What a digital twin is, and what it isn’t

Search the phrase and the first results are about turbines. An engineer builds a digital twin of a jet engine: a model that takes live readings from the real thing, so they can see a bearing wearing out before anyone hears it. Bridges have one. Wind farms, ports, whole cities. The point is the same every time. The model stays tied to the thing it copies, and it is only useful while it tells the truth about that thing.

Lately the phrase has picked up a second meaning: an AI that speaks for a person. Same words, very different thing, and the gap between the two is where most of the confusion lives.

The engineer’s version

A twin of a bridge is fed by sensors. Strain, temperature, vibration, traffic. It doesn’t guess; it reads. When a reading is missing, the model says nothing about that part of the bridge, because there is nothing to say. Nobody would trust a bridge model that filled the gaps with a plausible number.

That standard is worth holding on to, because it is exactly the one an AI twin tends to drop.

The person’s version

An AI twin is a chat. Someone opens a link, types a question, and gets an answer about a person who isn’t there. Most of what gets sold under this name is a large model that has been given a bit of information about you and told to act like you. The model fills in the rest from everything it has ever read. That is where the honesty goes. Ask it about a job you never had and it may invent one, because inventing is what it is good at.

The useful version is narrower and duller. You write a set of answers about yourself, in your own words. An AI reads those answers and takes nothing about you from anywhere else. A visitor asks something; the AI finds the sentence that covers it and answers from there. If no sentence covers it, it says so and points the visitor to you.

That is the whole trick. The AI has no sensors into your life. It has your page. So it should behave like the bridge model: read what is there, stay quiet about what isn’t.

What it can honestly do

Quite a lot, as it turns out. It can tell someone what you do, the way you would say it. It can explain why you do it, what you are proud of, what you are like to be around, what you would talk about for hours. People want to know you, and you can’t be in every conversation. A twin is the part of you that is there for them: the friend who moved away, the new classmate too shy to ask, the aunt who has never quite understood what you do.

It can also be a better first hello than a profile. A profile is a list. A twin can be asked a follow-up.

And it can keep the thread. Where a visitor asks it to, or where the two of you are connected, it keeps one short note about that person, so the next chat picks up where the last one stopped. They can read the note and delete it. You can read every chat your twin has, and visitors are told so before they write a word.

What it should never claim

Three things, and a good twin refuses all of them.

It should never claim to be you. “I” in the chat is the AI. You are referred to by name. A visitor should never have to wonder whether they are talking to a person or a piece of software.

It should never claim to know something you didn’t write. No guessing your salary, your politics, or your opinion of a company you never mentioned. The right answer, with your name in it, is “Deniz didn’t write about that, so I don’t know. Ask Deniz.”

It should never rate you. No score, no ranking, no verdict on whether anyone should hire, trust or date you. A twin that grades people is a different product, and not one anyone asked for.

Both meanings of “digital twin” can stand, as long as the second borrows the discipline of the first. A model that reads and reports, not a model that performs.

That is where we start: one page of your answers, one link, and an AI that reads the page and admits what it doesn’t know. From there you can connect with your friends, and your twin keeps the thread with each of them. If that sounds like enough, you can make your own twin now.