Not directly health related, but I think there’s some learnings available in terms of automated driving safety technologies. https://arstechnica.com/cars/2018/06/why-emergency-braking-systems-sometimes-hit-parked-cars-and-lane-dividers/ tells a good story of how the automated systems in vehicles that manage cruise control, lane alignment and emergency braking are in many cases discrete systems that don’t talk to eachother (so there’s no real ability to integrate their responses). Also may help illustrate the true gap between AI systems and the human element.
A ‘Autocruise’ system which sounds viable is mentioned later in the article seems viable to me, but spoon-feeding every possible piece of intel in order to build a set of responses is quite different from an AI that can dynamically take in new data and evolve its outputs accordingly.