The person who never existed

A few days ago, I came across a Pew Research Center study about synthetic people in research. My first reaction was simple: “If they aren’t real people, why would I ask them anything?” So I kept reading.^[1]

Put simply, a synthetic person is a profile an AI tries to represent. In Pew’s experiment, a model took on the role of real members of its panel. It received demographic information and each person’s previous answers, then answered the same questions from three surveys. Pew compared the human and synthetic results.^[2]

The idea felt strange and fascinating at the same time.

Imagine a couple in their mid-thirties with a young child, a certain income and a two-bedroom flat. We give an AI that profile and ask what they would look for when moving home. The answer might be perfectly reasonable: more space, nearby schools, peace and quiet, perhaps a terrace.

But that couple has not viewed a single home. They have not walked into one that seemed perfect and changed their minds because of the light. They have not debated whether moving ten kilometres farther out is worth it. They have not discovered that something they called essential may not have mattered quite so much.

The AI can build a plausible answer. But no one has lived it.

That is where my curiosity starts to itch.

Across the three surveys Pew studied, the average absolute error between synthetic and human results was 12.4 percentage points. It was higher for some groups. Pew also found less diversity in the opinions generated.^[3][4]

Another detail stayed with me: changing the model could change the picture. GPT-5.1 and Claude Opus 4.6 did not always represent public opinion in the same way.^[5]

So who are we listening to? A possible consumer? Patterns the model has learned about similar people? Or the AI’s own statistical personality, too?

I am still learning about AI in research, and I do not have a settled answer. I do not think these tools should be dismissed either. They may help us explore scenarios, test questions or form hypotheses before going out into the real world.

But I find it hard to see them as the destination.

Perhaps a synthetic person can help us imagine what is likely. That is exactly why we need to be careful not to confuse it with what someone is actually experiencing.