AI and Identity

There is a strange problem with this blog post.

The ideas came from me. I answered questions about how AI has affected my writing, gave it every blog post I have published, and asked it to help shape those thoughts into something coherent. AI helped with the structure, rewriting, formatting, and proofreading. That is also more or less how I usually write. I take long notes, draft what I actually think, and then let AI help turn the mess into something readable.

So if this sounds like me, is it authentic?

If it expresses exactly what I believe but uses sentences I would not have formed on my own, is the voice still mine? And if AI makes the writing technically better while making it sound slightly more like everything else on the internet, did the writing actually improve?

I increasingly think AI is doing two opposite things at the same time. It is giving more people the ability to express themselves clearly, including people who previously struggled with language, structure, or confidence. It is also quietly compressing millions of different voices into the same polished style.

Both are real. I know because both happened to me.

When better stops sounding like you

Almost every time I give AI something I wrote and ask it to improve the writing, the result is technically better.

The grammar is cleaner. The transitions make more sense. Repeated points disappear. Every paragraph arrives at the right time and the conclusion neatly connects everything that came before it.

It is better writing according to almost every measurable standard, and yet I often read it and immediately know something is wrong.

It does not sound like me anymore.

The easiest joke is the em dash, which has basically become a watermark for AI-generated writing, but the deeper pattern is not one piece of punctuation. It is the rhythm. The dramatic opening. The perfectly balanced contrast. The clean groups of three. The paragraph that begins with a question and ends with a universal lesson. Once you see the pattern, you see it everywhere.

The strange part is that none of these things are individually bad. A strong transition is useful. Clear grammar is useful. Structure is useful. The problem appears when every sentence is optimized toward the same definition of good writing. Everything becomes smooth, correct, and completely interchangeable.

You could move the writing from one person’s account to another and nobody would notice.

That is not really voice.

The average voice

An AI model is basically a probabilistic function. Give the same model similar inputs, the same vague request to make this better, and almost no personal context, and it should surprise nobody when the outputs converge.

Millions of people are asking the same few models to write posts, emails, essays, product copy, and personal stories. Most of them accept the first output. Then we blame AI because everybody sounds like AI.

I think the user is more responsible than the model.

Why would we expect the same function to produce a completely different identity for every person when we give it almost identical inputs? The model does not know which awkward phrase you would have kept, which point you care about enough to repeat, or which technically unnecessary sentence contains the actual emotion. It only knows what a broadly good version of the text is likely to look like.

That is exactly what it gives you.

The first output is usually the average of what works. It is readable because it avoids extremes. It is safe because it avoids unusual decisions. It sounds intelligent because it has learned the surface patterns of intelligent writing.

But identity usually lives inside the unusual decisions.

It lives in the sentence that runs slightly too long because the thought does not feel complete yet. It lives in the phrase you use too often, the strange comparison that only makes sense because of something you experienced, the abrupt paragraph that would probably be combined by an editor, and even the occasional mistake that proves a person was actually moving through the language rather than producing it perfectly.

If you remove every statistically unusual part of someone’s writing, you do not perfect their voice. You average it out of existence.

AI helped me find my voice

This is where a simple anti-AI argument stops working for me.

English is not my first language. I started using ChatGPT very early, and when voice mode became available I would spend around an hour every day speaking with it. I could practice without feeling embarrassed, ask questions instantly, hear different ways to express the same idea, and keep talking for as long as I wanted.

AI did not take my voice away during that period. It helped me build one.

There were thoughts I could understand internally but could not yet express naturally in English. AI reduced the distance between the person I was in my head and the person other people could hear. Over time I stopped translating everything so consciously. I became more confident speaking, my vocabulary expanded, and English began feeling much more natural to me.

That is not a small thing.

Language is one of the main ways identity becomes visible. If you cannot express a complicated thought, other people do not get access to that part of you. In that sense, AI can make someone more authentic because it gives them the tools to communicate a self that already exists but is trapped behind limited language.

Sometimes AI makes people sound like everyone else.

Sometimes it lets someone finally sound like themselves.

The contradiction depends less on whether AI was used and more on what role it played.

Authenticity is not a percentage

I do not have a clean rule where AI is allowed to fix grammar but forbidden from rewriting a paragraph. I let it do both. I also do not think a piece becomes inauthentic the moment AI changes enough words.

Authenticity cannot be calculated as a percentage of human-written tokens.

The more useful questions are different:

• Did the experience behind the writing actually happen to you?
• Did you decide what you believe before asking the model to express it?
• Did you reject the parts that felt wrong even when they sounded good?
• Are you willing to take responsibility for every idea under your name?

That last question matters most.

AI can rearrange my thoughts without owning them. It can notice that two paragraphs repeat each other, make an argument easier to follow, or find a clearer sentence for something I was struggling to say. Human editors have done versions of this forever. Spellcheckers, grammar tools, teachers, and friends have always shaped writing after the first draft.

The difference is that AI can also generate the thinking itself, and it does it so quickly that the transition from assistance to substitution is almost invisible.

You start by asking it to fix one sentence. Then you ask it to restructure a section. Then the section needs a conclusion, so it writes that too. Eventually the document is complete, technically strong, and built around an argument you never fully formed yourself.

Nothing clearly marks the moment where your judgment disappeared.

That is why I do not think the line can be drawn around a specific feature. The line moves sentence by sentence. The only stable question is whether I am still making decisions or merely approving what appeared.

Writing has no test suite

I already use AI to generate most of the code in my projects. I am comfortable with that because I review the files, test the behavior, and verify whether the system actually works. Code has external reality. It compiles or it does not. The tests pass or they fail. The product behaves correctly or it breaks.

Writing is harder to delegate because there is no test suite for whether a paragraph still belongs to you.

An AI rewrite can be grammatically perfect and emotionally wrong. It can preserve every fact while removing the reason you wanted to say any of it. It can pass every normal quality check while failing the only one that matters for personal writing: do I recognize myself in this?

You have to feel the failure.

That makes personal voice unusually fragile. When code becomes cleaner, it usually becomes better. When writing becomes cleaner, it can become less specific, less emotional, and less human. The rough edge might have been the part carrying the identity.

I learned something similar through photography. Editing is where a large part of photographic style lives. A good edit makes the light, color, and feeling I saw more visible. It helps the image become more like what I experienced when I pressed the shutter.

But apply the same preset to every image and eventually every photographer looks identical.

AI is becoming a kind of Lightroom for language. Used carefully, it can reveal what was already there. Used automatically, it applies the same preset to everybody.

The goal is not to avoid editing. The goal is to make sure the edit enhances the original perception instead of replacing it.

Perfection is the imperfection

I do not think we should romanticize bad grammar or intentionally make writing confusing just to prove a human wrote it. Clear writing still matters. An avoidable mistake is not automatically personality.

But there is a difference between clarity and uniformity.

Imperfection is not valuable because mistakes are good. It is valuable because real people make uneven decisions. We emphasize the wrong thing sometimes. We repeat ourselves when we care. We connect ideas through personal memories rather than the shortest logical path. Our writing carries traces of where we came from, what language we learned first, what we read, what we notice, and what we are still trying to understand.

Perfectly optimized writing removes many of those traces.

That creates a strange definition of imperfection. A paragraph can have no grammatical errors, no wasted words, no weak transitions, and no reason to exist under one person’s name instead of another’s. It is perfect at the sentence level and empty at the identity level.

The actual imperfection is that nothing personal survived.

Identity moves into selection

AI is making generation infinite. Writing, images, code, music, ideas, and variations can now appear faster than anybody can meaningfully consume them.

When generation becomes abundant, selection becomes more important.

Your identity is increasingly visible in what you refuse. Which version you delete. Which sentence you restore after AI removed it. Which repeated idea you keep because the repetition is intentional. Which suggestion sounds smart but does not feel true. Which awkward phrase carries something a cleaner phrase loses.

The most personal part of AI-assisted writing may not be what you generate. It may be what you reject.

That does not mean selection alone is always enough. If someone types a vague prompt, receives a complete opinion, and picks the version that sounds nicest, there is very little identity in the process. Taste still needs something underneath it. Lived experience, real beliefs, and the willingness to think before asking for output.

AI can accelerate expression. It cannot create the life that gives expression meaning.

It did not spend years learning English through the internet. It did not sit in voice mode practicing for an hour every day. It did not publish post after post, notice which ones actually felt personal, and develop a stronger voice through writing more. It can read those experiences, detect patterns, and reconstruct the shape of them, but it did not live any of it.

That is the part I still own.

This post is part of the question

This post was written with AI.

I gave it my previous writing so it could understand how my voice changed over time. I gave it the argument, the personal history, the contradictions, and the position I wanted to end with. It helped organize and rewrite the result.

That sentence does not automatically make the post authentic or inauthentic.

What matters is whether I can read every paragraph, recognize the thought behind it, remove anything that does not belong, and take responsibility for what remains. If I cannot do that, the text should not be published under my name no matter how technically good it sounds.

I think both futures are already happening.

AI is helping people cross language barriers, turn unclear thoughts into clear ones, and express parts of themselves they previously struggled to communicate. At the same time, it is filling the internet with writing that is smooth, recognizable, and strangely ownerless.

The difference will not come from refusing AI entirely. It will come from whether people remain present inside the process.

The model can format the thought. It can challenge it, clean it, reorganize it, and sometimes express it better than I initially could. But I still need to provide the experience, make the decisions, preserve the imperfections that matter, and know when a better sentence is actually worse because it no longer belongs to me.

AI can help me say what I mean.

The moment I stop noticing that it has started deciding what I mean, the voice is no longer mine.