Well machine learning is applied science. But inside there are theoretical work (like encrypted machine learning or optimization theory) and practical work (like text2speech, or other tasks)

Disagree. Theory is always interested in building better, simpler protocols, asking questions about lower bounds and possibility. See almost all of fundamental crypto research.

How do you suppose you're going to provide theoretical assurance of privacy and correctness without science?

Maybe I'm wrong but I see this as applied vs fundamental, not "without science" issue.