Yeah, I sure would hate for a cosmic ray to flip a bit while an instruction is en route, and instead of gently cutting out my appendix, having the robot suddenly deprive me of a kidney.
How likely is that, versus the possibility that the surgeon operator has an aneurysm or seizure mid procedure. Or the nurse hands the surgeon the wrong chart?
I think the point is that no additional software/hardware error is added on top of the human error. Human error + software/hardware error should come as close as possible to an operation using human hands (only human error). Ideally, the benefit of a hardware/software product should produce lower human error too.
Without knowing anything about the hardware/software and transport messaging layer, it could be very likely or unlikely depending on how well or poorly the product is implemented.