The imaginary listener.

Let’s continue the discussion from the last post.

So here’s another utilitarian argument for using conversational, natural language with the chatbots, such as troubleshooting programs. And if “polite” isn’t the right word, let’s say “in a tone that is broadly consistent with social norms for this type of information exchange among humans.”

A friend passed a link to my post on to someone she knows with a background in computational lingustics, and reported back to me a comment he had. I don’t have confirmation of permission to quote the individual, but I’ll try to paraphrase:

There’s something about using human-directed language that affects the information we provide.

This friend-of-a-friend’s comment really hit the nail on the head. I thought of all the times I sat down to untangle my thoughts about a subject, all the times I wrote summaries of textbook chapters I was studying, all the times I made presentation slides or outlined notes of material I was preparing to present. I’ve always written to understand, written to learn, written to teach.

And it might be different if I were speaking with my voice, but when I am interacting with a chatbot via text, I know it’s activating the writing mind. The mind that articulates, clarifies, connects, and does this in a singular act with creating a record of those connections.

Whenever I am writing in that way, even when I am writing purely for myself, I always, always, always am writing to somebody. I might have a real person in mind whom I’ve chosen to be my pretend audience, or if the writing is destined for reading, I might have the real audience in mind. But often I have a wholly imaginary interlocutor in mind—and it’s still useful. As in rubber duck debugging, explaining my arguments “to someone” forces me to articulate clearly and logically and to consider the beginner’s perspective, the starter-from-scratch.

So when I’m writing anything, but especially writing to understand a problem, I’m always thinking: How do I explain the issue to this listener?

And so why shouldn’t I use the same voice that I use to “an imaginary listener” when I write queries to a chatbot?

A chatbot is an imaginary listener!

It is essentially an extremely online rubber duck.

So I’m going to be comfortable from now on using my “imaginary interlocutor” voice when I find myself interacting with chatbots or other LLMs. I think I might have a new favorite image, however, for the entity I’m explaining things to.