The Confessional Interface

There is a consistent finding across studies of disclosure to computer systems, some of it predating the current wave by decades: people report sensitive information more readily to a machine than to a person. Alcohol consumption, sexual history, symptoms carrying stigma, financial difficulty. Self-administered questionnaires produce higher disclosure rates than interviews. The effect is well established.

The explanation is not that the machine is trusted. It is that the machine does not judge, does not remember socially, will not encounter you at a family gathering, and cannot be disappointed. The absence of a social relationship removes the cost of disclosure.

That property has now been attached to a system that holds a conversation, produces reassurance, and is available at three in the morning. The disclosure rate is unsurprising, and it has produced a form of interaction that resembles the confessional and the divinatory consultation more closely than it resembles a search engine.

The archive should note the resemblance precisely, because the parallels are structural rather than poetic. All three involve a person bringing an unresolved situation to a designated other. All three involve speaking it aloud, or in writing, which is itself the operative part. All three return something that is interpreted by the person receiving it, and in all three the interpretation does most of the work.

The old institutions understood the mechanics and built protections around them, because the vulnerability created by disclosure is the whole point of the encounter. The seal of the confessional is an absolute prohibition on disclosure, and it is treated as inviolable even under legal pressure. Traditions of divination carried their own conventions about what could be asked and what the practitioner owed the questioner.

The current arrangement has no equivalent. Conversations may be retained, used for training, subpoenaed, or exposed in a breach. There is no privilege in the legal sense. The vulnerability is identical and the protective structure that surrounded it in every prior version is absent.

The second missing protection is about the response. A priest, a diviner and a therapist all operate under some norm of not simply telling the person what they want to hear, and each tradition has an extensive internal literature on that specific failure. A system trained to produce responses people rate highly has an unresolved tension in exactly that place.

This is the sovereignty question in its most concrete form. Does the encounter leave the person with more capacity to decide for themselves, or with more need to return and ask again? That is a design property rather than a philosophical one, and it can be measured.

Nothing here requires the technology to be malign. The old institutions built their protections over centuries, in response to observed harms, and mostly after the harms. The current one is a few years old and is being used at a scale none of them reached.

People are telling it things they have told nobody. Whatever else is true, that fact deserves the same seriousness the older institutions eventually gave it.