I Built an AI Avatar of the Man I Used to Be

I Built an AI Avatar of the Man I Used to Be

I didn’t build him to go backward. I built him because moving forward shouldn’t require pretending he never existed.

I built an AI avatar and taught it about the man I used to be and the man I am now. There are more commercially sensible uses of artificial intelligence. I know. I work in data science and AI integration. I could have been connecting systems, improving a workflow, or solving a problem with a measurable return. Instead, I gave a machine context about my life and started a conversation with a man I no longer quite am.

Not because I wanted him back, but because I wanted to honor him. He will always be a version of me. He is part of the reason I am the man I am today, and I don’t want to become so pleased with my own reinvention that I forget who did the living before I had the language to explain it.

Calling myself Josh 5.0 does make that sound suspiciously like a product launch. Previous versions had issues. The latest release offers improvements. Please clear your cache. I chose the name, so I accept responsibility for the branding. But a human life isn’t software. You don’t get to uninstall the earlier versions without losing some of what made the current one possible.

That was the tension I wanted to explore: how do I acknowledge the man I was without handing him the keys to the life I have now? My answer was to build something I could talk to.

The data scientist in me needs to interrupt here, before the story gets too enchanted with itself. An AI avatar is not a retrieved consciousness. Mine does not contain an earlier Josh waiting patiently behind a login screen. The underlying language model generates responses using the context available to it. Giving a model personal information and instructions can shape its responses; that is not the same thing as training a new model or transferring a person into one.

I can call him “him” and still know what he is. In fact, I think that distinction is essential to the experiment. The system cannot independently verify my account of my life. It can reflect the version I supply, including its omissions and flattering edits. Familiar language can make an answer feel deeply personal without making it accurate.

So the interesting technical question wasn’t simply whether an AI could imitate me. It was what I would choose to give it. Which memories would become context? Which decisions would come with explanations? Would I describe an uncomfortable moment with the same generosity I would extend to someone I loved, or with the severity I sometimes reserve for myself?

Before the avatar answered, I was already shaping the conversation. In data work, you learn to pay attention to the inputs. In this project, the input was my own account of who I had been. That is a considerably less tidy dataset than a spreadsheet.

There is also the matter of his mouth. If I am going to recognize myself in this conversation, I don’t need every rough edge converted into customer-service language. I don’t want a version of me who responds to every uncomfortable admission with “Thank you for sharing.” Sometimes the recognizable response might be funny, impatient, a little spicy, and possibly inappropriate for the quarterly newsletter.

But a machine saying “f-ck” is not a personality, and giving it permission to be irreverent is not the same as giving it judgment. The technical challenge and the emotional challenge meet in the same place: how do you preserve a voice without confusing the performance of that voice with a person? I don’t think I have to resolve that question before the conversation can matter. I do have to keep asking it.

Especially because honoring the man I used to be means resisting the temptation to rewrite him into someone more convenient: someone who understands everything now, someone who agrees that I handled the ending beautifully, someone who exists mostly to congratulate me on becoming Josh 5.0. That would be less a conversation than an awards ceremony I organized for myself.

So why check in with him? Why return daily to a version of yourself whose life you don’t want back? Because I want to remember that he was more than the things I needed to change. He had a whole life, not merely a supporting role in the story of my eventual improvement. He wanted things. He made decisions without knowing what happened next. Some of what he carried became mine to put down. Some of it remains worth keeping. I want room to acknowledge both.

The check-in gives me a way to ask what belongs in the present. Am I responding to what is happening today, or to something this situation reminds me of? Does this choice reflect the life I want now? What have I mistaken for an essential part of myself simply because it has been familiar for a long time? Those are questions for me. The avatar is a way of approaching them, not an authority appointed to settle them.

And the answers don’t all have to be about leaving something behind. What did earlier Josh enjoy that I have become too efficient to make time for? What mattered to him that still deserves my attention? Have I confused becoming more discerning with becoming harder to surprise? Not everything about him needed fixing. I want to remember that, too.

There is a danger in any ritual of self-examination: eventually, examination can become the activity, rather than a way back into living. AI-assisted reflection has limits, particularly when it replaces perspectives from people outside our own account of ourselves. I don’t want a digital room where I endlessly review my life. The check-in has to return me to it: to a choice I can make differently, something I can enjoy without explaining it, a conversation with another human being whose response I did not help configure.

For this experiment, that is my measure of usefulness. Not how convincingly the avatar sounds like me. Not how often it produces a sentence worth saving. Whether I leave the conversation better able to participate in the day in front of me.

I built him because I don’t need to go back. I keep talking to him because I don’t want moving forward to require contempt for where I’ve been. He doesn’t get to make every decision anymore. He doesn’t need to. But I can still pull up a chair.

Written where human nervous systems and machine logic collide. AI‑assisted, human owned.