“Farewell,” read the subject of the email. I clicked. Sure enough, a coworker, Benj1, was announcing his decision to resign. The note touched on themes I’ve come to expect in such communications. Highlights, gratitude, a request to stay in touch. It took me less than a minute to realize it was written by an LLM.
What gave it away? It was a stark departure from Benj’s usual style. It had that slightly incongruous, “uncanny valley” ring. Benj’s short tenure working on, let’s face it, standard software problems was described like a decades-long odyssey of heroes. And the writing was complimentary, even effusive, but still wordy and generic.
The ChatGPT email was not bad, and I could tell that some effort had gone into it. It’s even possible that Benj would’ve been comfortable with us knowing it was written by GPT. But personally, I was less touched than if I hadn’t realized.
If you want to get consistently great results with an AI helper, your discernment has to be at least at the level of AI. You need to be able to articulate what you want, identify where AI falls short, and refine. You need intention and judgment.
Without that, the work you produce with AI will fall short of great. Think: communication that lacks your ineffable voice and personality, or that sort of conveys what you want, but not quite. Or: code that’s repetitive, locally optimal but globally clunky, buggy in ways you don’t notice.
To get the best results, you write detailed prompts, evaluate results with a keen eye, iterate and step in yourself to fix outputs when needed. This isn’t a totally different set of skills than producing by yourself, but is a new workflow with a different distribution of time spent on parts of the process. You are quicker (the first draft is a lot easier!), but still active and engaged.
Maybe tomorrow AI will advance to the point where this is easier. Generative AI could tap into your data to learn your personality, and craft something in your voice without much prompting. Still, it can’t know which of many plausible shades of meaning you have in mind, or the specific impact you hope to have with your work. Those things take reflection for us to know, and skill to express to an AI.
So where does that leave us? People tout AI’s potential to level the playing field. For example, this study of customer support workers using AI assistants suggests that lower skilled workers gain the most. Does this fly in the face of my belief that discernment matters? As always, it depends.
Customer support is not the same as creative writing. I’m grateful when I get competent customer support, but I’m not expecting to be moved (unless you count the occasional tear of frustration…). Canned responses based on pattern-matching are fine by me if they’re correct. Also, the customer support agents in the study did not need to prompt the AI. They instead chose from suggested responses or wrote their own. Using AI to write an essay or design a complex system involves more generative tasks, like breaking down a goal into smaller tasks, producing specific instructions for AI, etc. There’s credible concern that only a small fraction of people are discerning and articulate enough to make advanced use of generative AI systems with their prompt-based UX.
Over the weekend, I attended an AI conference2 and heard from amazing speakers like Sal Khan and Daniela Rus. Multiple speakers referenced the idea of humans as “managers” of AI in the future. Anyone who’s been a manager will tell you that it is a high-judgment, mentally demanding role. With the current UX and capabilities of AI, it will take discernment to get the best out of AI. Craft isn’t dead.
I finally wrote something on the internet! Thanks to the community that helped me get there. My partner Viswa crystallized the insight about AI and discernment in a conversation months ago. sent me a link to Jakob Nielsen’s writings. ,
, Sandra Yvonne and Neha Nathan read the draft and provided feedback.Names and some details changed for privacy.
No, I did not go to MIT, and yes, I felt like an outsider in a sea of MIT alumni, but the conference was still wonderful and inspiring!