In one of the early scenes in Inception, Cobb tells Ariadne that the people around them in the dream are projections drawn from his subconscious. The dream does not observe a world. It populates one from prior absorption. The architecture, faces, gestures, and social texture of the street feel real because they are built from real material. But the source is memory and expectation, not contact.
Much of the current enthusiasm for synthetic publics works inside a similar loop. Generated panels, personas, agents, and simulated stakeholders do not recruit participants or observe social positions. They recombine patterns from training data: prior research language, documented consumer segments, labeled attitudes, public discourse, and published verbatims. The output sounds like social research because it is built from the residues of social research and public speech. That is useful, but it is also the limitation. Synthetic personas are shaped by what has already been said about a population, not by accountable contact with the population itself.
For organizations under pressure to produce faster insight, this looks like a cleaner version of fieldwork. But that cleanliness is part of the validity problem. The dataset appears complete because the model fills the gaps.
They are oh so tempting. Synthetic respondents do not reschedule, drop out, or produce unusable recordings. They can be queried repeatedly and segmented instantly. For organizations under pressure to produce faster insight, this looks like a cleaner version of fieldwork. But that cleanliness is part of the validity problem. The dataset appears complete because the model fills the gaps. For Inception, Christopher Nolan borrowed Escher’s staircase for a reason: each local step is coherent, each transition follows a rule, and the structure folds back on itself without exit. A synthetic panel works the same way. It can produce internally consistent responses while remaining structurally disconnected from an actual population.
