Digital Trust

Declining trust in digital institutions is routinely diagnosed as a communications problem. The premise is that the system is sound and the public has misunderstood it, and the prescribed remedy is better explanation. Sometimes that diagnosis is correct. Frequently the public has understood the situation precisely and responded rationally, and the trust being withheld has not in fact been earned.

We are not interested in restoring confidence. Confidence that is not warranted is a liability — to the people persuaded, who then rely on something they should not, and eventually to the institution, whose eventual failure is met with a much sharper reaction for having been denied. Trust is an evidence problem before it is a communications problem. The relevant question is not whether people believe a system is trustworthy but whether someone with no reason to take the operator's word could establish that it is.

So we work on verifiable trustworthiness. Independent audit rather than self-attestation, by an auditor the operator did not select alone. Published performance including failures, at intervals fixed in advance rather than chosen once the results are known. Disclosure of the incentives shaping a system's design — what it is optimised for, and who benefits when it succeeds on that measure. And consequences for broken commitments that were specified beforehand and have actually been applied at least once, since a consequence never imposed is not one. This standard reduces reported trust in the short term, because meeting it means publishing things that are currently not published. We think that trade is worth making. Trust that survives scrutiny is the only kind worth building.