I think because it supports all “conductive” surfaces, which in practice will exhibit slightly different capacitive properties. Machine learning is one approach to derive the mathematic effects of these qualities on voltage signals while the user enters a training sequence on the surface.