January 2023: “Learn prompt engineering – it’s the wave of the future and your ticket to AI riches.”
January 2024: “Don’t waste your time learning prompt engineering – soon all prompts will be written by AI.”
Which of these is correct?
There’s a grain of truth in the claims that AI will automate prompt engineering. That doesn’t mean, however, that humans won’t still be needed to write prompts – or to write whatever inputs are needed to guide AI once prompt engineering has been automated.
AI Is Automating Prompt Engineering
The “AI is making human prompt engineering obsolete” camp is onto something. There are several ways in which the process of writing prompts is being automated:
- Large language models (LLMs) themselves are pretty good at creating and improving prompts. Try asking ChatGPT to fine-tune a prompt that you’ve written and you’ll see that it often does a great job. This approach doesn’t require any new technology or even a separate tool to automate prompt writing.
- Applications, such as PromptPerfect, Promptable, and Promptify, can optimize human-written prompts in various ways. You can then use those prompts in ChatGPT, MidJourney, and other prompt-based AI applications.
- AI tools, such as Leonardo and DALL-E, are beginning to integrate automatic prompt generation or offer it as an option.
Many users will be satisfied with these options because they enable “good enough” results to be produced without special prompt writing skill.
Human Prompt Engineering Skill Will Still Be Valuable
But what about when good enough isn’t good enough? What if you need to create an article, a piece of art, or a video that is unique, of professional caliber, or that expresses your unique voice more precisely than you can achieve using the kinds of generic prompt engineering automation I described above? In those cases, I contend that human skill at manual prompt writing will continue to be valuable, at least as part of the creative toolbox.
I recently saw a great reminder of this in the form of a set of tokens that a skilled graphic designer had found to be particularly useful at creating high-quality and imaginative artwork using Midjourney, as the result of having used that tool to create over 75,000 images. Take a look at his images and I think you’ll agree that they required real skill and many hours of experimentation, not just generic prompting.
Although AI-generated prompts will get the job done in many situations, specific skill at manual prompting and – dare I say it – the “old fashioned” manual skills of writers, artists, and coders will all still have their place in creating and polishing professional end products. Most valuable will be the higher-level skill of knowing when and how to use and mix old and new skills.
By analogy, although sometimes new species supplant the old in the natural world, much more frequently the new and the old occupy different and complementary niches. We humans coexist with bacteria that first evolved billions of years ago. In light of that, let’s not be so quick to assume that a much more recent technical skill will be completely replaced by relatively small technological advances so quickly.
AI Will Create Demand for Humans To Write New Kinds of Inputs
The claim that AI-automated prompt generation will make human prompt writing skill unnecessary also seems to reflect an ignorance of the impact that new generations of automation have always had on human skills in computer science.
First, humans programmed computers in machine language–directly in the 0s and 1s that computers directly understand.
Then assembly languages were created, which enabled programmers to write software code in ways that more closely resembled natural language.
Then, the introduction of so-called “high-level” languages, such as FORTRAN, COBOL, and C, enabled human programmers to write software code using even more abstract instructions.
We can trace the same pattern to the present day, through the introduction of structured programming, and then object-oriented programming, and so on all the way to the low-code/no code platforms that are now popular. You can even view the use of language model-based tools, such as ChatGPT and GitHub Copilot, as extensions of this trend which now enable humans to use natural language as a programming language.
At each stage in this process, you could say that the introduction of a new kind of programming language or technology made it unnecessary for humans to learn the skills that enabled them to succeed at programming using the previous generation of programming tools. Yet that would only be partially true, for at least three reasons:
- Human skill at using the previous generation(s) of technology can still be valuable in some circumstances, such as to obtain efficiencies that can’t be achieved using the latest technology, which tends to be bloated compared to its predecessors.
- Problem-solving techniques often are highly transferable from one generation to the next. General engineering education remains valuable even when a new layer of abstraction or automation is introduced by technology.
- The latest advance creates demand for a new set of human skills. Assembly language created a need for people who knew how to write assembly code. Fast forward to a year ago, and LLMs created a need for human prompt engineers. Even if prompts can be generated automatically, there will be a need for humans who are skilled at using systems that incorporate automatic prompt engineering.
Beyond Prompt Engineering: What’s Next for AI and Humans?
We’ve now seen that although there is some validity to the claim that automation of prompt engineering will reduce the need for humans to invest in developing prompt engineering skills, this truth obscures a deeper historical pattern: each time technology automates a human skill, that very automation creates demand for a new human skill, while also not completely eliminating the need for the old skill. Although we don’t know exactly what new skills will be enabled or needed in the aftermath of the automation of prompt engineering, we do know that humans will need to develop and deploy those skills manually – at least until the next wave of automation, when the pattern repeats itself.
