[netcan]

I'm not the OP, but I think he means this:

A bookkeeper/accountant might be manually entering invoices into a system, emailing managers to approve payments and such. It takes 3 hours per day. This work is "routine" in the relevant sense (it can be automated) but our young, congenial and soon-to-be-replaced-by-robots accounts payable woman (her name's Lynne) also does and is capable of doing "non-routine" tasks. She can call up suppliers to negotiate payment terms, have lunch with the manager in the next office to clandestinely suss out some questionable expense charges, etc.

Calling Lynne a "routine worker" is kind of problematic. She does routine work because it needs doing. When that work disappears, she's not out of work by default. Generally when technology creates efficiencies, the unemployment rate does not just increase by the number of people that did that work yesterday. People who are capable of doing other useful things do those things, if it's useful to do more of those things. There's still a question of whether the economy can use up these newly available Lynne hours. Often it can.

What happens when some technology automates a lot of "routine" programming work? Programmers make more program(s).

[mapt]

"Generally when technology creates efficiencies, the unemployment rate does not just increase by the number of people that did that work yesterday"

Perhaps not, but it does increase. Because somebody else was calling suppliers and having lunch with the manager yesterday.

[ryandrake]

Not necessarily. As a stereotypical "knowledge worker" my current work activities can be classified very broadly into two buckets: 1. Low-value routine grunt work that I am required to do and 2. High-value non-routine work that is ultimate optional but the more of it I do the better results I get at my job. I try to spend at least 10-20% of my work time figuring out ways to automate more and more of bucket 1 so that I have more and more time to do bucket 2. There's an unending amount of bucket 2 work available for me to do should I have time for it (trust me).

And here's the magical thing! Once all of bucket 1 is automated, a lot of my bucket 2 tasks start becoming routine--patterns emerge and I can imagine ways to even chip away at them with automation, allowing me to do even more high-value work, in theory making me more valuable to my employer.

[paulryanrogers]

My guess is this varies widely depending upon the business. Some are more compatible with automation than others.

[netcan]

Dynamic effects are really hard to talk about deterministically. Generally speaking, technology is always creating efficiencies and unemployment usually ranges in a range, over the long term.

Some technology advances seem to reduce unemployment. When factories were doubling efficiency several times in an average career-span, employment in manufacturing kept going up and so did pay. This was a long period. It started with farmworkers (replaced by automation) urbanising to work in the rapidly advancing, high tech manufacturing sector. The factories kept automating more and more. Efficiency went up. During long periods labour got a good cut of these gains, salaries rose.

At other times and other technological events (including trade which is economically very similar to technology) unemployment went down in the sectors closely linked to the tech.

[hnal943]

That's an unfounded assumption. There's lots of work that goes undone because it's cost prohibitive. Technology in this case increased overall value because it allowed Lynne to do something that no one was doing before.

[mapt]

If it was cost-prohibitive yesterday, why isn't it still cost-prohibitive today?

Maybe it would be doable at a lower salary. Let's pay Lynne half as much as yesterday for this new job.

[hnal943]

Because technology gave us levarage, and lowered the cost. Lynne's pay is ultimately set by the market. Lower her pay and see if she leaves to make more somewhere else.