Chapter 3: Instruction

In this chapter, you will learn how to instruct a chatbot to do what you ask and how to create, through language, a productive collaboration between you and your artificial assistant.

Briefly about instruction

You may have heard the term prompt engineering before, which refers to formulating good prompts or requests for a chatbot. The term can make it sound as if instructing a chatbot is an entirely new science that requires special skills to master.

Here is the good news: you are probably already quite a skilled “prompt engineer”! The same qualities that already enable you to work with your colleagues are exactly the ones you need when you instruct a chatbot: being able to explain what you want and mean, give examples, ask follow-up questions and guide a conversation in the right direction. These are communication skills you use every day when you work with other people, and they are also the ones you need when you work with AI.

So, instruction is about being able to:

• Communicate clearly what you want from the chatbot
• Maintain an ongoing dialogue in which you guide the collaboration in the right direction and get the chatbot to act appropriately along the way
• Have a sense of how the chatbot responds to different requests – and the imagination to try new strategies

In the next modules, we will look at what these skills look like in practice.

The three Ps

In Chapter 2 on delegation, you were introduced to the three Ps: Product, process and preferences. There, we used the three Ps as a framework for planning your collaboration with the chatbot.

In the same way, you can use the three Ps when it comes to instructing your artificial assistant. You simply need to turn what you have already decided into verbal instructions that the chatbot can decode and act on. It cannot read minds – so anything you want it to know, you need to tell it yourself:

1. Product: What is the purpose of the collaboration? What are the success and quality criteria? What is the context?
2. How should the collaboration process unfold? Should the task be solved in several steps or phases? What are our different roles in the collaboration?
3. Preferences: What particular role or “personality” should the chatbot take on? How should it behave? How should it not behave?

All of this is something you can tell your chatbot to create a productive collaboration. Either all at once or as the interaction develops.

Prompting

As mentioned earlier, “prompting” well is not an exact science, but rather the ability to have a constructive conversation with your chatbot. Still, there are several basic techniques worth keeping in mind when you need to instruct a chatbot:

• Expand the context – and at minimum state the purpose of your collaboration so it understands what you need help with
• Give examples of how the chatbot should respond so it knows what you want
• Specify your criteria for what a good response looks like, so the chatbot knows, for example, your specific wishes regarding format, length and tone
• Break the collaboration into several steps so the chatbot only needs to deal with one task at a time
• Define the chatbot’s role and personality so it knows how to act
• Communicate clearly and collegially – just as you would with a colleague (the collaboration often works better that way for both parties)

These are some general techniques you can use when you instruct your personal assistant, but over time you will no doubt find your own tricks for getting it to act appropriately. Experiment your way forward.

Metaprompting

If you want to improve your collaboration with the chatbot, it can be helpful to use a particular technique known as metaprompting. Here, you ask the chatbot itself to ask follow-up questions, improve your request or offer its own suggestions for how to approach a task.

For example, you can ask the chatbot to interview you before it produces an output. That way, the chatbot asks about the context it needs in order to solve the task satisfactorily.

Iterative prompting

When you instruct a chatbot, you often make the mistake of writing one very long prompt or request, then sending it off and hoping for the best. This approach often leads to major frustration because the chatbot is rarely able to give you exactly what you want on the first attempt.

If you take only one thing away from this chapter on instruction, it should be that good instruction usually takes the form of a dialogue in which you continuously adjust and refine your request to reach a useful result. Some call this approach iterative prompting, but you could also simply call it having a conversation. And most of you already know how to do that.

EXERCISE: The artificial Question-Asker

If you want to practise having a conversation with your chatbot, this is a good place to start:

1. Tell the chatbot about a task you would like help solving. For example, this could be creating a structure for a presentation, a project plan, advice on solving a professional challenge or something similar.
2. Ask the chatbot to ask as many questions as it needs answered in order to solve the task satisfactorily. (Tip: Ask it to ask one question at a time!)
3. Answer the chatbot’s questions. If at some point you feel that the conversation is going in circles, simply ask it to give its suggestion for the task.

Worth trying: 5 tricks for instructing

As you get to know your chatbot, you will probably find your own “tricks” for getting it to do what you ask. Until then, you can try ours:

1. Give the chatbot an ultra-specific role: “You are an experienced Head of Communications with 20 years of experience in crisis communication, who always speaks directly and despises corporate nonsense.”
2. Give the chatbot negative instructions, meaning what it should not do: “Reply without clichés.”
3. Ask the chatbot to generate several different versions or suggestions: “Give me 10 different versions so I can choose the best one myself!”
4. Ask the chatbot to ask about the context itself: “Ask me as many questions as you need answered in order to fulfil my request.”
5. Ask the chatbot to formulate the request itself: “I would like your help with [Purpose], but I do not know how to phrase it as a prompt. Can you?”

Conclusion

You now know how to instruct and interact with your new artificial assistant, so you can hopefully build a long and productive collaboration.

In the next chapter on Monitoring we will take a closer look at how you assess and improve your collaboration with the chatbot. The key point: use your critical sense and professional judgement.

Rules of thumb for mastering instruction

1. Write what you need the answer for (the chatbot cannot read your mind)
2. See it as a process (rather than trying to formulate the perfect prompt on the first go)
3. Be imaginative and also try unusual ideas (you never know what may turn out to work well)