When ATMs were introduced to banks in the 1980s, people thought bank tellers would quickly stop existing. Over time, each bank branch did end up employing fewer tellers. However, ATMs made it cheaper to operate bank branches, so more bank branches opened and there were eventually more bank teller jobs over all.
This comforting trend ended several years ago.
Here's a blog that plots teller employment numbers up through 2015:
Teller employment peaked in 2007 at over 600,000. It's down to 472,000 as of 2018 and projected to decline another 12% by 2028.
You need to keep following the numbers to distinguish "the prediction of technological unemployment was premature" from "the idea of technological unemployment is mistaken."
> Teller employment peaked in 2007 at over 600,000. It's down to 472,000 as of 2018 and projected to decline another 12% by 2028.
By that point we're not talking about the effect of ATMs anymore, which were introduced in the late 1980s/early 1990s.
This effect is more likely due to online banking and banking apps. We don't know yet how this change reallocated
> "the prediction of technological unemployment was premature" from "the idea of technological unemployment is mistaken."
Not really.
If you look at the other example (spreadsheet apps and bookkeepers) you see that even if a job stops existing, the remaining tasks get allocated elsewhere over time.
This overall dynamic is supported by the researchers in labor economics (see papers linked in the post)
This doesn't mean the transition is smooth, or even benefitial, for the job holder, but it doesn't translate to long run unemployability.
Economists often assume unlimited human wants and limited resources, giving rise to scarcity. Under the assumption of unlimited wants it is perfectly reasonable to predict that 40 years from now paid employment will remain as common as it was 40 years ago even if we don't know what the jobs will actually be.
I have a somewhat different perspective. I believe that human wants expressed in market economy terms -- e.g. "market demand" -- are satiable and becoming increasingly satiated among households with buying power (top two income quintiles in the US, for example [1]). There's limited demand for more consumption of services that already existed in the 20th century. You can only go to the doctor, the dentist, or the barber so often before more-frequent visits become a chore instead of a luxury.
New inventions may spur different demands but scope for aggregate growth is limited by hours in a day. Scrolling through Instagram replaces going to the movies and maybe some day exploring virtual reality replaces scrolling through Instagram. That reallocates entertainment demands but it doesn't necessarily mean more aggregate consumption or entertainment-employment.
There is big growth coming in home health aides, elder care, and similar jobs. But that's due to an aging population rather than the development of new service jobs for the 21st century.
Big growth in demand for "gig economy" services? Maybe. We'll see after the VCs stop tolerating losses.
[1] The unmet wants of the lower income quintiles are less significant in a market economy because they don't have growing buying power. Two former bank tellers aren't going to find employment walking each others' dogs.
Even if the decline in teller employment is not due to ATM's, the parable still suffers if another convenience innovation came along and did kill jobs.
We don't know if innovation killed jobs here, we only have a graph showing there are less tellers.
It is expected that specific job titles will be obsoleted by technology. That is not a major issue when talking about the net effects on jobs. Society is littered with job titles (like Smith, Turner, Cartwright, Ackerman, etc, etc) that used to be everywhere and are now specialist roles or extinct.
The problem isn't that automation kills jobs. Those jobs reallocate over time.
We've seen automation act on the economy since the 1820s and nothing about deep learning or big data changes these dynamics.
The problem is that those who do lose their jobs have their lives altered for the worse, and those who get the better parts of automation don't share the benefits with the ones losing from it.
Their inflation index is literally the official CPI index plus a constant (~2.8 last time I looked). He's that lazy about his scam.
The maintainer of shadowstats isnt in the business of maintaining an honest set of metrics, he's in the business of selling misinformation around hard to measure things.
This is exactly what happened but I assumed that the real cause of more branches was different. When banks stopped having to keep banking and investments separate, they wanted all the customers they could get. When you are averaging 10+% in the stock market while only paying 1% to consumers, you want as many consumers as possible. Then the only reason people would switch would be if a specific branch was more convenient for them or had more free ATMs. This became a question of customer acquisition. Plus you can write off the building and have a solid real estate business (ala McDonalds) for free. This also is what has pushed the banking consolidation trends so much. Banks could no longer get new customers via switching so they starting buying all the local players to keep their balance sheets growing.
Also worth noting that this is absolute number of tellers, not tellers per US resident. Automation is very clearly reducing the demand for tellers on a per-potential-customer basis. For example, if we had seen the ratio remain flat, starting in 1980, we'd have expected a 50% growth in the number of tellers. But in fact it was flat (at least on the original article's graph).
I can't find numbers on what fraction of the population was unbanked in the '80s, but I can only imagine it was higher than it is now. That also is not accounted for in the parable.
As someone who has worked at a financial in a rural area, I can assure you these people are still going to need tellers. I mean can you imagine explaining daily why someone gets charged an NSF fee or telling them that their check is a fraudulent check (sent by a scammer)? Also these people hate technology more than communists. So trying to sell them online banking is like telling them to become atheist socialists ready to fight for the revolution.
I'm a swede, who lived in the US 2009-2011. Back then I was amazed (and my friends as well when I told them about it) about how behind the US were tech wise when it comes to banking. From what I understand, we're in the forefront in Sweden, and now with all the bank startups, it's just taking off for real.
In Sweden the number of offices almost halved between 2007 and 2019. It takes time for these kinds of shifts to really take off. As a tech guy, I believe tech guys are often to optimistic about how fast non-tech _humans_ adopt tech (not because of tech, but because the change in lifestyle, which takes a lot of time for people). It's all the standard curve of adoption (first hype, then dispair, and then it truly happens and diffuses into massmarket).
Another note on that -- this is with increasing population and as they say, more bank branches opening than ever, so the automation effect has been very strong.
"Similarly, as tasks are automated in the modern economy (such as manufacturing tasks) workers will shift their time into other tasks like the growing service economy"
I like to call this the "lump of labour fallacy fallacy". It falls down because complete fungibility isn't actually a thing. You can't turn a dumpster diver into a brain surgeon, no matter how many hours of training you apply. As tasks are automated new jobs and tasks are created generally at a higher skill and intelligence level.
This excludes an increasing fraction of the workforce all of whom still need something to do of service to others for their own mental wellbeing as well as to fulfil the social contract requirement of reciprocity.
Exactly. New jobs will appear, but they won't be jobs that most of the displaced workers can get, or reasonably be expected to get. Do we leave them on the wayside, or do we treat humans with dignity, and realize that life is more than work, that unemployment is not always a problem, it can be the solution.
> It falls down because complete fungibility isn't actually a thing.
Agreed.
Studies on those who lost their jobs in the manufacturing towns (Autor, Dorn & Hansen 2016, linked in the article) shows that losing your job to automation is a harrowing experience.
The jobs reallocate over time but it's a bumpy process that can ruin the lives of those affected.
> As tasks are automated new jobs and tasks are created generally at a higher skill and intelligence level.
Not really. Automation tends to hollow out middle skill jobs, leaving high skill and low skill jobs available.
The jobs reallocate over time but it's a bumpy process that can ruin the lives of those affected.
This is why broadly speaking the historical Luddite's were right, from their own perspective. It may the best thing to do for an overall economy, but pretending people won't get screwed over in the process isn't doing anyone favors.
Not addressing this fundamentally as a matter of policy is a choice. I'm not saying it is the wrong choice, just that being transparent about what that means would seem to be the best course.
being transparent means the admission that public policy has failed the vulnerable and those in need. I dont believe this problem will ever be "admitted to" - and it's more likely that the gov't will let it "blow over" (read, let those who got automated out of a life to fend for themselves, and eventually they either die/find new opportunities on their own, or the family slack picks them up).
same things happens on stock markets - shrinking number of mid-cap (and overall number of) companies,and growing influence of the largest corporations leads to inequality of stock markets.
few winners, much more losers.
as mediocre companies go bankrupt, people who cannot catch up to changing job market leave job market as well, or remain underemployed & overworked on several low-paying jobs
There are issues with some of the earlier points, for example:
> We automate tasks, not jobs
We automate both. One popular example is trucking. There are 3.5 million truck drivers in the US alone. And of course this is an area where complete automation is on the horizon. The tech exists, it's getting close to ready for large scale deployment and big players are working on it.
The industrial revolution comparison is apt but only to a point. The rate of innovation in tech is speeding up and our ability to automate jobs away faster than we can adapt along with it.
That said the article goes on to focus on what I think is one of the key points in the automation conversation: wealth inequality.
If we don't come up with a way to share the wealth produced by increasing automation, the current system will funnel almost all of it to the top.
> If we don't come up with a way to share the wealth produced by increasing automation, the current system will funnel almost all of it to the top.
I think this is a discussion that requires more nuance than is generally used. I think you're talking just about money here, but I'm not sure.
Without that nuance, there's a good argument to be made that the current system that has made powerful computers accessible to and affordable to all but the most impoverished in the world is an excellent example of sharing wealth. The industrial revolution similarly made cheap textiles available and cheap shipping has dramatically cut food prices.
All of those are arguably sharing wealth. But none of them are sharing money.
On the other hand 'trickle down economics' has been pretty categorically debunked by reality; it can't not be about money entirely either.
So it isn't all about money, but if the monetary gains of increased productivity are captured by a shrinking fraction of the population, nothing good happens in the long run.
Exactly. Even if wealth is shared, and all humans technically receive everything they need to live life, that does not mean the world is avoiding assorted moral hazards, such as "human zoo" condition. Nor that the human condition on the whole is optimal.
100% agreed. Automation both requires capital and reduces ongoing expenditures due to salaries, which has the combined effect of a higher barrier to entry for people launching a new business where high automation is required to be competitive and at the same reduces redistribution of the value added. So it’s not so much that money is funelled to the top, but maybe more like it stops trickling down...
> We automate both. One popular example is trucking. There are 3.5 million truck drivers in the US alone. And of course this is an area where complete automation is on the horizon. The tech exists, it's getting close to ready for large scale deployment and big players are working on it.
Assuming you are right, I wonder if this couldn't eliminate trucking altogether given a bit of time: with greater concerns for the environment, governments have tried to pull measures to reduce trucking, but have often be opposed by truckers.
If self-driving eliminates truckers, who will then defend trucking?
Automation has interesting dynamics in the complex system that is today's society. I am not sure if the situation described above holds for the U.S.A. for instance, but it matches my understanding of the one in France, at least.
> And of course this is an area where complete automation is on the horizon. The tech exists, it's getting close to ready for large scale deployment and big players are working on it.
That's incorrect. We're really far off still from level 5 self driving cars if you follow developments in that area.
However, the major task in the truck driving job (highway driving) is automatable already. Other ones (urban maneuvering, loading/unloading, etc.) are much harder to automate.
> That said the article goes on to focus on what I think is one of the key points in the automation conversation: wealth inequality.
I'm the author, thanks for echoing my reading on the dynamics.
People focus on an imaginary problem when a very real one is happening underneath our noses.
The rate at which we're approaching full self driving capability (especially with trucking where, for many routes, it's almost all freeway driving) is fast enough that it's safe to say those 3.5 million truckers will not be ready for the change. Even if it takes 10 years they won't be ready for it.
And that's just the easiest example. Past revolutions which have decreased the cost to provide basic needs happened considerably slower than the digital revolution is happening.
I don't think it's an imaginary problem at all. Overblown yes, I completely agree... but if that's what it takes to keep the conversation moving, I'm ok with that.
The urgency, I think, is not "all the jobs will go away tomorrow", but that it's a big enough problem that the time to start solving it is now. And unfortunately the world is really loud these days, it takes volume to be heard over the noise.
This is like reading a global warming denier. The earth has heated up and cooled down before. Global heating has never been catastrophic before so why would it be now?
This guy literally uses graphs and charts from the industrial revolution and and generalizes the results over to automation as a concept. And then he confidently states that this proves that the concept of automation benefits the highly educated. Yeah, well when high level signal processing is automated that won’t be true will it?
Just like global warming, this isn’t a problem that can be understood by looking at localized effects in the recent past. It’s a problem that can only be understood through first principles reasoning. Greenhouse gasses heat the planet. Therefore global warming is a thing. Humans are high level signal processing machines. Therefore the automation of high level signal processing will displace humans in the global market. It’s just inescapable. And yet people mental-gymnastic their way out of it.
And whenever I see “alarmist” in someone’s argument, it’s never a strong argument. The alarmists are right.
It's the opposite. The reading from the consensus experts is that the labor effects of automation aren't unemployability, it's worsening job prospects and increasing inequality.
Those are different things which require different solutions.
> And then he confidently states that this proves that the concept of automation benefits the highly educated.
I'm the author. I linked to 30 or so sources in the article, most of them academic publications from the most respected researchers in the field (David Autor, Daron Acemoglu, etc.)
The article is basically a literature review and vulgarisation for laypeople audiences. It's not some theory I just came up with.
> It’s a problem that can only be understood through first principles reasoning.
3 papers by Daron Acemoglu linked to in the article do that and test their theory against recent trends to see if it holds.
Wow, hello. I’m about to clock In the for the day so I’m afraid I can’t address your sources until later tonight. But in the meantime, saying that other people think this is true doesn’t advance your argument in my mind. I don’t care what “experts” say, although I have no problem picking it apart, because there are no experts in AI that doesn’t exist yet. Nobody has first hand experience studying a real world example of what we are talking about. High level signal processing is the only thing that humans monopolize. The more of that that gets automated, culminating with AGI, the closer we are to the obsolescence of humans. It’s just blindingly simple. And please don’t call me an alarmist. I’m genuinely concerned here. I don’t ruminate in this because it’s fun.
> Humans are high level signal processing machines.
True, but that's not all humans are. They are also social. They are also creative. At least some of them are also deep logical thinkers. So...
> Therefore the automation of high level signal processing will displace humans in the global market.
That might not be the catastrophe you make it out to be, because humans are more than just that. Humans that have jobs that only expect them to be high-level signal processors are, I suspect, rather rare, maybe even nonexistent.
> Job automation underpins the explosion of economic growth we’ve seen since the 1830’s industrial revolution.
What underpinned the explosion of economic growth in the 1830s to 1850s was opium money we stole from china and whale blubber. What underpinned economic growth since the 1850s was oil and population explosion along with dumping large populations in small areas in order to boost consumption.
We know for a fact that automation doesn't underpin economic growth because if it did, Japan and most of europe would be the centers of economic growth. Why has automating Japan lagged so far behind china or the US since the 90s? Their population has stagnated. While China added 300 million more people and the US added 100 million people in that time, Japan added hardly any.
Here are the list of the fastest growing economies. Automation isn't the first word to come to mind when I think of many of these countries.
Economic growth is essentially population growth + moving that population to cities so that they can consume more efficiently. If automation truly underpins economic growth, that list of GDP growth rates would be led by primarily america, european nations and a few east asian nations.
This comforting trend ended several years ago.
Here's a blog that plots teller employment numbers up through 2015:
https://greatdisruption2018.com/the-project/the-parable-of-t...
Here's the BLS data and outlook for the future:
https://www.bls.gov/ooh/office-and-administrative-support/te...
Teller employment peaked in 2007 at over 600,000. It's down to 472,000 as of 2018 and projected to decline another 12% by 2028.
You need to keep following the numbers to distinguish "the prediction of technological unemployment was premature" from "the idea of technological unemployment is mistaken."