Given how many bugs (no pun intended) the average programmer is capable of inserting into programs, I'd be much happier with an adaptable-to-cockups human than a pesky "computer says no" entirely automated solution.
Unfortunately, medicine today isn't deterministic. A lot of decisions are made with 'instinct' -- a fancy word for intangible experience. Human error is always going to be an issue. What might prevent human error in the design of said automated systems?
I can agree with that. But you've made two implicit assumptions here:
1. The disease is diagnosed and identified as 'common'
2. There are no other complications because of say
alcohol abuse, or a genetic condition with kidneys and so forth