Part of the syllabus of Bioinformatics and Simulation of Bioprocesses (BSBP), a subject of the Industrial and Environmental Biotechnology Master at Complutense University of Madrid, is a brief introduction to Python programming. Most students have never been exposed to code development before and find elementary theoretical concepts such as loops and conditionals relatively easy to grasp, but hard to apply in practice to novel problems they have never faced previously. Furthermore, their background is mostly in biosciences, but many have not taken any computational biology or bioinformatics subjects. To worsen matters, BSBP requires that the student's new knowledge of Python be immediately used to solve certain complex tasks of Biorreactor design, such as parameter estimation. We have similar concerns regarding related subjects, particularly the Integrated Laboratory of Biophysics and Bioinformatics (ILBB), of the Biochemistry Degree. In this context, and with the advent of artificial intelligence (AI), many students are now trying to alleviate their frustration with these subjects by resorting to code-generating automatic tools, but their lack of experience hampers their appreciation of the quality of AI responses.

We have started a new approach to BSBP consisting in handing out a collection of simple, systematic programming exercises that are not scored, but just for practice. We randomly solve some of them in the classroom, starting a discussion about possible solutions and comparing the students' with ours. In addition, after some time, we publish detailed comments for all of them. We found that this strategy is very well received, so we are considering using it to introduce AI training in the subject. Our goal is not to discourage the use of AI, but to provide students with some critical understanding of the capabilities and limitations of the technology, by comparing AI and human solutions, both to whole exercises and to a breakdown in manageable pieces.

In order to produce problem statements that are more suitable for AI, we organized a mixed working group, including former students of ILBB, junior researchers in bioinformatics and some teachers, to reformulate the exercises in our collection and to break them into appropriate pieces. Our goal is to train students in analytical skills, so that they realize how AI responses become better when they request more specific, atomic and unambiguous tasks. Before actually bringing this idea to the classroom, we wanted to test it within our working group. Here we report on our experience trying to get more accurate responses from several AI Large Language Models for biochemistry- and biotechnology-oriented Python exercises.

Keywords: Biochemistry, bioinformatics, computational biology, critical thinking, self-learning, motivation.