Prompt Design vs. Prompt Engineering: Making Generative AI Accessible

We need more equity and accessibility in AI

Prompt engineering is a terrible term. "Terrible" might be a bit strong. Let me rephrase. Prompt engineering sounds impressive, but it's not quite appropriate or accurate. Even worse, prompt engineering may deter some people from even trying generative AI.

What's wrong with 'prompt engineering?' The first problem is that engineering is a scary word to many people. College engineering programs are notoriously difficult (at least that's the general perception). Imagine telling a new AI user that they really need to learn prompt engineering.

What? To use this thing I have to be an engineer? I don't think so. Forget it.

Yeah, that's going to go well. The thing is, for many uses, you don't need to be a highly trained expert to use generative AI effectively. That's one of the beautiful things about generative AI. You can get useful results almost from the very beginning. When I help my students learn generative AI, they often get pretty good results in our very first lab. Sure, there are lots of nuances and special techniques that you can use to improve results and expand uses, but a little bit of knowledge goes a long way with generative AI; you certainly don't have to be an engineer.

Engineering is not only complex it's also quite precise. Tiny engineering errors can lead to huge consequences, so engineers are trained to be very careful and precise (and that's a good thing). Large language models don't operate that way. If you calculate the same engineering equation with the same inputs a hundred times, you get the same answer a hundred times. With generative AI, identical prompts typically lead to different outputs, especially for creative tasks. That's not engineering.

I'm advocating for people to start using "prompt design" instead of prompt engineering (although there's room for both terms, as I'll discuss later). Prompt design bring different things to mind. The word "design" connotes a less precise, more iterative process of refinement rather than the application of strict scientific principles and formulas. Design also evokes an emergent process in which the final product evolves from a non-linear, sometimes messy process. (It's important to note that I didn't originate the idea of using "prompt design." I'm sure I heard it somewhere, but I can't recall where.)

In engineering, errors are typically bad things that must be corrected. Design is more forgiving of errors, recognizing that errors are part of the process. In fact, design often views errors as critical learning opportunities along the path to the final outcome.

Design's stopping rules are also fuzzy when compared to engineering. With design, there's often an "ah-ha" moment in which the designer recognizes that they've gotten the design right (or at least good enough for the current purpose).

In my experience, the design process I've outlined above is pretty close to what happens when most people use generative AI. We start off with a sometimes vague idea, and probe different avenues and make multiple refinements until we get what we want. When I go back and look at some of my chat sessions, I'm struck by all of the meandering that takes place. I tend to pursue a lot of different tangents, some of which are productive and some of which only tell me that I need to take a different path. Many of my chats are messy, rambling, and wonderful.

So, I'm a big fan of the term prompt engineering as a replacement for prompt engineering. I think it's a more accurate description of the process of using generative AI, but that's not the big reason I'm a prompt design advocate. The big reason is that it's a friendlier term. As I said at the beginning, engineering is a scary word for many folks. Many of us have been designers of a sort since we picked up our first crayon, but I'm not sure the same can be said of engineering. Although there are certainly highly skilled designers, you don't need to be skilled to engage in design activities. A little knowledge is all you need to get started. Sure, design is a different process with better results when a highly skilled designer is at the helm, but often deep skill is not required to get an acceptable result.

If people think a new technology will be hard to use, they're much less likely to use it. Decades of research into innovation adoption has confirmed this over and over. (It’s also intuitively obvious.) Whether a new thing will be compatible with existing ways of doing things is even more important. If something seems wildly different from normal ways of operating, people are much less likely to use the new thing. That's one reason that the chatbot interface for generative AI has been successful. People are used to chatting, so interacting with a chatbot seems natural (kind of). But if people hear the word "engineering" they're likely to think using AI will be complex and highly technical, which isn't very compatible with past experience for most people. Design sounds much less complex and is more compatible with past experience for most of us.

That being said, there is a place for prompt engineering. There are times when (relatively) precise fine-tuned prompts are the way to go, especially if uses are going to be deployed widely. For example, if a company wants to roll out an AI-based customer service chatbot, they need to do careful engineering to ensure that the bot will answer appropriately and consistently. That sort of use requires rigorous testing and optimization. So, there are many instances in which prompt engineering is necessary. But for the casual user, prompt design is the way to go.

Let's look at definitions of both terms to help better illustrate the differences. (By the way, I used Claude 3 Opus to create these.)

Prompt Engineering: The process of creating precise, fine-tuned prompts to elicit specific, desired outputs from generative AI models, often using technical knowledge and rigorous testing.

Prompt Design: The iterative, user-friendly process of crafting prompts for generative AI models, allowing for exploration, learning from errors, and gradual refinement to achieve the desired outcome.

At the end of the day, it's just a term, but words matter. So, let's start to demystify prompting to make generative AI more inclusive and accessible to regular people to help others gain the benefits of these magical tools.