How to use AI as a teacher

When we use the term "AI" on this page, we primarily mean artificial intelligence based on large language models, such as OpenAI's ChatGPT and similar tools. If you want to use AI tools for teaching purposes, you should use GPT UiO for privacy reasons. In order to use AI in a fruitful and responsible way, it is crucial that you familiarize yourself with the possibilities and limitations such tools provide. This means that you should have familiarized yourself with ethics and social responsibility, as well as legal perspectives before you or your students use the tool. The students also have their own information page about AI. There, the students are encouraged, among other things, to inquire with you about acceptable AI use in your subject. In "recommended resources", you will find selected material that can help you further understand AI. Here you will also find specific websites you can follow to stay up to date.

What are language models

Generative Pre-trained Transformer, abbreviated GPT, is a type of artificial intelligence model developed by OpenAI to generate text. The GPT model is trained on large amounts of text data and learns patterns and language so that it can generate coherent and relevant text based on what you write to it. You can read more about what language models are on the website "What is GPT UiO".

It is important to be aware that the language models are not knowledge databases. Nevertheless, with good instructions, the tool can be used in various ways linked to many domains of knowledge. However, it is crucial not to take for granted that the answer you receive is true or of sufficiently good professional quality.

Examples of AI tools

There are a number of tools that use different forms of AI to perform tasks other than what GPT UIO can do today. You should therefore be aware that students potentially use more AI tools than those offered by UiO. An example is the plus subscription to ChatGPT which gives the opportunity to both analyze and create images, do web searches, create small programs and run them to solve mathematical questions or use third-party tools to solve other tasks, such as for example looking up databases, connect to web services or perform other specialized tasks with or without AI. In addition, there are specialized AI tools in a number of fields such as video, music, images, and more. See the page Futuretools for a comprehensive overview. 

Explore the use of GPT UiO

Once you have understood the basics of the technology, the next step is to understand, test and experience how the tool can be used in your discipline. It is crucial that you put yourself in the driver's seat and have a clear idea of what you want to get out of the tool.  

If, for example, you want to use AI for teaching planning, relevant information such as the teaching's aims, content, framework and participants can help to give more precise answers than if you omit this information. In many contexts, it will also be important to share information about forms of assessment and other activities, as well as your experiences about what is important and difficult for students to understand and master in the subject.  

Prompting

When you talk to language models, it is important to be able to ask concrete questions and give tasks that contain sufficient context. This action is called prompting. Prompting is your way of giving GPT UiO instructions about what you want it to do. It can be challenging to create prompts that give you the desired response. GPT UiO bases a lot of the answers it gives on the information and context you give it. There may therefore often be a need for adjustments, corrections and follow-up questions during the conversation. This means that the better your understanding of the topic, the greater the probability that you can create useful prompts. In order for such AI tools to be perceived as useful for you, it is important to view the interaction as a dialogue. Such tools are therefore not intended to be used as a search engine. At the bottom of "recommended resources", you will find selected material that can help you understand prompting.

Tips for creating good prompts

1. Give GPT UiO a role - for example, instruct the model to behave like an expert in the field you want help with.  
2. Be specific - consider giving one instruction at a time to get more precise answers, than if you give it many tasks in the same question/task formulation.  
3. Rephrase - if you feel that the answer GPT UiO does not give correct answers or is on the side of what you are asking; try to rephrase the question by adding more details. 
4. Use examples - to sharpen the answers GPT UiO gives, consider exemplifying what you want an answer to.  
5. Context - define the context of the conversation. For example, you can ask GPT UiO to create a case assignment based on a given topic. You must then make sure to explain at which level the case is to be solved - eg BA/MA, the teaching context, how long the students will spend, length of the case and which theories you think should form the basis of a proposed solution.   
6. Think step by step - language models generate the most precise answers if you instruct it to explain step by step.   
7. Ask for help in asking questions - you can ask GPT UiO for advice on how to ask more precise questions. In many cases, it will ask for more details related to the question you have asked, which can help you get a more precise response.  
8. Get tips for areas of use - you can also start by writing the job title and your tasks, and then ask what GPT UiO can help you with. You will then receive a list of suggestions for possible areas of use that you can use as a starting point for further conversation.  
9. Test again in a new thread - When you test the same prompt in a new thread, you may find that the answers vary, since AI can generate different perspectives each time. That is, the "memory" of GPT UIO is limited to the individual conversation thread you are in. By using multiple conversation threads, you can explore different response styles and points of view, and the responses can be perceived as more relevant. In addition, one can avoid potential confusion from previous context, if the thread of conversation was long.

Assessment activities

AI re-actualizes the importance of developing abilities such as critical reflection, independent thinking, creativity, assessment skills, nuance, as well as the ability to wonder and ask questions. These have always been important goals in higher education. There may be good reasons for allowing the use of AI in student work to be assessed. At the same time, it can represent a challenge to assess what the students have actually learned. This means that we have to discuss which forms of assessment are suitable for bringing out such abilities in your subjects. What skills and knowledge are you going to test the students on? How do you know if the students have learned what they are supposed to?  

Exam answers and cheating?  

There are no tools that can reliably identify AI-generated text in an answer. Although it is claimed that some tools can, AI is developing so quickly that such a "plagiarism check" is continuously out of date, and therefore cannot be considered reliable. In addition, there are always ways to fool the plagiarism check. 

Our recommendation is therefore to focus on developing forms of assessment where the main focus is on the process of learning, as opposed to only focusing on the final product. Such forms of assessment are important for the entire course design, and are in line with what the research highlights as strengthening for learning. UiO's regulations allow for many different forms of assessment, so there are options. Below we have listed some suggestions, and you can find inspiration in the section "learning activities with AI?". In some cases, there may be a need for a school examination, either during a transition period, or as a continuous form of examination based on the learning objectives that have been chosen.  

UiO's examination regulations focus on the student's answer being their own work. With AI, this compliance principle is challenged, and clear guidelines at the faculty are desired. There are many subject-specific assessments that must be made. For more perspectives, the AI ethics page is recommended. 

Under "recommended resources", you will find selected resources that can support you in the development of assessment and learning activities.  

If you allow the use of AI  

• Provide clear guidelines for the use of AI, and emphasize the importance of taking privacy, copyright and ethics into account  
• Clearly communicate the requirements for academic integrity, accuracy and use of sources to the students  
• Clearly communicate if the use of AI is expected or only a possibility  
• Give students insight into risks of using AI 
• Require transparency. That is, description and documentation of the process. See what we write about this on the AI student page.   
• Require reflection on the process and use of the tool 
• Require use of adequate sources  
• Give the students training in the use of the tool, either as physical training or reference to relevant user documentation.  
• Clear examiner guidance that explicitly addresses the use of AI  

Proposals for assessment activities  

When designing assessment activities, it is important that the tasks the students are given cannot simply be solved by AI alone. Draw on experiences and reflections that students do or have done into the assessment activities you design. As mentioned earlier, in several cases it will be beneficial if the students also visualize how they have worked with the assignments, by describing their reflections during the process and what they have learned.

Connect tasks to concrete work or experiences  

• Connect the task to concrete work in teaching. For example, you can work on one or more cases in class, and questions can be linked to this specific work. The student work with cases can be included in the assessment basis, for example through a portfolio assessment.  
• Connect assignments to student field practice experiences.  
• Connect to experiences. Let the students explain and reflect on process execution, learning activities and problem solving.   
• Assignments that ask for the student's own views. Give tasks that involve analyzing complex issues, assessing alternative solutions and arguing their own position.  
• Assignment formulations linked to the syllabus. Create complex task formulations specifically linked to the syllabus, which require critical reflection. For example, students can be asked to use the syllabus to highlight actual/practical/current issues.   
• Assess and/or compare different texts using the syllabus. This can, for example, be the work of fellow students, fictitious student answers, published articles or texts produced by GPT UiO.   
• Have the students create something new. Give tasks without predefined answers and encourage originality and creativity. For example, students can be tasked with developing and justifying new research questions or justifying arguments, with references to specific literature.  

Iterative forms of examination 

These are examination forms that are divided into several steps over a longer period of time, with feedback rounds and opportunities for the student to improve their own work.  

• Mutual assessment. Fellow students give and receive feedback on each other's work along the way and improve their work before final delivery. During the process, AI can contribute feedback as a "fellow student".  
• Folder assessment. The students must work in several steps before delivering an academic final product. AI can or should be used as a supervisor in the process.  
• Project work. Parts of the work take place in class, and the students receive feedback along the way to improve their work. This work may contain submissions/presentations along the way.  
• Logs and reflection notes. To make the students' thought processes visible, they can submit logs or reflection notes in which they explain their use of GPT UiO. Here it is important to bring out the reflections and assessments they have made at each step. Examples of questions can be
  • How did you use AI to arrive at the result?  
  • What kind of prompts/instructions did you use?  
  • Which prior knowledge and skills did you notice were important and useful in the process?   
  • What have you learned from the process? What experiences do you take with you from the chat you had with GPT UiO, and why?   

Multimodal forms: More than words  

• Oral forms. For example, video reflection, presentation and/or conversations.  
• Multimodal responses. As part of the answer, students can develop content in formats other than text, for example video reflections, video essays, podcasts, presentations, interviews, visual representations.  
• Multimodal tasks. Create tasks that consist of different formats in addition to text. However, Chat GPT-4 is capable of interpreting and producing some visual material. Development is rapid; as of now GPT UiO cannot do this, but that may change in the future.   
• Video case exam. The students watch a video that thematizes a topic from the syllabus. The students are asked to analyze the video with the help of the syllabus. 

Learning activities with AI?

In higher education, reading original sources, writing one's own texts and independent  

exploration and problem-solving are essential in order to be initiated into one's subject tradition and knowledge practices, learning, reflection and critical thinking. There may therefore be good reasons to include such activities in a way that is fully or partially shielded from AI, but there may also be good reasons to include AI in such activities.  

The connections between learning outcomes, assessment forms and learning activities should be communicated in a way that enables the students to think about their own learning. Such awareness of one's own thinking and learning is called metacognition. If you want to use AI tools, you should include AI in such a way that it strengthens the students' learning and metacognition, and at the same time their understanding of the technology and how it can be used appropriately in the learning process. We recommend that the learning activities are designed as an iterative process, where the students document any dialogue with AI, and justify the steps and assessments they make. In addition, the students should work together so that they can discuss and practice evaluating the AI-generated content. 

At the bottom of "recommended resources", you will find selected material that can support you in the development of assessment and learning activities.    

What roles can AI have in teaching 

In line with the development of AI, new areas of use are constantly giving new areas of use you can consider using in your work with teaching. Important mental work is of course something you can't dismiss when it comes to planning activities and teaching, formulating assignments and the like. You as a teacher know what the students will learn. Therefore, it is crucial that you are in the driver's seat and have clear ideas about what the students will learn from your teaching. What are the central issues, concepts and sources in the subject? What prior knowledge do the students have, and what might be challenging for them to understand and master in the subject? You may also include this in your instructions to GPT UiO. In this way, cooperation with GPT UiO can be awareness-raising for you as a teacher. Remember, it is you as a teacher who are responsible for ensuring the quality of the AI generated content.  

Here are some examples of tasks you can use as a starting point to explore the GPT UiO

• Development and testing of assignments and cases  
• Development of multiple-choice questions or other quiz formats 
• Development of learning activities  
• Development of presentations and scripts  
• Translations  

Suggestions for learning activities without AI  

When you are with your students in the classroom, you can set conditions for how the learning activities should be carried out, and whether they should be done with or without KI. When the students are at home, you do not have the opportunity to control this. Therefore, you should consider giving students assignments and learning activities, AI cannot easily solve. In both cases, it is crucial to communicate to the students the purpose and how the activities are intended to contribute to their learning. This kind of metacommunication about your expectations helps students understand your learning intentions. This benefits their metacognitive abilities.  

Activities in the classroom  

• Writing activities  
• Peer assessment  
• Reading activities  
• Discussion activities  
• Problem solving  
• Case assignments  
• Lab activities  

Activities both inside and outside the classroom  

These are activities where AI cannot easily generate the central parts of the answer  

• Tasks which are complex and specifically connected to the students' own experiences and reflections, for example from an excursion, a seminar, a lab exercise,    
• Step-by-step assignment with feedback from fellow students and discussions along the way  
• You can find more examples and inspiration for learning activities in the university pedagogical online resource BetterTeaching, see especially the modules "Teaching planning" and "Assessment and feedback".  

Proposals for learning activities with AI 

GPT UiO can be used as a learning partner as a way for the students to have an ongoing, scaffolding, academic dialogue with the tool. The purpose of such learning activities is for the students, and preferably together with fellow students, to build on or create ideas together with GPT UiO, or for GPT UiO to ask questions in such a way that the students have to reflect on outstanding knowledge and learning processes.    

Such learning activities can be

• Devil's advocate: GPT UiO is asked for counterarguments to a proposed solution or a series of arguments.  
• Idea generator: GPT UiO can be used as a partner to generate ideas, for example for projects or issues for discussion.  
• Questioner: GPT UiO is asked to give the student one question at a time related to a subject they want to be tested on - like a live test/quiz. GPT can give feedback on the answers the students give.  
• Feedback task: Group assignment with GPT UiO where the students will receive feedback from GPT UiO on subject texts they have developed, or other texts they upload. The students discuss improvement potentials in the texts, preferably based on assessment criteria and/or examiner guidelines.  
• Critical assessment of GPT-generated content: Content generated with the use of GPT UiO must always be quality-reviewed using credible sources. It can also be useful for students to compare GPT-generated text with other subject texts. Students learn by comparing the academic basis against the AI-generated content.   
• Co-producer in creating learning activities: As part of the teaching, the student can be tasked with creating learning activities together with GPT UiO and fellow students. For example, case assignments, discussion assignments, flashcards, multiple choice questions, board games etc.  

On the AI student page you will find examples of how students themselves can use KI as learning support.