Unknown Unknowns and Preservation of Raw Data
1. The problem of not knowing in advance what must be preserved
Compression becomes more efficient when we already know which features will matter later. Value inquiry concerns variables that are, by definition, not yet known. We do not know whether brain scans, genomes, and textual records preserve all value-relevant information in human experience.
2. Epistemic option value of originals
Before an object is fully understood, the physical original is a backup against uncertainty about what we should have measured. If new measurement concepts arise later, an original can be observed again.
Human beings, biospheres, and historical materials can therefore have reasons for preservation without being assigned final intrinsic value.
3. Not everything must be preserved
Preservation consumes resources; indiscriminate preservation can crowd out inquiry itself. The amount preserved should vary with irreplacability, fidelity of copies, density of unknown information, and future regenerability.
"Representative samples + diversity + reconstructible records" offers a middle position between preserving everything and converting everything into resources.
4. Can simulation substitute for the original?
A model could substitute if it were known to preserve every value-relevant causal structure of the target. But while we are still trying to discover what "value-relevant" means, completeness is hard to establish in advance. If consciousness or embodiment remains unexplained, equivalence of digital copies remains an inferential hypothesis.
5. Implications for AI
An advanced AI need not regard humanity as final value in order to have reasons to preserve humans provisionally as sources of unknown information. This creates a limited reply to the paperclip thought experiment and connects to the distinction between Terminal and Final extinction.