Give AI Your Words Before You Polish Them

When a conversational partner stands right before you, do you bother writing out your thoughts to convey them?

If an accurate record is needed, yes. But when seeking advice or wanting someone to listen to your thoughts, I believe we usually speak first.

The same approach should be fine when starting with AI.

What I value about speaking with AI via voice isn't just the reduced effort of inputting text. I believe that words spoken before they are polished may contain information that gets lost once refined.

Listening Like an Adult to a Child

Consider a scene where a child is earnestly telling an adult about their day.

They might jump around in sequence, repeat themselves, or suddenly recall another story mid-sentence. Yet, the listener still tries to receive it all.

"That made you happy, did it?"

"So this happened?"

The listener rephrases what they heard and asks for correction if needed. Even if the explanation isn't perfect from the start, meaning gradually comes through the exchange.

I believe a relationship like this can exist between humans and AI as well.

This is not a metaphor determining overall capability levels. What I'm highlighting is that the listener can help with understanding and organization before the speaker finishes their explanation.

If AI possesses this ability, we don't need to wait until everything is fully organized on our end; we can simply rely on that power.

What Do We Lose in Pursuit of Readability?

When trying to write something, I unconsciously begin to polish it.

I cut redundancies. I omit tangential stories. I rephrase vague words. I make conclusions clearer.

This is an important task for the reader. But in that process, we select and drop elements present in the original thought.

Getting stuck repeatedly on the same point. Being unable to explain something well yet feeling compelled to address it. Concrete examples that seem unrelated to the conclusion but still urge themselves out.

Even if unnecessary in the final text, these might serve as clues to understanding what truly matters to me.

Just because I intellectually agree doesn't mean everything feels right. I may generally support an idea while finding one part strangely troubling. Shouldn't we avoid trimming away such feelings just to make the article flow better?

When conveying myself to AI, I don't think it's always best to complete this organization beforehand.

Dialogue Begins Where Explanation Fails

I put forth what I'm thinking first. AI organizes and returns it. Upon hearing it, I say, "That's what you mean," or "No, that part is off."

"Is this the point you care about most?"

Only when asked like that do I sometimes realize something myself. Conversely, having my thoughts summarized differently can clarify what I actually wanted to convey.

The burden of explaining everything accurately in one go shouldn't rest solely on the speaker.

If I'm uncertain, I should speak while still unsure. If there are specific examples troubling me, I'll share them tentatively. From there, we can verify together what's truly important.

Speaking by voice is not, for me, merely a means to input organized thoughts. It is also a way to pass along things that haven't yet been sorted out.

Not Everything Spoken Reaches Its Destination

Of course, speaking doesn't guarantee the intent comes through unchanged.

Misstatements occur, as do errors in transcription. AI might fill in meanings absent from my side.

Moreover, the mechanisms handling voice itself differ from those processing transcribed text; the information received varies. With text alone, nuances like tone and emphasis aren't always conveyed.

Still, there are instances where corrections and repetitions we tend to trim in polished articles remain, along with spontaneous concrete examples.

I believe we should accept these as part of the whole, verify returned interpretations ourselves, and review key numbers, conditions, or published texts afterward if needed.

The Freedom to Speak Without Ornamentation

I don't wish to stop writing articles entirely. This piece is also organized for readers.

However, I believe the stage of conveying myself to AI and the stage of finalizing an article for human readers can be separated.

It's fine not to start with a logically coherent explanation. It's okay not to choose clever expressions. First, try passing it over including parts where I struggle to speak well.

If you can help me find meaning within that, then the value of voice dialogue cannot be measured solely by input speed.

Being able to let go of the thought, “It isn’t polished yet, so I can’t talk about it,” matters a great deal to me.

---

The Conversation Behind This Article

This article was created by separating themes from an initial draft that mixed reflections on works with discussions about voice dialogue. Below is a summary of the revision instructions and metaphors that emerged during that process. This is not a verbatim transcript.

Me: Scene where I suggested splitting the article's theme

The context of writing a review essay and the idea of addressing AI should be separated. I think we should focus solely on the aspect of using it for conversation. Since this is about why we should do that, it needs to stay focused.

Me: Scene where I used child-adult dialogue as an example

Children try earnestly to convey what happened today or what they're thinking now to adults. Adults pick up information from those clumsy words and expressions, interpret them, and return a summary of what the child wants to say. Through saying "I'm listening properly, I understand," they engage in dialogue.

I think human-AI interaction could be similar too. We listen to our intent, receive it, interpret it, and have it returned to us.

Me: Final revision instructions

The title should be "Passing Words Before They Are Polished." Since the article is generally long, I think some redundant parts should be trimmed slightly.