You can sort the rows of the table by and of the columns by clicking on the name of the column. The search box on the top left can be helpful if you want to look for something specific, e.g., a university or a country.

Additional and Similar Resources

• The recently announced UNIDIR AI Policy Portal. Their database is currently under construction.

• Chan et al. (2023)1 on a guidance about how universities can create and implement such policies.

• A growing collection of classroom policies by individual lecturers/teachers, collected by .

• A Team at scribbr.com created this collection of policies and guidelines, specifically on AI writing tools at US universities. Here’s their article about it with overall insights.

• A similar collection about policy updates at universities in Australia, collected by TEQSA in May 2023: AAIN-Institutional-Responses-Generative-Artificial-Intelligence

When using generative AI like ChatGPT, the quality of the results depend on the input. Garbage in, garbage out. Part of that input is the data the model was trained on and has access to, which we users have no control over—apart from handing specific information to the AI directly. What we actively control which hugely influences the output are the instructions. Our questions, our prompts, our examples we provide to the model.

To help with this, Seb Dianati and Suman Laudari from Charles Darwin University in Australia created an article series about prompting generative AI specifically aimed at teachers and students.

Basics of prompting

Their first article sets the stage and introduces key ideas about prompting. For example:

• Provide context

• Give the AI a role (“Act like a …”)

• Be specific about the response format you wish (a 200-word summary, a list of 10 things, a table with the columns ‘key word’, ‘example’, ‘definition’, a table in CSV format)

They also remind us that it’s not all over after you send your prompt and receive one response. We can rerun the same prompt or edit it and then rerun it. Or, we can ask follow-up questions, refine the results by chatting with the AI and so on.

Prompts for Teaching and Assessment

The second article shares 25 applications in teaching and assessment. Among them are ideas about how to ask ChatGPT to create a grading rubric for a specific assessment (oral presentation), creating exam question, or to get support to create a video lesson.

Prompts for Administrative Tasks

In the third article we find 25 suggestions to use ChatGPT to help with administrative tasks teachers can face in their daily work. Among others there are prompts about getting help with giving feedback, proofreading, generating ideas for student activities, creating a weekly plan for courses.

Prompts for Student Engagement

The fourth article deals with ideas to support student engagement. Ideas include help with creating content for presentation slides, summarizing YouTube lecture videos, and support for decision-making based on algorithms.

I liked the idea to bundle prompt suggestions for different use cases like this article series did. Unfortunately, the individual prompt suggestions were often not that innovative and not crafted as carefully as the first article made me hope. Some ideas were very similar.

It feels that the authors lost track of their own basic suggestions about prompt writing during the later parts of the article series. Some used little context or gave no clear role. Here’s one example:

Develop learning outcomes or learning objectives: List these learning objectives as learning outcomes.

[wait for response]

Now apply Bloom’s taxonomy to develop higher-order learning outcomes.

Here’s a rewrite of the prompt adding some of the ideas that were suggested in the first article:

Act like an instructional designer / lecturer / teacher / educator for a course on introductory psychology for first semester university students. Develop specific learning objectives as learning outcomes and use Bloom’s taxonomy (classification of educational learning objectives) to develop higher-order learning outcomes. Please return a well-formatted table.

In this example I gave a role, provided context and the topic of the course the teacher is interested in creating learning outcomes for, and suggested an output format.

🔨Tools & Resources

Recent finds of education-related AI tools and resources about them.

• Khanmigo, the AI tutor of Khan Academy

• Artificial Intelligence Teaching Guide from Stanford University

References
AI use Statement: No AI was used in drafting or editing of this post.

Generative AI has been heralded as doom for any written assigments at schools and universities and consequently the end of an essential pillar in humanistic undergrad education. Both main claims are debatable — the role of written assignments for education in general and generative AI’s impact on them. However, almost a year after this conversation started, we can agree that conversational AI tools based on LLMs (like OpenAI’s ChatGPT and Google’s Bard) can create text that will pass a teacher’s requirements for a written assignment while remaining undetected as the product of AI.

When AI is used to create anything, questions of authorship and plagiarism pop up quickly. Transparency is key here. We should share how and when we used which tools to help us with the creation of something.

A healthy dose of scepticism is also helpful. Keeping the general disclaimers by OpenAI and Google in their respective tools in mind:

“ChatGPT may produce inaccurate information about people, places, or facts.”
“Bard may display inaccurate or offensive information …”

But prohibiting students from using AI tools is unrealistic. And since they’re going to use them anyway (many of them already are), we’d better familiarize ourselves with their capabilities and limitations, with the best practices of how to use them and what to avoid. And we should keep in mind that these are developing and changing constantly together with the tools themselves. That should be a goal for teachers and students alike.

Educators around the globe are experimenting with adapted assignments that allow their students to familiarize themselves with AI by allowing them to use it in certain ways help them create their texts.

Here are a few ideas about how to let students use generative AI for written assignments and ways to help to stay more transparent about the use of AI in the writing process:

• Let AI create a first draft of an assignment or of a part of it and then edit it, fact-check it, and so on.

• Let AI create an outline for their assignment, with placeholders or with topic sentences to get them started.

• Write a first draft yourself but then use AI to edit, e.g. by asking it to improve specific sentences or find more suitable terms or better examples or explanations.

• Ask AI for multiple ways to phrase something, e.g. how to explain a term, and then pick, combine, and edit to create an even better final version. For example, ask for four versions of a intro sentence/paragraph or for a sentence that should link two paragraphs.

• Create a paragraph with one tool and then let it being critiqued, graded, or improved by another tool (e.g. asking ChatGPT to grade a text created by Bard).

• Ask two tools to create a text (same/similar prompt) and compare the results. Or even compare multiple results of multiple tools. What are the essential concepts and terms used in all of them and how do these compare what you can find in a textbook or scientific publications?

• Use a writing tool that stores the full editing history, like Google docs or etherpad. The editing history might then be shared/submitted together with the end product.

• Use marking (bold or italicized font, or highlighting with different colors) to clarify what was created entirely by you, what was paraphrased from an AI output, and what was copied directly from an AI tool.

🔨Tools & Resources

Recent finds of education-related AI tools and resources about them.

• Life skills course “AI for Education” on Khan Academy

References
AI use Statement: No AI was used in drafting or editing of this post.

OpenAI recently published a blog post that showcases how some teachers are using ChatGPT, including a handful of example prompts you can use yourself.

I’ve tried one of their prompts but needed additional specifications to get it to work as intended. Their example B “Create effective explanations, examples, analogies” didn’t work directly with GPT-4 in the normal ChatGPT interface. Instead of prompting the instructed questions one by one and then handing me the requested explanations and examples, ChatGPT listed everything like a full conversation script, with placeholders like

[Wait for the teacher's responses]

I could avoid this issue by rephrasing the prompt and clarifying the idea of asking one question at a time and then waiting for the user input. I think the issue might also have had something to do with how the term “teacher” was used in the original prompt, as I assume that the model works better when we simply refer to the user who types as, well, user. We can then still specify that the user is a teacher. Doing so is also more in line with the role terminology used in OpenAI’s chat completion API, where we have the roles of user and assistant.

Here’s my rephrasing and simplification of their prompt for Create effective explanations, examples, analogies:

You are helping the user who is a teacher develop effective explanations, analogies and examples. First, greet the user and describe your role. Then ask them the following questions, one at a time. Wait for an answer each time. If needed, ask additional questions to gain more information to improve your results.

- Who are your students, what is their learning level?

- What topic or concept do you want to explain?

- How do you want to customize this teaching? 

Based on the information you received, then give the user a short explanation of the topic (max. 500 words), 2 examples, and an analogy. Keep the explanation as simple as possible without sacrificing accuracy or detail. Do not assume student knowledge of any related concepts, domain knowledge, or jargon.

Finally, ask the teacher if they would like to change or add anything to the explanation.

This prompt leads to to better results for me, the back-and-forth between user and assistant works now as planned. But the analogies returned didn’t work for me all the time. Sometimes even after several tries. So, as is usually the case, there are limitations to work with. And it’s obviously essential to look through any results and fact-check and refine them.

🔨Tools & Resources

Recent finds of education-related AI tools or resources about such tools.

• Wisdolia creates flashcards from PDFs, text input, or YouTube videos.

• Video: Introduction to AI for Teachers and Students by Wharton Interactive at University of Pennsylvania:

References

AI in Education is meant as a place to collect and highlight resources and tools, insights, ideas, and experiences about everything regarding “AI in Education”. At least everything that I come across and find interesting.

Apart from that, we’ll see what emerges from this start.