Population level statistics has lead to a large number of algorithms that have been proven wrong for medical conditions.
BMI, dietary cholesterol, vitamin D, etc...
Auto differentiation is an incredible tool, but I don't see how it will truly be a 'miracle' like vitamin D, which in hindsight is clearly over fitting combined with a correlation that wasn't acausal connection.
ML is extremely powerful, but not for individuals for the same reason statistics is great for populations but not individual.
This is the 'common sense' problem that both fields struggle with and need to correct for. Ethical problems with empirical testing in humans complicating the problem when applying ML or statistics to medical problems complicates that too.
Personalized intake recommendations.
Comprehensive health assessments from various health sources (labs, wearables, etc).
Monitoring vitamin trends through health assessments w/ time & overall wellness.
Reading and applying latest research/clinical trials to all the above.