Since the industrial revolution began there has been speculation about the long-term effect of automation on employment. You can create plausible-sounding but completely opposing just-so stories: that automation will free humans from drudgery to focus on more important work or leisure, or that automation will destroy employment. The first argument is one made by the now silly-seeming futurists of the 1950s who predicted a 20-hour work-week for Americans.
That mid-century naivete works fine if all you want is to maintain 1955's level of productivity - but these decisions do not result from central planning. They result from individual humans that seek status and are competitive and want more wealth if there's more wealth to be had. (Fortunately, this is the path that results in economic growth, rather than stagnation and class sedimentation.) Consequently, the real work-week actually grew slightly in the second half of the twentieth-century.
The second argument - that the machines will put us all out of work - is a more pessimistic version of the same argument. In this scenario, the 20-hour work week exists not because that's all we have to do to maintain our current economic growth rate, but because there just isn't any more work to be done - the machines are doing it all.
There are examples in our past we can point to. Yes, there were labor shocks in the U.S. in the 1970s and 80s, due to automation and offshore competition; yes, the populations of many Midwestern American states are lower today than they were a century ago, because those economies are mostly dependent on agriculture, and one person with machines today can produce what many did a century before. But our population as a whole grew, and became richer in that time, even if in certain industries or geographic areas at certain times there were disruptions. No one in 2009 is complaining that the beaver pelt industry is moribund, because people learned other trades besides trapping. Likewise there are industries besides agriculture, and as economies expand, new industries appear, and with them new needs.
That's why, it's hard to see why we'll suddenly all be out of work, unless we're talking about full self-repairing AI, in which case we'll have other problems. There are many more examples of automation expanding work that we can point to now. What is most relevant for the future is "white collar automation". It's certain that the automation of physical labor that the world experienced in the last century and a half (and which it's still experiencing in many quarters) is not the end of the road, and that computers will start automating white collar tasks much more than they already have. (Forget about outsourcing to India.)
Have there already been "cognitive" tasks automated where we can gauge the impact on the industry in question? Yes. That computers and calculators are widely used does not seem to have put accountants out of business - except, of course, for those accountants who refused to learn how to use a computer - and now one accountant can produce what many did a century ago. The same is true for other white collar professions. The answer is that in the future, on some level, we will all be programmers. (Compared to our parents, aren't we already?) We'll use the much-afeared automation as additional tools to increase our productivity. Just don't be that guy in the office who won't learn how to use Excel, and you'll be fine.
One challenge to the coming white collar automation blitz is the existence of administrative sinecures. The invisible hand has clearly not yet done away with them in the private sector, nor in government or academia. Since their existence flies in the face of economics, it is likely that their persistence is best explained
by universal irrational aspects of human psychology that will exist even in competitor organizations. Once the expert systems start coming on-line in earnest, will these positions survive? Will they, over time, make them seem more and more suspicious and non-contributory? One down-side to the new automation is that sinecures are not quantized; it's likely that some percentage of your time is sinecurity, and the inefficiencies that allow it to persist will be erased, or at least known about. Your little sick days to catch up on soaps? Your argument that you really need to audit that company in Miami? Your coworkers' simulations will catch it.
Alternatively, it's possible to imagine that while producers are staying ahead of the automation blitz by learning to use it to produce, will sinecurists will learn new strategies to justify why their positions are "more art than science" that
can't be reduced to vulgar numbers by mere machines.