authors note that future work should go further. The questionnaire evaluates courses, but a separate competence test would be needed to measure individuals' AIEDTEC competence directly. Future research should also examine whether the proposed four-phase development model is supported by long-term empirical evidence.

Currently, schools and universities are under pressure to respond quickly to AI. Many institutions are drafting policies, building AI literacy modules or adding tool-based training. The framework suggests that these efforts will be incomplete unless they combine technical knowledge, pedagogical knowledge, cognitive development, systems thinking and ethics.

The research also challenges institutions to rethink how they prepare students and educators for AI-rich learning environments. Training people to use AI prompts is not enough. The deeper need is to understand how AI changes learning processes, how technology should be aligned with educational goals, and how risks should be anticipated before tools are deployed at scale.

On the whole, the study contends that AI can support powerful new forms of learning, but only when educational systems develop the competence to use it responsibly. Without that competence, digital transformation may produce tools that are impressive but pedagogically weak, ethically risky or harmful to learners.