Introduction
Erving Goffman described social life as a stage on which selves are performed and managed. This essay reframes that vocabulary in the language of partial observability: interactants observe cues and infer intentions, competence, and moral character that remain latent—hidden state in a social POMDP.
Front stage, back stage, and belief
Front region performances correspond to observable emissions; back region preparations correspond to latent dynamics. Trust, embarrassment, and “face work” are belief updates when the hidden process leaks into view. Goffman’s micro-sociology is therefore continuous with Bayesian theories of mind—though affect and dignity add normative weight pure likelihoods omit.
Applications
Online identity multiplies fronts; latency and editability change what “sincerity” can mean. Institutional roles—clinician, teacher, officer—bundle expectations that constrain how much latent strain may surface. For AI evaluation, the lesson is sharp: benchmarks that score static worlds miss the interactional heart of intelligence.
History and method
Goffman’s ethnographic imagination drew on mid-century fieldwork; updating his framework requires attention to digital ethnography, platform governance, and cross-cultural variation in face norms—topics the essay sketches without pretending closure.
Conclusion
Hidden selves are not bugs in social life; they are structural features of coordination. Good maps of interaction must model what actors cannot—and should not—fully disclose in real time.