The AKILAS (Adaptiver KI-Assistent für die Schule) project aims at developing an adaptive AI-based learning assistant for German-language school children which will help teachers select and evaluate study exercises for students according to their individual needs. Funded by BMBF and running from 2021 to 2024, the project brings together the fields of NLP, AI and Education Sciences and is carried out by the University of Potsdam and the the University of Magdeburg, in co-operation with solocode GmbH in Berlin.
Intelligent tutoring systems (ITS) have been studied since the 1970s. In addition to multiple-choice exercises, which are easier to process automatically, many “classical” English-language ITS have also included the automatic evaluation of free text answers from students using NLP methods, e.g. by computing the semantic similarities between students’ answers and a pre-defined set of reference answers. In addition, ITS interact with students using natural language, give feedback, provide explanations and ask follow-up questions in a manner similar to human instructors.
Moreover, the aspect of adaptivity has gained increasing interest in recent years. From an educational science point of view, it is highly desirable that teachers teach according to each students’ individual learning patterns and needs. In practice, of course, it is impossible to provide a teaching staff member to each student for personalised instruction. Adaptive ITS attempt to narrow this gap by using AI methods to provide e-learning with the possibility of catering for individual students' learning paths and needs. Not only will an adaptive ITS provide formative and targeted feedback to students as part of its automatic task evaluation module, it will also suggest suitable study material to individual students based on its knowedge of students' progress and learning habits. As an assistant to teachers, adaptive ITS systems make it possible for a single teacher to provide personalised instruction and supervision to multiple students in heterogeneous groups.
Three research groups from the universities of Potsdam and Magdeburg are working on the following tasks:
- 1) The NLP work packages focus on the evaluation of students’ free text answers and the generation of adaptive, targeted feedback. (University of Potsdam)
- 2) The AI work packages focus on the generation of learner types through privacy-preserving machine learning. (University of Magdeburg)
- 3) The education science work packages focus on the empirical evaluation of ITS systems with regard to its overall benefit to teaching and learning. (University of Potsdam)
Additional insights from an industrial perspective are provided by solocode GmbH, who is an experienced provider of digital learning apps for German schools.
In developing our ITS components, we pay attention to the relevant privacy data protection requirements. Moreover, the AKILAS project maintains continuous exchanges with external experts from our ELSI advisory board, where ethical questions for implementing ITS systems are being discussed.
Research Goals for our subproject (NLP)
- Using the latest NLP technologies, the ITS evaluates students’ free text responses and recognises systematic misconceptions and knowledge gaps.
- For the responses, the ITS provides targeted feedback to students in natural language. Taking into consideration insights from the learner types generated by the AI group as well as from the education scientistis research on effective feedback in e-learning, the system will adapt to individual students’ linguistic preferences in terms of formality, structural complexity, jargon terminology, etc.
Prof. Dr. Manfred Stede
Our lab is responsible for the NLP work packages as well as the overall project coordination.
Prof. Dr. Katrin Böhme Inklusionspädagogik – Förderschwerpunkt Sprache, Universität Potsdam
Prof. Dr. Sebastian Stober Artificial Intelligence Lab, Otto-von-Guericke-Universität Magdeburg
AKILAS runs from February 2021 to January 2024.
- Xiaoyu Bai and Manfred Stede. A survey of current machine learning approaches to student free-text evaluation for intelligent tutoring. International Journal of Artificial Intelligence in Education, pages 1–39, 2022. URL: https://link.springer.com/article/10.1007/s40593-022-00323-0. [Bibtex]