Together with Bielefeld University and LMU Munich, the research project DeepWrite and the start-up KlausurenKIste have advanced the digitalization of grading in legal examinations. The digital statute book developed by the start-up LexMea was also used in the process.
The pilot project, initiated by Prof. Marie Herberger, offered students at Bielefeld University the opportunity for the first time in the winter semester 2024/25 to complete a fully digital law exam as part of a mock examination in enforcement law. The exam was written on a computer using a digital statute book and was graded and assessed not only through traditional human marking but also with the assistance of AI.
In the summer semester 2025, this initiative was expanded at LMU Munich to include a mock examination in administrative law. In the winter semester 2025/26, the same opportunity was also introduced at University of Potsdam.
The aim of the pilot project was to explore the digitalization of legal examinations and the associated challenges related to grading, feedback, and assessment within the legal context.
In the virtual university course “Administrative Procedural Law,” ("Verwaltungsprozessrecht") offered by Professor Pache at the University of Würzburg, students were given the opportunity from the summer semester 2025 onward to participate in a previously unique teaching format: a law exam with AI-supported feedback. The project was implemented in cooperation with the interdisciplinary research project DeepWrite and was carried out במסגרת the vhb special funding initiative "AI meets vhb".
The aim of the project was to further develop legal education through the targeted use of artificial intelligence (AI) and to explore new forms of individualized support. On a voluntary basis, students were able to submit an exam in administrative or administrative procedural law, which was then evaluated by a current GPT language model. The students received detailed feedback on both the formal structure and the substantive legal reasoning of their written opinion, supplemented by an AI-based assessment.
The focus was not only on individual learning support, but also on the systematic testing and academic evaluation of the potential applications of AI-assisted grading tools in legal education.
The Hochschule für den öffentlichen Dienst in Bayern (HföD) in Hof, in cooperation with the research project DeepWrite, offered several AI-supported examinations within its Diploma program in Public Administration (Diplom-Verwaltungswirt/in (FH)).
The intensive collaboration began in the summer semester 2024 with the so-called “Compliant Teamwork Exam,” a practice assignment that students were able to complete collaboratively via the HföD’s learning management system ILIAS. In the winter semester 2024/25, the initiative was expanded to include an additional digital exam completed individually.
The focus was on AI-generated feedback for legal exams written by first-semester students. Before submitting their exams, students had the opportunity, on a voluntary basis, to have their written opinions reviewed by current GPT models and to revise their answers accordingly. They received both formal and substantive feedback from the AI and, depending on the specific exam, also an AI-based assessment.
The pilot project also included a comprehensive field test and an evaluation of the AI-supported grading assistance.
The FAU Erlangen-Nürnberg, in cooperation with the research project DeepWrite, offered an online course on examination techniques in regard to the "Gutachtenstil" via the Virtuelle Hochschule Bayern (vhb). From the start of the winter semester 2023/24, the use of artificial intelligence (AI) for grading short legal cases in the opinion-writing style for first-year students was tested in this course.
In the online course, students had the opportunity, on a voluntary basis, to have their texts reviewed by ChatGPT-4 and iteratively improved with AI support before applying the legal opinion style in actual exams. The AI was specifically intended to point out missing components and suggest linguistic and stylistic improvements.
The pilot project also included an evaluation of the AI-assisted grading and a comparison between feedback provided by human tutors and that generated by the AI.
In a keynote lecture at the AI & Law conference hosted by recode.law, Dr. Martin Zwickel from FAU Erlangen-Nürnbergdiscussed the use of ChatGPT in legal education and the collaborative project.
