Your Critical AI Toolkit
Part 3: Developing your skills to use AI
How to effecitvely communicate with an AI tool
/prod01/yorksjacuk/media/content-assets/safe-images/1600-x-1000/Blended-learning-teams-call-1600x1000.jpg)
There are lots of skills and knowledges you need when using AI.
In Part 1 of this toolkit, we explored knowledge and skills for what AI is and how to identify it, and in Part 2 we explored critical questions and knowledge to help you understand and evaluate AI's possibilities and limitations. These are all key components to effective decision making around using AI.
In Part 3 of this toolkit, we will bring that knowledge and those skills together and focus on final skill of effective communication with an AI tool. For this, we are focusing on multi-purpose AI chatbots like Microsoft CoPilot, ChatGPT, and Gemini.
Communicating what you need
If you have a task that needs completing by someone else, or you need some help with a task you are working on, you need to give instructions.
Task: What makes a good set of instructions to complete a task? Reflect on assignments you have had, objectives you've been given at work, or tasks you have at home, and make a list of when you have experienced effective and less effective instructions. Describe what was in those instructions.
Your list of instruction components may include:
- Clear verbs stating the action (imperative verb statements are common, such as 'Make me …', 'Design a new …', 'Create two new …')
- A list of things to include, such as ingredients, features, or parts
- Constraints that say what not to do
- Connectives that say what order to do things in (first, next, then)
Depending on your audience, your instructions may be more formal or informal, more polite or impolite.
Critical thinking questions
- How confident are you at giving instructions?
- Is breaking down tasks into their components something you find easy or challenging?
- What opportunities at university, at work, or at home, do you have to develop your instruction skills?
Questions for those who teach, support and/or assess learning
- What opportunities do you have in your curriculum for students to develop their communication skills in breaking down tasks, giving instruction, or delegating, for instance?
- How might a competency in communicating instructions be relevant in a range of contexts beyond AI?
- Could AI be used to test and improve your assignment instructions or assessment marking criteria?
Basic prompt engineering
When talking to an AI chatbot, the words we use form instructions. We call these prompts.
Task: Watch Microsoft's 'AI in a Minute: Prompt Engineering' and find out:
1. What is prompt engineering?
Critical thinking questions
- How confident do you feel to explain prompt engineering to others?
- For each task you use AI for, how successful is your prompting? Is it helping you to complete tasks more quickly and efficiently, or taking you more time with prompting and evaluating?
- How could you observe over time if an AI tool is changing how it responds to you? What do you think it is learning about you from the prompt you have given it?
- To what extent are you experiencing any of the limitations or ethical concerns during your prompting and evaluation? Can you use your prompts to take the limitations or ethical concerns into account?
Questions for those who teach, support and/or assess learning
- Can you offer prompts to your students to get them started in using Generative AI in a way that is appropriate to your contexts?
Improving prompt engineering
Providing an AI chatbot with a prompt built from a role, task/instruction, and some context is a good foundation. Prompt engineering is a skill that can be developed further and enhanced from that foundation.
Prompt frameworks are suggested ways of structuring your prompts to optimise the results. Think of them like templates for what to include and in what order.
Task: Use a Google search to search for 'AI prompt framework'. Look at the AI summary it provides. How many prompt frameworks are listed? Do any look similar to the basic prompt components we have looked at?
There are many different prompt frameworks because AI is growing and evolving, and AI engineers have different ideas about the best way to write prompts. People who develop prompt frameworks are doing the same kinds of evaluation that you have been learning in this toolkit: testing and then reflecting on what a tool has responded, why it might have responded in that way, and then refining prompts again.
Example prompt frameworks
Task: Watch RISEN Framework
A RISEN prompt should contain the following components:
| Prompt component | What is this for? | Example |
|---|---|---|
| Role | To tell the AI chatbot what persona it needs to take. Think of this like a job - you are telling the AI tool who you want it to be. | You are a notetaker. |
| Input | To provide the AI chatbot with a question, statement, or other information. | Summarise this meeting transcript. |
| Steps | To provide the AI chatbot with the instructions to complete the task. | Start with a list of attendees, then summarise the main key discussion points, and then provide an action list at the end. |
| Expectation | To provide the AI chatbot with a clear outcome - what do you want it to produce at the end? | Produce a concise summary that is formal and suitable for use in a business context. |
| Narrowing | To provide the AI chatbot with any limits or restrictions. | The summary should be a maximum of 500 words. |
Completed prompt:
"You are a notetaker. Summarise this meeting transcript. Start with a list of attendees, then summarise the main key discussion points, and then provide an action list at the end. Produce a concise summary that is formal and suitable for use in a business context. The summary should be a maximum of 500 words."
Task: Read this summary of the CREST Framework by the AI Literacy Institute.
A CREST prompt should contain the following components:
| Prompt component | What is this for? | Example |
|---|---|---|
| Context | To tell the AI chatbot background information. | You want to summarise the conversation from a business meeting about a project plan. |
| Role | To tell the AI chatbot what persona it needs to take. Think of this like a job - you are telling the AI tool who you want it to be. | You are a notetaker. |
| Example | To give the AI chatbot an example that you want it to follow. | Here's an example of the kind of summary minutes you usually write: [include example] |
| Style | To tell the AI chatbot what tone or format it should follow. | Use a professional tone. |
| Task | To tell the AI chatbot what it should do. | Write summary of minutes from this transcript and include a list of actions. |
Completed prompt:
"You want to summarise the conversation from a business meeting about a project plan. You are a notetaker. Here’s an example of the kind of summary minutes you usually write: [include example]. Use a professional tone. Write summary of minutes from this transcript and include a list of actions."
Task: Watch What Is the 5-Step Prompt Framework and How Does It Work?
A TCREI prompt should contain the following components:
| Prompt component | What is this for? | Example |
|---|---|---|
| Task | To tell the AI chatbot what you want it to do, including what role it should take and what the output should be. | You are a notetaker. Write a summary of minutes from this transcript and include a list of actions. |
| Context | To tell the AI chatbot more information about what you need. | You want to summarise the conversation from a business meeting about a project plan, using a professional tone, and include a list of actions. |
| References | To give the AI chatbot examples it should use to guide how it completes your request. | Here are examples of minutes - use the same format and style. |
Completed prompt:
"You are a notetaker. Write summary of minutes from this transcript and include a list of actions. You want to summarise the conversation from a business meeting about a project plan, using a professional tone, and include a list of actions. Here are examples of minutes - use the same format and style."
You then Evaluate the response and then Iterate (make changes to the prompt and try again).
CoT prompting includes a basic prompt, plus 'perform the task step-by-step'.
Task: Watch What Is Chain-of-Thought Prompting in Generative AI?
CoT prompting can be a powerful learning tool for you as an AI user, because it asks the AI chatbot to reveal to you its logical reasoning, or, how it has come to the response it has.
The simplest way to use CoT prompting is to give an AI chatbot a prompt and then add 'perform the task step-by-step' as the final instruction. For example:
'Identify keywords in this article abstract. Perform the task step-by-step.'
The AI chatbot will provide you a step-by-step description of what it has considered in forming a response, followed by the response. You can use this to understand more about how the AI chatbot has interpreted and processed what you asked for, and then further improve and refine your original prompt.
Critical thinking questions
- Is using a specific prompt framework important to you? For example, are any of the frameworks easier to remember, easier to work with, seem more effective?
- Are there tasks you want to use AI for that a prompt framework would help? Are there tasks where it might be challenging to write a prompt in the way a framework expects?
- Is this prompt engineering knowledge and skill valuable or important to your profession, industry, or future employers? How confident would you feel to explain the value of this knowledge and skills?
Questions for those who teach, support and/or assess learning
- Are your students using prompt frameworks already?
Summary
In Part 3, we have focused on communicating instructions, creating your foundations in prompting an AI tool, and developing and advancing those prompting skills to make your communication with AI chatbots more efficient and effective.
How you make use of these skills, along with your evaluative skills, is your choice. Use the learning in this toolkit to critically engage and make informed decisions on where AI fits into your learning, your work, and in other parts of your life.
