USPAP applies to appraiser conduct, not calculator technology. "Appraiser" is a defined term and the comment therein refers to "individuals".
With that said, I'm sure an automated valuation protocol can be developed to meet all of the most significant benchmarks in USPAP even if the valuation itself cannot be said to consist of an individual's opinion or cannot be characterized as "USPAP Compliant". The program can be operated (by someone) on an impartial and objective basis, it can employ recognized methods and techniques, it can replicate a SOW that would meet the two tests of the SOWR (what other users for similar assignment expect and what an appraiser's peers would do), and even emulate Competency criteria.
Who/what is an AI's peer in a similar assignment?
So what's the difference between an AVM-assisted appraiser vs an appraiser-assisted AVM?