Interesting post proposing deliberately avoiding single-sourced IDs: "A pragmatic solution with no unique ID is the opposite. There's no up-front data cleansing. ID is continuously inferred, the system copes with duplicates, and structure is applied retrospectively. Ther's a presumption of ambiguity which makes us proceed with appropriate caution. Quality increases with volume and over time. There's little up-front cost. "
This is pretty well what we have now, isn't it? Driving license, electoral register, NHS records, etc. all contribute to identity.