we're already in this regime for airline pilots and (anecdotally) the results aren't great; the french airbus that went down over the ocean a few years ago was because of pilot error after an emergency condition (icing) disabled the autopilot.
I don't know if that's the typical case for autopilot tagging in its human, but it seems difficult to motivate and train experts who will only work 0.5% of the time. Expertise comes from experience.
Unless the pilots are required to be in the simulator the whole time the autopilot is running the plane, which would be interesting and cool.
We also have no idea of how fast autonomous systems will 'evolve' (with programmer help at first) and 'learn' once authorized to roam free. For their constrained domains, I suspect they'll be up to 'expert' performance pretty quickly; plus the rules will change as reaction times go down, attention per vehicle improves, and communications improve.
SFAIK, a "patch" to the icing problem has been deployed. I suspect it'll be like that - undiagnosed defects emerge, get patched, repeat, lather, rinse. Somebody once called this "like simluated annealing" to me ( I lack the materials chops to say if that's true or not ).
The problem with this is that it's very hard to communicate the realities of this to "end users". They may have Luddite streaks to start with and may be looking for an excuse to ... defect from the relationship with the machines for very real reasons that have nothing to do with the actual operational virtues of the machines.
You can't really tell them that they are wrong, either.
The fact that you and I both know the icing story is evidence that this is unusual.
Then again, is a system that works 99.5% of the time and delegates stuff to humans 0.5% of the time more or less reliable than a system that relies 100% on humans, with their microsleeps, inattention, mistakes etc.? Has air travel, e.g., become mostly less or mostly more reliable by means of automation?
Hmm -- the question shouldn't be 100% human vs 0.5% human, the question is 0.5% human vs 0% human. Can somebody helicopter into a new situation and make a decision quickly about what to do.
Flight may be a bad example because the time constraints are so short. Changing the scenario to an automated fedex truck lost on a rural road, the truck can pull over and wait a few minutes for help. In this case the solution probably isn't to tag in an expert but rather to call the delivery recipient for directions (same as a human driver would do).
This is so spot on.
Before were able to automate away all this work we'll augment peoples abilities with really smart tools. This concept will end up not necessarily destroying entire job sectors but their job definition will be dramatically different, in the same way a computer used to be a room full of people doing math.
But the story of why people have jobs they hate - if they have jobs at all - may need a villain . You can't ignore narrative reasoning, no matter how flawed it is.
I don't know if that's the typical case for autopilot tagging in its human, but it seems difficult to motivate and train experts who will only work 0.5% of the time. Expertise comes from experience.
Unless the pilots are required to be in the simulator the whole time the autopilot is running the plane, which would be interesting and cool.
We also have no idea of how fast autonomous systems will 'evolve' (with programmer help at first) and 'learn' once authorized to roam free. For their constrained domains, I suspect they'll be up to 'expert' performance pretty quickly; plus the rules will change as reaction times go down, attention per vehicle improves, and communications improve.