Sooner or later, all knowledge workers will probably be working with AI tools to enhance their productivity. Then the big question is, how do we interact with these AI tools? Is it by chatting, speaking, writing, turning dials or by configuring variables? There are many ways things can develop and different use cases require different user interfaces.
If we believe that the chat interface will be the way to go, how we word our questions becomes the center of attention. If we want a AI chat bot to write us a recommendation letter, it will give us one. But if we want to be specific, we must include that in our message, like write it in old English style. This process of honing in on the language and wording of our message to get AI to produce specific wanted results, is called prompt engineering.
So, what is prompt engineering exactly?
A basic example of prompt engineering is using AI image generation to generate new images based on a written prompt. A prompt is the words or sentence you write to describe what you want, that is then fed to the AI system. If we tell the AI system to give us an image of “a horse riding a unicycle”, we will get many interpretations of that theme. Some of the images are more relevant to us, some are total misses. Maybe we wanted the image to be in the style of Salvador Dali. So, then we update our prompt to “a horse riding a unicycle in the style of Salvador Dali” – and now we get a resulting image that more closely matches what we wanted. And this iterative process can go on, for a long time, until the results satisfy the requirements.
Now, if we believe that is the way we are going to be using AI tools, that eventually leads to the following scenario: Every knowledge worker will be doing their jobs by writing prompts! The software developer writes “implement a function to send REST API calls to a test server” and voila! – the AI system implements the wanted function. Same goes for the marketer, who writes a prompt asking for “A marketing pitch for a new cool product” and AI will give a satisfying result. And the list goes on, you get the idea.
I think the idea is really fascinating – how to express your idea in natural language so that the AI system understands it the same way you do? We are so used to communicating with humans that we don’t realize how hard it can be to convey our ideas clearly and efficiently. Humans do so much error correction in, first the signal processing part (what words are said) and second, in understanding the meaning of those words (understanding the context).
We have the first part figured out – how to transcribe audio into words and feed them to an AI system, like you can talk to your smart phone. But the second part is still under development. We can get quite close to what we want, maybe through trial and error, but still, we wish that computers could understand us as easily as other humans. But maybe it’s not too long until we get there. Until then, keep honing on your prompt writing skills.
Prompt engineering, Wikipedia, Available: https://en.wikipedia.org/wiki/Prompt_engineering
About Tuukka Virtanen
Test automation consultant with technical experience in test automation and quality assurance. TMap Next certified Test Engineer with knowledge in test planning and execution and test design techniques. Master of Science in Information Management. Indie game development as a side project. Creative and visual thinker. The latest assignment included web and mobile game test automation with Appium and Robot Framework in an Agile customer project and regression test automation for websites.
More on Tuukka Virtanen.