The statistical relationship between skull and face that AFA3D learns from modern French CT scans may or may not approximate the relationship that held in Upper Paleolithic central Europe.

The AFA3D approach has significant advantages over manual methods.

It is replicable: given the same skull scan and the same database, the algorithm produces the same result.

It is not subject to day-to-day variation in an individual practitioner’s judgment.

It sidesteps some of the specific subjective choices — about nose shape, ear shape, lip form — that human reconstructors must make.

And the geometric morphometric framework allows the uncertainty in the reconstruction to be formally quantified and communicated.

But the choice of reference database is not a subjective judgment call that the algorithm makes — it is a decision made by the researchers who design and deploy the system.