AI Innovation Lab (for VET)
The Future of Testing: Where AI Meets Pedagogy.Next-Generation Assessment
Welcome to the AI Innovation Lab, your digital space for experimentation, discussion, and real-world solutions. Here, we’re turning the challenges of AI into a new standard for legally compliant assessments in vocational education and training.
The world of generative AI is transforming the educational landscape at lightning speed. We’re not waiting to see how the assessment of tomorrow will evolve—we’re building it.
“The goal is not to prevent AI from being used in exams, but to design exams so well that AI enhances them rather than undermining them.”
The lab is not a theoretical seminar. It is a space for action centered around three innovation clusters:
- Visionary Task Culture (Exam Task Innovation): AI can generate pure knowledge-based questions without any randomness. We develop task formats that challenge students’ judgment and reflective skills—designed so intelligently that AI becomes a partner rather than a “ghostwriter.”
- Intelligent Processes (Exam Administration Innovation): We are rethinking exam organization. Experience how digital tools not only digitize administration but also radically simplify it and make it more accessible.
- Precision & Workload Reduction (Exam Grading Innovation): How can AI support us in grading without undermining pedagogical autonomy? We test feedback systems that save time and enhance objectivity.
The Innovation Lab is based on the research report titled “AI-Driven Innovations in Vocational Testing: Empirical Approaches to Next-Generation Assessment through the Establishment of a Collaborative AI Innovation Lab.” The research findings serve as the foundation for integrating AI into the test development process in a targeted and resource-efficient manner.
In addition, as part of the Lab, we compile findings that we consider scientifically interesting or relevant to practice. So the Lab is growing—join us!
Three Innovation Clusters
Scope for innovation
Cluster A - Innovation in Exam Questions Smart Tasks
Cluster A analyzes exam questions from commercial vocational training in the context of the increasing use of AI. The focus is on the assumption that open-ended questions requiring extensive justification and a high degree of judgment are less susceptible to AI and are therefore particularly suitable for exams. The study examines the extent to which existing tasks can be solved with AI support and thus need to be designed to be “AI-proof” or whether they produce AI-induced errors. To this end, 102 exam tasks are analyzed across 14 AI systems using a test prompt to assess structural features, typical error patterns, and required AI competencies.
Cluster B - Innovation in Exam Administration Next Gen Admin
Generative AI has the potential to revolutionize training in business-related professions. This places new demands on trainers, particularly with regard to the need to develop and apply skills in working with this new technology. The integration of AI requires an awareness of the legal framework that applies to the use of AI and that ensures the responsible use of the technology while upholding ethical standards and data protection. In Cluster B, these issues are discussed with exam administrators in order to derive reliable conclusions about the potential of AI as well as the risks associated with its use within the context of the exam process.
Cluster C - Innovation in Exam Evaluation AI Assisted Grading
AI systems can play a valuable role in assessment processes. With this in mind, Cluster C investigates how effective AI is when used as an assessment tool. To this end, human grading is directly compared with AI-generated grading as part of the analysis. The goal is to use detailed analyses of this comparison between human and AI-generated grading to identify patterns that allow for the classification of task categories: those in which AI evaluation yields results nearly as consistent as those of human graders, and those in which this is not the case—or where the discrepancies are so significant that they distort the assessment results.
You may also be interested in...
Learn more about Cluster A The Augmentation Trap Model
Artificial intelligence promises short-term productivity gains, while at the same time potentially undermining professional expertise over the long term. The Augmentation Trap model illustrates in particular why delegation can ultimately result in a dangerous loss of expertise that cannot be replaced.
Learn more about Cluster B The Trendslop Effect
Artificial intelligence raises expectations of clear strategic guidance, while at the same time tending to reproduce popular management ideas. The Trendslop Effect explains why convincing answers can mislead companies despite appearing to provide clear and persuasive strategic guidance.
Learn more about Cluster C End of Traditional Online Tests?
Autonomous AI agents complete standardized examinations faster than participants and achieve top grades. Dead-Loop Learning illustrates why conventional online examinations may lose their credibility as evidence of competence when autonomous AI agents are able to complete them successfully.
New Research Impulses for the AI Innovation Lab
Insight Report World Economic Forum Education Readiness for the Age of AI
Learners and teachers already use AI—often faster than examination regulations, data protection frameworks, and institutional processes can be adapted. The World Economic Forum’s report of June 2026 identifies precisely this gap as a central challenge.
Its “AI Readiness Framework” outlines 16 areas for action, ranging from data governance and infrastructure to academic integrity, teacher development, assessment, and concrete learning experiences.
The report warns of cognitive dependency, misinformation, non-transparent records of performance, and the loss of human interaction. At the same time, it shows how AI can enable personalized learning, timely feedback, and reduced administrative burdens.
Why is it worth reading? The report translates the abstract debate into concrete questions for education policy and assessment practice. It is of particular relevance to Cluster B of the AI Innovation Lab because it considers responsibilities, transparent rules for the use of AI, human oversight, and institutional prerequisites together.
(June 2026)
Academic Legal Opinion with Practical Recommendations Legal Framework for AI in Vocational Education
The use of AI in vocational examinations touches on several areas of law. Article 12(1) of the German Basic Law (GG) protects occupational freedom, while Article 3(1) requires equal treatment of all examinees. The Vocational Training Act (BBiG) and the Crafts Code (Handwerksordnung) require examination institutions to assess candidates’ actual level of training validly, fairly, and with legal legitimacy.
The EU AI Act defines AI systems in Article 3. Articles 6 and 8 set requirements for risk management, transparency, and human oversight, while Article 52 provides for labeling obligations for AI-generated content. Articles 5 and 6 of the GDPR require lawful, purpose-specific, and data-minimizing processing; Article 25 requires privacy by design, and Articles 15 to 17 safeguard data subjects’ rights. Data should be protected and, where possible, processed within European infrastructures.
German copyright law and the German Civil Code (BGB) govern training data, examination materials, contracts, and liability. Responsibilities between education providers and AI providers must therefore be clearly defined.
The legal opinion distinguishes between different levels of risk. Greater criticality entails stricter requirements for documentation, auditing, conformity assessment, and human oversight. Examination regulations must transparently specify permitted AI tools while ensuring that candidates’ own work remains identifiable. AI detectors are of limited use due to statistical uncertainty and discrimination risks. Instead, context-based transfer tasks, reasoned arguments, and personal hearings are recommended.
Key message: Legally compliant AI-supported examinations require a clear legal basis, risk-appropriate procedures, data protection, and human responsibility. Assessment should consider not only the final result but also the traceable process leading to it.
(March 2026)
AI in Vocational Education and Training What Can AI Actually Achieve in Vocational Education?
Which effects of AI are empirically supported, and where does technological optimism begin? This systematic review analyzes 26 studies of AI use in vocational education and training. Intelligent simulations appear particularly promising for supporting practical skills in realistic situations. Tutoring systems and chatbots can support knowledge acquisition, task completion, and self-regulated learning.
However, the authors identify clear research gaps: many findings are based on short-term studies and self-reported assessments. Authentic workplace learning settings are almost entirely absent. Particularly revealing is the tension between learner-centered objectives and often highly directive AI systems.
Why is it worth reading? The article looks beyond the general success narrative and shows the conditions under which AI could genuinely strengthen professional judgment, reflection, and independent action. For Cluster A of the AI Innovation Lab, this provides important impulses for authentic, transfer-oriented assessment tasks.
(June 2026)
– This website is currently under construction –
