It is very difficult to compare the safety aspects of automation with manual flight. An aircraft is designed, built, and tested against known and tried standards, often hard learnt from experience - (FAR / CS 25); the human design and operation has no such reference book.
Both technology (automation) and humans have their positive attributes, each able to excel in certain situations, but whereas technology alone might provide a known standard of reliability, the human is subject to error (so too is technology, as discussed previously via the human in design – so triplicate, etc).
Automation has been used where either the human physical and/or mental capability is limiting, e.g. autoland in low visibility; the activity is driven by the situation.
Humans are good at solving problems, etc, the unknowns, although still subject to error.
As the general operating situation (the state of the industry) puts more pressure on the human, the apparent error rate increases; this is often interpreted as declining skill, but actually it is a sign of reduced safety margins. Thus, looking at the situations in which automation and humans work and their respective safety margins might provide a more meaningful comparison.
If the industry continues to demand economic efficiencies which constrain human flexibility and approach the limits of physical and mental capability, then automation, with an appropriate level of reliability for the situation has to be considered, e.g. Auto flight, RNAV, RVSM.
Alternatively, if the industry still needs the human flexibility, then the situation in which the human operates must be constrained to accommodate the limits of human performance, and the susceptibility to error.
Unlike automation, human reliability cannot be specified, thus the situation has to change to take account of the human.
I suspect that the practical solution lies somewhere inbetween these alternatives – technology and the human working together.
In current operations, we have automation and we have an industry ‘situation’; perhaps the problem is that we haven’t fully understood where the limits of each are and how to adjust reliability and situations to maintain a high standard of safety. We need to improve our understanding of both ourselves and the ‘situation’ (and equipment) in which we work.
A significant human attribute is that we can ‘create’ safety by manipulating situations, choosing where and when to use automation, but only if the background situation (crew selection, training, organization) provides sufficient margin from the limits of human performance.
If the reliability of automation is such that failure requires human intervention, then the residual system, indications, and operating environment must enable realistic human performance, this includes the level of training, e.g. Cat 3B autoland degrades - land - as the system is reliable; Cat 3A autoland fails – GA (perhaps the alternative of a manual landing has a reduced safety margin – pushing the boundary of human limits).