Turbulence in the Flight Path - Job Elimination?

The New York Times posted an essay yesterday (prompted by the Watson Jeopardy Challenge) that explores the question of whether increasingly powerful AI will cause major disruptions in the job market, or if human adaptability combined with "IA" (intelligence augmentation) will let us keep finding useful things to do - for a while, anyway.

One example they give is how rapidly ATM's replaced bank tellers, even for a task that generally involves physical interaction - giving or receiving various pieces of paper.  I'm not sure that's a telling point - true, there are an increasing number of online-only banks who don't employ a single teller - but even at physical banks with ATM's right outside, they have plenty of tellers.  I know I can deposit a check at an ATM, but when I (rarely) get a paper check these days (thanks, mom!), I still go to a teller to deposit it.  My gut rationale for that is I expect it to process faster, since I presume they empty the ATM at the end of the day and then do all the stuff inside.  Though I seem to recall something about ATM's that actually scan/OCR/process checks in real-time now (and a similar smartphone app).  Still, plenty of tellers around from what I'm seeing.

But the larger "threat" is in areas where there is no practical physical interaction by default - customer service, mostly.  According to Wolfram Alpha (go IA), as of 2008, there were 2.233 million people employed in this position.  How much of these jobs can be automated now?  I think the problem for the moment is validation...anyone can take a stab at some heuristics for what to do in what situation, but I don't see a CEO turning their operations over to WatsonCS without some data that suggests there won't be widespread customer revolt as a result, and collecting that data seems likely to be arduous and expensive.  But Customer Service is, in a basic sense, about trying to achieve some defined minimum average level of satisfaction for customers who contact the company, with the least expense possible.  Based on the iffy quality of service I get from humans these days, it seems like it'd be hard for AI to do much worse.

It's going to be interesting in the next few days/weeks seeing conclusions drawn based on Watson's performance.  I wonder how many of them will take into account the fact that Watson might be capable of much, much more than it shows, given the rules of the challenge don't allow an active Internet connection.

It's sort of like debates about enhancements in sports.  Some of those rules are about protecting the athletes, since some enhancements can be dangerous, but some are arbitrary, aimed at protecting the "purity" or "sanctity" of the sport/game/contest. When are contests where increasingly available technology has to be excluded to keep things "fair" going to stop being as relevant in our society?  I've never been much of a sports fan, but I'd probably be more interested if safe enhancements were making things more exciting.  I think it's telling that a lot of people these days watch the commercials more than the actual Big Game.