From Programming to Chatbots: Digitalization Strategies in Chemical Engineering Education

Education 4.0 marks a significant shift within the field of education by incorporating cutting-edge technologies, including artificial intelligence (AI), virtual laboratories, augmented and virtual reality, into the educational process. The main objective is to equip students to face the challenges posed by the digital transformation and thrive in an increasingly automated world. This new paradigm emphasizes interactive and personalized learning experiences that promote digital literacy, critical thinking, and creativity. It seeks to educate technology-proficient professionals capable of adapting to fast-evolving environments and fostering innovation. To address the new challenges and requirements imposed by Education
4.0, this PhD thesis aims to develop digitalization strategies to provide students with the skills required for their future professional career. Such strategies include creating programming courses tailored to (bio)chemical engineers and automating repetitive tasks through generative AI to support teachers.

Programming, especially in widely used languages such as Python, is one of the new strategies adopted in the Education 4.0 paradigm to strive for a more digitalized education. To address this need, this work proposes a structured pedagogical framework for the introduction of Python into the (bio)chemical engineering curriculum. This work focuses on the implementation of sPyCE, a series of Python courses for (bio)chemical engineers, resulting in two open-source courses: Chemical Reaction Engineering in Python and AI for Chemical Engineers. The first two weeks of all courses in sPyCE aim to introduce the basics of Python. This material has been tested and introduced in various courses and workshops. Chemical Reaction Engineering in Python was embedded in a Bachelor’s course at the Technical University of Denmark (DTU). Students found the course to be useful, however some found having to learn programming to be difficult, and the learning curve to be steep. By the end of the semester, they managed to solve complex exercises in Python, such as combining mass and energy balances and solving systems of differential equations. Although the results in the final exam did not indicate an improved
performance that can be pointed towards the introduction of Python as firstly hypothesized, students attained higher grades in their home assignments compared to previous editions of the course that did not utilize Python. This indicates that the educational resources were comprehensive and effective, allowing students to achieve the learning objectives and successfully use Python to solve the assignments. Teachers believe that incorporating programming encouraged critical thinking among students, suggesting that this might result from the requirement to make assumptions for executing the code. The second course in sPyCE, AI for Chemical Engineers, aims to teach a wide variety of AI techniques, focusing mainly on machine learning and deep learning approaches. These methods cover different areas within the field of AI, such as natural language processing, signal processing, and image processing. Moreover, the course is designed to also teach the basic workflow of data-driven modeling including data pre-processing, data analysis, and basic visualizations. sPyCE received good feedback from both students and educators, suggesting that the material is well designed and enables students to learn. Additionally, the fact that the material is well received by the community is highlighted by the large number of users saving and downloading the code from GitHub. Generally, the feedback on this initiative is more positive when Python is introduced as part of an advanced course (e.g., Master’s level course) or external workshops, rather than embedded in an already established course.

Another digitalization strategy adopted in Education 4.0 is the use of AI-powered tools. In fact, research suggest that, if applied correctly and ethically, AI could benefit both educators and students. Recent years have witnessed the rise of large language models (LLM), which are AI models trained on massive amounts of text. Their ability to extract information from text could offer a powerful tool for improving the learning experience of students. However, before embedding AI tools in an educational context, a rigorous ethical assessment should be performed. Therefore, this PhD thesis also presents an investigation of the implications of using AI in education, supported by quantitative and qualitative survey data. Among the findings, respondents believe that AI models should not replace but rather support teachers, highlighting the quality of the generated output as their main concern. They suggest that this could lead to potential time savings and the opportunity for teachers to focus on quality teaching and other important activities. Respondents also show openness to the use of these systems in an educational context. As part of this effort to digitalize education, this thesis presents two examples of how LLMs can be used to automate educational processes. The first model, ChatGMP, is developed to automate a repetitive task and provide an interactive and fun experience for the students. The chatbot has to perform a question-answering task in a mandatory interview exercise. To achieve this, two consecutive years of interviews were recorded and used as input to an LLM. In spring 2024, ChatGMP was tested in a Master’s course at DTU, where three groups of students volunteered to perform the interview with the chatbot rather than a teacher. Students managed to gather meaningful information and relevant documents to write a report of the exercise. It was well received by both students and teachers. All teachers were impressed with the responses of ChatGMP, stating that it was a successful experiment and thus it will become an integral part of the course in the next editions as well. The second example of LLMs in education investigated in this thesis is provided by FermentAI, a chatbot built to answer exam questions of a Master’s course taught at DTU. Although this model was not tested with students, the metrics used to assess its performance indicate its reliability and potential for future uses. Additionally, this work demonstrates the vast applicability of LLMs in education, which could potentially be used to solve a multitude of diverse problems. To further improve the prompt given to LLMs, this work also aims to provide a proof-of-concept of how to leverage inductive logic programming (ILP) to learn rules and user preferences.

To facilitate the discovery and accessibility of these digital strategies, some of the tools implemented, such as sPyCE and FermentAI, are embedded in BioVL, an e-learning platform developed to teach (bio)chemical processes. The main objective of this work is to provide a platform to students where they can learn the principles of (bio)processes and how to model them, as well as where they can interact with a bespoke chatbot to clarify their doubts.

Finally, to foster transparency and open-source research that is accessible to everybody, the majority of the code is released on GitHub.