FCH30806 Advanced Food Chemistry: See the study handbook for learning outcomes and further course details

• Master
• Number of students: 93

The usage of AI chatbots in this course was mainly research driven: to investigate whether students find the use of AI chatbots for setting up experimental designs useful. In case they do, this use could inspire students during brainstorming and the designing phase for their experiments. The idea was that this application would lead to more complete experimental designs, with blanks and controls that some students might not have thought of themselves.

In this course, students were allowed (but not obligated) to use AI chatbots for creating an experimental design. In a lecture at the start of the course, they were provided with information regarding AI chatbots. Some options listed that could support them with, were to give ideas for a research setup (including blanks and controls), or to find more background information. The lecture included some guidance on how to design good prompts, and information on a few AI chatbot options (ChatGPT and Perplexity) and their pros and cons. Some examples of good prompts that were provided in the lecture were:

Prompt 1. I have this research question: What is the effect of pH and temperature on pectin modification by PME? What is a correct experimental setup, including blanks and controls, to test this?

Prompt 2. Can you specify which pH range and temperature range are relevant for modification of apple pectin?

Regarding the chatbot options, Perplexity was preferred, since it provides sources, but it was not mandatory to use a specific tool. Lastly, the lecture contained a reiteration of the exam boards’ rules. It was made clear that AI chatbots were allowed to be used as one tool for pre-laboratory work, not as the only source, and not for other tasks (e.g. report writing).

At the end of the course, students were asked to fill in a survey on their experiences using AI chatbots. This included questions on whether they used AI in the assignment, what their opinion of it was, whether it was useful, and whether they trusted the correctness of its output.

54% of survey respondents indicated they used AI chatbots. To the question on whether it was useful for setting up a research question and hypothesis, these users gave an average 3.4 (+/- 1) out of 5; they gave an 3.3 (+/-1) out of 5 for whether it was useful for setting up an experimental design – this seems to be a moderately positive evaluation. Through conversations with students, it became clear that when the course ran last year the workload was high, and students often had to work into the night. This happened a lot less this year. Though, because other changes were also made in the meantime, it’s not possible to attribute that (fully) to the use of AI. Evaluation is still needed on whether this assignment will remain the same next year.

The most important tip is to test the use of AI chatbots in your assignment yourself beforehand. In this case, tests were conducted on what Perplexity and ChatGPT produce if you prompt them for an experimental setup. This testing gives you as lecturer more insight into which problems students could run into, it shows you what (not) to ask, and subsequently you can provide your students with clearer (prompting) instructions.

Secondly, don’t give students too much freedom. Ensure there’s a clear distinction between what is allowed and what not. This will decrease the students’ fear of cheating/accidental plagiarism.

Building upon that, make sure that you give students clear instructions including concrete examples. Don’t just say ‘be careful, AI might hallucinate’ – instead show an example of this happening. Explain the possibilities of different tools, warn students about their possible downsides, and let them know how to check and deal with the tools’ output.

Lastly, it’s important that students know AI is only one tool, and more importantly an external one. The course materials are already very relevant and specific to what they need, and should thus never (fully) be replaced with AI.

Julia Diederen and Bake de Rink (researchers) and [email protected]