The trouble here is that those pilots were (sadly) never going to successfully fly those planes well in those high-pressure, high-variability cases where an autopilot performs poorly. If you can train a human how to make a judgment call, you can train a robot, so it should never surprise us that robots are best at the things we understand best, and that we’re worst at the things robots are worst at. More old-school practice might make for a slightly better pilot, but probably not enough to make a large difference. Those human pilots are there mostly as a good-enough safety net for when the mechanized idiot-savant that usually flies the planes fails in a novel and spectacular way where a human would not.
Humans and machine work in concert best when the human can leverage the machine to simplify data that still needs a human judgment call: A robot may not be able to interpret the meaning of the numbers in a spreadsheet and recommend a business-specific course of action, but it easily sift through thousands of line entries and sum them up in some easy-to-navigate tables and charts that let the human quickly search the possibilities.
They also work well together when the machine saves time that would normally be wasted on ancillary task. For example, doing the math to design, say, an electronic filter or a mechanical support is tedious. Optimizing the design for trade-offs is even more so (ESPECIALLY when the trade-offs are not linked by one of those nice, smooth, continuous functions where your professor promised you a solution exists). But you can run a much less-tedious simulation, and you can automate the simulation process to generate 10 or 20 different optimizations (or iterate 10 or 20 times to get the best result if you’re lucky enough to have one of those nice, smooth relations. If that’s a task that takes a significant part of every day, then yeah, you’re being automated out. If it’s a task you do once a month or less – when you have to spend a lot of time refreshing your memory or getting up to speed – then the machine is just helping you get it off your plate faster by doing to the fine-detailed remembering and calculation for you.
And some people are inherently rare and in no danger of being fully automated out. For example, we just don’t have enough librarians to do the jobs in a lot of places. When you automate the process of checking in a book, you don’t so much let them cut a librarian (because they don’t – they have other tasks that actually require a human) and you save a manager from figuring out how he’s going to hire an additional librarian who effectively doesn’t exist on the job market.
And a lot of times, this all leads to what Tenner would call a “revenge effect” of technology: as something becomes cheaper and easier to do, you don’t usually spend less time or money doing it: you just spend the same effort to do it MORE. Spreadsheets used to be so tedious as to be very rare. Now we do them every day. Did that create fewer jobs because management can figure out where they stand more easily and get rid of “dead wood”, or did that create more jobs for spreadsheet drones? Hard to say. I’m not going to give you that old saw about buggy whip makers, or the claim that new tech creates as many new types of jobs as it obsoleted (they don’t). But I am saying that some of these jobs go obsolete a lot more slowly than others.
This is not to say that people aren’t going to be screwed. They are. In the very long term, even white collar jobs will be in danger. But that is a bit far off.
