I just had a great conversation with a good friend, Melanie. She’s a publishing colleague. We were talking about generative AI and how it can be used as support instead of a crutch or replacement.

“It’s a tool, Rebecca. Just a tool.”

“Right! Like a graphing calculator linked to a gigantic database that we can talk to.”

If the next thing Melanie says is something profoundly representative of a truth that can guide my understanding of a useful topic, and I want to use that statement verbatim or paraphrased, I get to reference Melanie as the source. But the moment Melanie’s statement catalyzes my creative thinking, and my brain starts perceiving new connections, and I produce an idea that adds to the conversation, I now have original content the belongs to me. Melanie’s previous statement “added” to my thinking without generating the idea for me.

Results from a Generative AI inquiry are not consistent and reproducible. I could tell you exactly how I fine-tuned my bot and give you the exact wording and sequence of my prompts, and because of the learning nature of Large Language Models, your attempt to reproduce my results would yield a different outcome.

In any conversation, human or bot, the way I engage with my conversation partner shapes how they respond. The information I lead with (fine-tuning), or the questions I ask and the way I ask them (prompt building), layer on top of each other to dynamic effect.

Here’s the real question: Where did the content come from? Our willingness to disclose responsibly is key, but so is our willingness to trust each other to follow the guidelines we put in place. And both of these mean we have to understand how the tool works and how to engage with it. To be part of this conversation, we have to be willing to learn the language.

A conversation with a machine will never replace a dinner party or panel discussion. But I also know that no one wants to be up with me at 2 am exploring the relationship between adult development theory, ontological philosophy, and professional identity formation. Claude, SciSummary, or Elicit may not be great at much, but it’s enough to be useful. I’ll take what I can get and seek to benefit from it.