AI two ways

Two ways I am incorporating AI into my coming course

As with most academics, my first course of the year starts next week. There will be many small changes this year, but one bigger change is that I am both expecting, and encouraging students to embrace AI as a fundamental part of their study process. Both theory, and increasingly evidence, suggests that AI can help students learn, though this comes with many challenges to getting it right.

With that in mind there are two actions I am taking to facilitate this.

1. Prompts in the study guide

For each of my courses I produce a study guide that week-by-week tells students what I expect from them in terms of work. It includes both pre-work to do before taught sessions, and follow-up work for after sessions (both of which include watching videos I have produced). Typical activities include reading book chapters, reviewing material from lectures and workshops, doing practice tests, and working on assignments.

This year I am adding a new element in the form of suggested prompts to use with an AI chatbot. I believe that the use of this AI to seek explanations, to clarify ideas and get examples is becoming fundamental to how students learn. I expect that many students will be doing this anyway, but asking the right questions can improve results, so my aim is to get them started with some good questions.

This may sound like it could generate a lot of extra work, but I can work with AI to get the prompts in the first place. Let’s consider the first week of my new course as an example.

This is an introductory microeconomics course, similar to hundreds of others taught at universities around the world. In the first week of the course I aim to introduce students to what economics is, what differentiates it from other social sciences in its use of models, and to apply this to two examples. The first example is the classic model of demand and supply. The second model is an economic model of voting. I look at the latter as I like to demonstrate to students that economic methodology can be applied to topics other than the more obvious ones of consumption, production investment etc. Anywhere a human decision is made, economics methodology can help us understand what is going on.

I used the following prompt:

I want to suggest some prompts to my students for use with Ai chatbots such as yourself. They will be exploring introductory economics, specifically how economists work with models, supply and demand, and also the economics of voting. Can you suggest some prompts that would help them in this task?

I decided to give the new ChatGPT o1-mini model a spin for this task. It ‘thinks’ about its answers more, but does so more quickly than the main model (but sacrifices some performance for this).

Here is an excerpt of the response I got:

1. Understanding Economic Models

a. Basic Explanation and Purpose

• Prompt: "Explain what an economic model is and discuss its importance in the field of economics. Provide an example of a simple economic model and describe how it is used to analyze economic phenomena."

b. Real-World Application

• Prompt: "Choose an economic model (e.g., the circular flow model, the supply and demand model) and illustrate how it can be applied to understand a current economic issue or event."

c. Advantages and Limitations

• Prompt: "Discuss the advantages and limitations of using models in economics. How do economists ensure that their models remain relevant and accurate despite these limitations?"

d. Comparative Analysis

• Prompt: "Compare and contrast two different economic models used to analyze market structures. How do these models differ in their assumptions and outcomes?"

e. Model Assumptions

• Prompt: "Select an economic model and identify its key assumptions. How might changing one of these assumptions affect the model’s predictions?"

The response goes on to include prompts in the categories:

• Exploring Supply and Demand

• Investigating the Economics of Voting

• Integrated and Higher-Order Thinking Prompts

• Tips for Students Using These Prompts

Everything is set out very clearly, the level of unprompted extra detail is excellent. Whilst a couple of the answers were not quite what I was looking for, this was due to some vagueness in my original prompt, and can be rectified easily with a suitable request for clarification and some extra detail.

It is worth highlighting one of the ‘Integrated and Higher-Order Thinking’ prompts. Under the subtitle ‘Critical Evaluation’- the suggestion is:

"Critically evaluate the effectiveness of using economic models to predict political behavior. What are the strengths and potential pitfalls of this approach?"

I think this is an excellent example of the higher level thinking I want my students to have, and is one I have used in the study guide. It is interesting that the AI recognises that such questions are desirable, without being specifically asked.

I have repeated this process for all the topic areas in the course, and each week of the study guide now has a selection of these suggested prompts that students can use as conversation starters with the AI. In each case I did not use all of the suggested prompts, but picked my preferred ones, and some minor editing was needed in some cases.

The type of questions being suggested, are very similar to what I have experienced from students over many years of teaching this topic, and I think accurately represents the type of issues students can find difficult and that they need to explore.

In this case the prompts that are provided both help speed up the task, and in at least some cases, are genuinely better than what I would have come up with even after a substantial time thinking about it. Both speed and quality are assisted by AI in this example.

In addition to these prompts, I will provide generic prompt templates to students that help achieve extra detail, though in reality I rarely use these tricks, finding that fairly straightforward questions get the results I want most of the time. It is also worth emphasising that I won’t just leave students to get on with this on their own, we will have in-class activities working with the AI on prompting, and this skill is actually built-in to the assignment for this course (which I will write more on soon).

What if the AI gets it on wrong? This will happen on occasions, but this happens in all other media as well. Certainly I have seen mistakes in textbooks, on academic websites and in published papers, not to mention in the media, YouTube videos and from university staff! Everyone and everything is fallible, and from my experience the current generation of AI chatbots are not doing much worse than their rivals. Students should be critical and again this is something I am making fundamental to their assessment in a way that should encourage them to think carefully about their engagement with AI.

The modern learning environment should combine traditional elements such as reading, problem sets and writing, with modern elements such as videos and the use of AI chatbots. Imagine a student with a textbook open and an AI, when they get stuck or need clarification the AI can help and direct them further.

As a final point in this section, an alternative way to generate these would be from AI aggregation of all the material on the course. If you first uploaded slides, study guides, reading lists, assessments, and then asked for prompts that come from this, it should generate good results and I will experiment with this soon and report back!

2. Custom GPT

As a complement to the prompts in the study guide I have also written a GPT specifically to support the students on this module. Most of what a GPT is, is simply a structured prompt that asks it to respond in a certain style, emphasise specific things etc.

In this case I want it to be formal, but supportive, and I want explanations to link to core concepts such as opportunity cost and trade-offs which are at the core of economic thinking. It is nice to have some control over this, but the power still comes from the underlying AI model.

The real added value in the GPT is that I can attach documents in the background that give this GPT unique knowledge, that can combine powerfully with the AI raw AI output. In this case I attach the module handbook and the study guide (including the suggested prompts I have added) such that the GPT ‘knows’ the learning outcomes for the module, the assessment structure, the tasks I am asking students to perform, and the reading. With this knowledge it can give more detailed responses to student queries, linking answers to specific aspects of the course. My instructions to the GPT ask it to do this at the end of every answer¹.

I have done this exercise before, in late 2023 for another module, but my efforts went to waste for that cohort, because free subscribers to ChatGPT did not have access to GPTs. This has now changed, and free subscribers do have access, though with use limits².

Conclusion

Both of these methods are experimental, I have not done them before and will see whether they make much difference and how students feel about them. I will survey my students and can also track the metrics of how often the GPT is used, in order to assess the effectiveness of this intervention.

I know for certain that my students are almost all avid AI users but continue to need guidance on how best to use it. I want my students to be independent learners who can harness AI to help them, and I think giving them ideas on where to start their conversations, and having a customised AI that knows the details of my course, are a step in that direction. Ignoring AI, telling students not to use it, or simply leaving them to their own devices to work it out, are not good options in my mind, a guided path to AI-mastery is the path I want to take.

Good luck to everyone for the new academic year, there will be more updates from me soon!

1. I discussed my design principles for GPTs in other settings here:

2. Whilst the discussion here is quite focused on ChatGPT, it is also worth appreciating the range of free uses for modesl such as Claude: https://support.anthropic.com/en/articles/8324991-about-claude-pro-usage