The Full Automation Fallacy
futuresofwork.co.uk
futuresofwork.co.uk
- I can resonate with the author that automation can easily create new tasks (work) that didn't exist before.
- These tasks generally speaking are of a more cognitive nature to the ones they replace.
- If deployed right automation takes out the robot from the human to allow the human to focus on these cognitive tasks which all being equal are more rewarding and satisfying.
- From our own deployment of automation (software in services sector) we have not seen a fall in number of humans in the business where we deploy automation.
- However, we have observed these businesses adopting human capital at a much lower rate than they were in the past as their business grows in favour of technology.
- Labour costs will have to come down a significant amount for to reverse this trend where business are seeing success through automation.
- Automation is a not a one-shot exercise. For it to be effective firms need to establish (literally write down) what their intention are from investing in automation. What is the measure(s) they are trying to optimise and improve.
- Firms need to re-evaluate these 3,6,12 months after deployment periodically and refine the automation machinery. This itself is one of the new tasks to emerge from deploying automation and lends itself to individuals who are more analytical and can make sense of vast amounts of data.
If your interested in looking beyond the obvious with Automation I can highly recommend also reading 'Ironies of Automation Baindridge 1983' https://ckrybus.com/static/papers/Bainbridge_1983_Automatica...
And that's the 'problem' if you ask me. Some of the following may seem to be elitist, which is not what it's meant to be but we have to recognize that not everyone has the same mental capacity.
As a software developer, we like the above. We live for it. We are so "lazy" that we'd rather spend 8 hours automating something than do the same 5 minute task once every day. Heck I'd wager some software developers have spent way more than 8 hours on automating a 5 minute repetitive task.
Thing is, these new tasks that are left over, are now 'beyond' some of the people in the labour force. They just don't have the cognitive ability to perform them any longer.
Take computing and software development or hardware development itself as an example. Almost anyone could be trained to work in a car factory. Sure for some tasks you might be better suited if you're very skilled in certain manual labour functions than someone else. Some of these are taken up by automation now. The people being replaced with this generally would not be able to design and build the robots or design and write the software running these robots though. They're a different set of people. Of course there will be some overlap, because you will have had some really bright people doing some of those jobs that were also bored out of their mind but hey that's the job they could get in 1951... They're probably the guys you will have sticking around to do some of the more mundane but still rather skilled in comparison tasks like 're-parameterizing of the robots' and the like. The other tasks that are left over are now even dumber and repetitive than the ones before.
Or take burger flipping. Literally anyone can do that with minimal instruction. How many of the burger flippers (yes I get that there are some really bright university students flipping burgers for tuition money, let's remove those) could program the drive through wait time counter or design the hardware if given some sensors, an Arduino/rpi and some LEDs?
A lot of the writing around automation has suggested that there will be a single turning point, or a very short period, where all of this job loss happens (because all businesses will adopt it instantly for profit, right?), but that's not really the case even when you find a good automated solution to a problem. If not a single point in time, they say it will happen quickly enough that people can't re-skill quickly enough and will be jobless.
So although fast food workers might fail at programming those things if you show up with a pop quiz, it's not really a good test. What we should do instead is offer them opportunities to be able to learn to do that over time (so long as they're interested in that kind of work) instead of trying to find a way to cull people who can't do it immediately.
Those are the people you can train up, to a certain point. Like you said, there might be people in that workforce that were just never given the opportunity but are actually pretty smart. But there's also the ones that just can't do anything above the level of what they were doing in that factory before, even if given the opportunity and training. There's a difference in training someone to perform X mundane manual task vs. Y mundane manual task vs. actually learning something completely new and different that requires 'thinking'.
It's the same with "traditional" jobs, like say a shoe maker. Today, you mostly have a few cobblers left that know how to fix your typical "suit's" shoes if they bought proper ones (the Allen Edmonds and the like). When shoe production was automated, the actual job of shoemaker disappeared and the replacement jobs dumbed down significantly, so that "anyone" could do them. You still have some actual proper shoe makers left. The ones that in previous times would have be the master shoe makers that invented new ways of sewing or perfected previous techniques, invented new ways of combining pigments for awesome looking shoes etc.
Such a fall in absolute numbers would not be expected if the reorganisation made an organisation more competitive (assuming it's just one of many in its industry), so perhaps you're doing something right.[1]
> - However, we have observed these businesses adopting human capital at a much lower rate than they were in the past as their business grows in favour of technology.
This also is expected. It's what happened with the introduction of ATMs in bank branches.
However those are side issues.
One of the major points of the article is that labour shifting is sold as automation. So we need to know: how much labour and cost has been shifted onto the customers and/or suppliers of the organisations you automate? How much has been shifted out to piecework contractors? What has the effect been on the families of the employees?
Without answers to questions like these we can't really make a proper asssessment of the full social benefit of your automation efforts.
1. A cynical take on automation projects is that they are often opportunities to shake out some of the dead wood that any large organisation accumulates--accumulates because, as other have noted, the incentives within large organisations act to increase the headcounts in various fiefdoms.
The other half of that cynical take is that in other organisations they are opportunities for one group of managers to gather more power over others. This is probably the case in hospitals, where specialist bureaucrats have an advantage over doctors, so "automation" projects shift clerical labour onto doctors (and nurses and patients) while "upgrading" admin work into endless and unnecessary "cognitive tasks".
Only one of those is paid poorly. Hint, it's not the machinist, whose productivity has massively improved thanks to automation.
Software developers are well compensated despite the extremely high degree of automation in their jobs.
The reality is that humans consume a fixed amount of resources and have a variety of jobs that produce value. As long as the value you produce is bigger than the resources you consume you can employ an infinite amount of people in a job. Automation increases your productivity and thus increases the amount of value you can produce.
Those burger flippers are compensated poorly because you need a lot of them to run a restaurant that serves cheap food. If you had a way of increasing their productivity 2 fold you can afford to pay them twice as much. I said afford.
You still need proper government policies that drive unemployment down to reap these benefits. However, without automation you are doomed to a low salary, no matter how well the economy is doing.
I don't see how that's meaningful. If it takes skilled human effort, we expect it to be well compensated. What does it matter whether that skill makes use of much existing technology?
A racing driver relies on an impressive stack of technology. Should we consider their job to be under a high degree of automation?
> humans consume a fixed amount of resources
This could not be more wrong. [0][1]
> Those burger flippers are compensated poorly because you need a lot of them to run a restaurant that serves cheap food
This doesn't explain why they're poorly compensated. It implicitly assumes the competitive survival of the restaurant, and treats employee compensation as a function of the employer's needs. That doesn't work. The employees are poorly compensated because people will do that job for that level of compensation.
Employers don't tend to pay more than the market requires them to.
> If you had a way of increasing their productivity 2 fold you can afford to pay them twice as much. I said afford.
What does it matter what the employer could afford to do, if we're talking about the actual compensation of the employee?
[0] https://en.wikipedia.org/wiki/List_of_countries_by_energy_co...
> If it takes skilled human effort, we expect it to be well compensated.
Because this is true
> The employees are poorly compensated because people will do that job for that level of compensation.
I would rephrase to:
“If the supply of labor is low relative to demand, and the purchaser has the ability to pay, we expect it to be well compensated.”
We don't actually pay people based on skill. Take the example of an old man making phyllo dough by hand - he gets paid based on the weight of the dough produced, not based on the work he puts in or the skill to produce the dough.
This was really prevalent in the Industrial revolution, where expert artisans were replaced by factories.
But not the price you can sell at, unless the demand curve also shifts in proportion to the increase in supply.
> Those burger flippers are compensated poorly because you need a lot of them to run a restaurant that serves cheap food.
They are compensated poorly because there exists other supply of burger flippers willing to flip burgers at that level of compensation, and because burger purchasers are only able/willing to pay a certain amount for burgers.
> You still need proper government policies that drive unemployment down to reap these benefits.
You need government policies to restrict the supply of labor because the increase in supply of labor is not being met with a proportional increase in demand for labor. Hence overtime, vacation leave, parental leave, minimum age, taxpayer funded education, etc.
McDonalds made $11 billion in 2019. They employed 210,000 people. They therefore made ~$52,400 per worker in profit.
A $10 / hour burger flipper makes ~$19,500 per year.
Those burger flippers are compensated poorly because of labor market competition. The value they provide only puts a ceiling on what it is possible for them to earn for their job to exist.
The actual prevailing wage is determined by, frankly, how desperate the low skilled pool of labor is. If unemployment goes up, that means greater desperation. If public sector compensation goes down, that means greater desperation. If social services are cut that means greater desperation. If overseas workers in a country that practices chattel slavery starts taking orders, that means greater desperation. If prison workers start getting in on your industry, that means greater desperation.
All of these things mean wages go down.
This all has absolutely fuck all to do with automation. It is largely about power.
The Roman empire also had mind-bendingly high levels of income and wealth inequality and they weren't exactly automating everybody out of their jobs. They were a society built on the selective application of extreme violence.
Automation does provide the perfect excuse for paltry, declining wages though, largely because it's nobody's fault. If I were an industrialist I'd for sure be gently guiding the pointy pitchforks in the direction of the robots while I outsource, use prison labor, lobby for reining in public sector wages/pensions, cutting social services and gorging myself on gargantuan profit margins. I'm squishy and they are not.
The other useful aspect of the automation narrative is that it plays well with outsourcing. Americans don't see the labor intensive factories that leave its shores in search of more impoverished, cheaper, possibly even slave labor.
What they do see are the factories that didn't bother offshoring - the ones that required a small amount of highly skilled labor to maintain automated production lines, where the headache of offshoring wasn't worth the hassle.
> McDonalds made $11 billion in 2019. They employed 210,000 people. They therefore made ~$52,400 per worker in profit.
Don’t forget that McDonalds is mostly a franchise business. They could have 0 people flipping burgers and still make 10 billion profit
I'll try recalculating for Starbucks, since not having franchisees makes the calculations easier:
Starbucks in 2019:
* Made $26.5 billion dollars
* Employed 346,000
* Average barista wage ~$12 / hour = ~$25,000 per year
* Average profit per employee = $76,589
Hold on a minute, that's Starbucks' revenue, not profit. You haven't accounted for any non-labour expense.
From its financial data (https://finance.yahoo.com/quote/SBUX/financials?p=SBUX) (Yahoo link for convenience), that $26.5b revenue is directly offset by $19b in cost-of-goods-sold (including some salaries already paid). Other overhead and depreciation accounts for another $3.5bn, and interest expenses contribute $235m.
The net pre-tax profit for Starbucks is $3.9bn, and if that were distributed equally among all employees (from your figure) it would result in about $11,200/worker over and above their existing salary.
That's closer to the break-even-wage-rate that Starbucks could conceivably offer, if it's intended to have zero return on capital (about $19bn in current assets for FY2019). And that figure is volatile; if you repeat for this fiscal year you'd calculate about $3200/worker.
When Amazon made zero gross profit for a decade (or was it two?) it wasn't because they weren't making profits.
Check the financial statements again. The $26.5b figure you cite is gross revenue, of which $24.4b is operating revenue. From that revenue you deduct the $19b cost of goods sold (including salaries) to find $7.5b of operating profit, and from that you deduct depreciation and administration expenses.
New investment does not show up as an expense on a financial statement, since it builds capital -- at the accounting level it's a transformation of one asset (cash) to another (land/equipment/etc).
There's lots of money that goes to franchise owners though.
An operation like McDonald's will end up paying higher wages due to automation, but to fewer employees. As the revenue per employee goes up things like training and churn become more expensive than increasing wages.
Even human decency only gets you so far. Restaurant margins are thin, and a different restaurant owner won’t hesitate to start selling burgers at a lower cost if lower priced labor is available. And most people are very price sensitive when it comes to commodity food items.
That is not the reality at all. People in industrialized economies today have, on average, vastly higher standards of living and thus consume vastly more resources than in centuries past.
While a typical person 100 years ago may have been happy with 2 or 3 pairs of shoes, a half dozen changes of cloths, a bicycle and a radio. A person of similar socio-economic status today has a dozen pairs of shoes, a hundred articles of clothing, a car, a tv, multiple radios, a mobile phone, etc, etc. This massive increase in physical possessions for the same (or less) labor output is due in large part to automation increasing the overall efficiency of dozens of industries.
What the "automation will take all our jobs" and "UBI for everyone" proponents seem to fail to realize --just as people who espoused similar views in 1910 failed to realize -- is that there is no limit to human desires. People can never have enough. People will always want more and better. Because people don't measure their wealth in absolute terms, but rather terms relative to those around them.
Thus, even with increased automation, the increased demand will help to provide more employment (if in somewhat different jobs). This has all happened before and will happen again.
However, the article also berates self-checkout systems, which everybody I know perceives as a net benefit to their daily lives. It goes on to pretend that filling in a form on a webapp to send your leave hours to HR is somehow "doing HR’s job", omitting the obvious time savings, persistence, and error reduction for everybody involved. Finally, the article of course freely shows us the AI effect [0] as if it is a novel insight, by quoting a "technology critic", whatever that means, to tell us that actually nearly all achievements surrounding AI are just VC hype and mechanical-turk type shams.
I don't get why everything always needs to be so polarized. When we invented plastic all of a sudden everybody started doing everything with plastic, even if it did not make sense [1]. After a couple decades, some uses of plastic stick, others don't. Its natural selection on a cultural/industrial level. Why can't we allow automation/AI to go through the same process, without pretending it's somehow making our lives worse, or treating it like it's magic.
Dressing the misbehavior of industrial capitalism as a failure of technology is intellectually dishonest.
[0] https://en.wikipedia.org/wiki/AI_effect [1] https://www.pinterest.com/pin/393713192402286646/
It is polarized since people are looking at how it may affect themselves rather than society as a whole. If there a 10% job loss is projected over their working life, they are concerned that they will be in the 10%. Even if their job is secure, they may be concerned that the threat of automation or displaced workers from other fields will force their wages to stagnate or decline. It is difficult to argue these worries away since they affect individuals and no one knows how automation will affect any given person in the long term.
Granted, some of the concerns are about the nature of the job rather than the material loss. Consider something like teaching, where there will be no real substitute for people in the foreseeable future. That being said, what teachers do will change. There will be more emphasis on technical skills and technologically mediated communications. In a profession which attracts people who want to help others, the increasing role of automation will likely be seen as a drawback rather than a benefit.
It feels unproductive to me that a blog sporting such a claim looks at the automation problem, and instead of critiquing our cultural connections of work, wealth, and value, critiques the automation systems trying to improve the future. I think being a luddite is neither critical, radical or a response to anything that is emerging. Its a knee jerk reaction based on fear, like you are saying.
You don't know me, but I want to provide a counter anecdote. My wife and I are both people that see self-checkout as a net negative. The error rate is just too high. They'd be fine if the error rate was low enough so that delays or required assistance was rare. It's not rare.
The best shopping experience is when there are plenty of human cashiers and plenty of lines open.
For instance, for me to understand your problem I would have to know what you mean with "error rate", which I am sure I would not have to ask you if we shared the same implementation.
There's no actual automation in grocery store style self-checkout. All they did was take the same laser scanners that the cashiers had been using since the early 1990s and replace the cashier with the customer (after adding a scale to try and cut down on theft). True automation would be something like the Amazon Go grocery stores where you just fill your cart with items and leave and it figures out how much to charge you. It's the same thing with automated airport parking payment systems - some make you stop, insert the ticket, then insert your cash or credit card (not really automation, just replacing the employee with the customer) whereas other read your EZ-Pass or other toll tag as you just drive through the exit (actual automation).
One level up from this is that you take the scanner with you in the store, and so the process of shopping and checking out is combined into one, again here the benefit is time savings.
The best systems I know allow you to just drop your items in a bagging area and it is automatically scanned, and then you pay. Sure, you still pay manually, but paying is a minor part of checkout, and so most of the checkout is automated.
The systems like amazon are another step above that (skipping the presenting and paying of items), but for me checkout is already a much better experience than it was.
The point is not really that we can't do better, the point is that self-checkout is working, and working well, right now.
EDIT: To get back to the question: is anything actually being automated? I would suggest so, yes. Although maybe its not directly clear, previously the cashier was doing the work of validating that you are paying the correct amount for the correct items, and that you have in fact payed, and so are allowed to take the items home with you. Now that work is shared between me and a couple machines, obviously some of the responsibility, and so work, has transferred from the cashier to the machine, because otherwise the responsibility would completely be with me, and I could take the whole store home with me without paying anything!
I for one absolutely love self-checkout at groceries, self-checkin at airports, etc. The parallel processing benefit is obvious, and ignorance is bliss when it comes to operator errors. And the more such self-service procedures become mainstream, the better we as a species become in reducing those errors. As a frustrated Android user in 2010, I used to think the general population wouldn't be able to figure out smartphones. Boy was I wrong.
It's intellectually dishonest to separate technology from the manner it is used. You introduce a powerful new technology, and people will use it to expand their powers, whether or not it's being used in the way the inventor intended.
I don't disagree - full automation isn't happening anytime soon. We'll stick with partial automation, where 5 people do the work that 100 people did previously.
That said, the author used a few pretty terrible examples to illustrate his point.
The Flippy example from 2018 was an interesting choice, considering after reworking the robot it is currently in use in dozens of restaurants throughout the US and is being trialed by White Castle. Yeah, version 1.2whatever from 2018 wasn't ready yet - that happens. If the cost benefit still makes sense, the robot will be improved and version 4.5 might be the one to succeed. It's not over.
The author writes about doctors doing administrative work - that's just a misuse of the software and a failure of management - there are actually more administrative staff in hospitals than ever before, though they don't seem to do anything additional. This is not because of a failure of automation, but because of a failure to align incentives. The manager is incentivized to increase administrative staff as it enlarges their personal power within the system, and is also not incentivized to make the work of doctors easier in any way.
I think a better conversation would be about why production remained steady while workforces decreased in the midwest. Why Jack Welch's methods became popular despite long-term failures and how we can reset business and investment ethics to allow for a greater wealth distribution based on worker input. What the goals of automation are and how we can move away from employer based healthcare and understand the realities inherent in a system where "Automation...recomposes the workforce, isolating and rearranging tasks, altering job descriptions, and hollowing out middle-tier occupations."
That is shortsighted, given the progress we have already seen.
Many researchers are aware of the shortcomings in AI and beginning to work on them. And that wave of research is growing. It is strange that people discount the possibility of this research paying off. I think they just aren't aware of it.
As AI and robotics improve, it will disrupt the workforce more and more. How that plays out is something that we should consciously direct with foreknowledge.
1. I got the impression that he felt this "fauxtomation" was the final iteration of any service practicing it.
To me, "fauxtomation" follows a fairly established startup methodology of doing things manually before automating. It allows sufficient time for engineers to observe the full scope before attempting to automate. In addition, it is provided as a sort of API to the end user while true automation efficiencies can be implemented incrementally with the in-house team.
The end goal of these services is almost certainly full automation. To a greater extent, I suspect this is the intended future of entire gig economies ("fauxtomation" at scale) created by the likes of Uber and DoorDash who are pushing for further advancements in technology to replace the comparatively high cost of human labor.
2. Where was the "partial automation" alternative to "full automation" in this article?
It seems that most of the described technologies fall short because the scope is too big. They do not allowing enough flexibility for fixing (and learning from) errors or accommodating unplanned use-cases. If a product (like self-service checkout) typically requires training for an employee to use properly, it is not ready for the average consumer.
I have heard the following quote used frequently:
> What can be automated, should be automated.
What is not given enough airtime is the inverse:
> What cannot be automated, should not be automated.
The immediate future is more cyborg than robot. A modern website is an assortment of modular APIs glued together. Some have referred to smartphones as a human appendage, and I think this is the proper analogy to view technology as a whole over the next decade: upgrades rather than replacements to our biological limitations.
Instead of replacing an entire industry with an industry-sized vending machine, large amounts of untrained workers will slowly be replaced by a smaller amount of trained practitioners who know how to use better tools. What this will mean for society as a whole, though, is anyone's guess.
Saying another robotic arm start-up has failed is just shooting fish in a barrel.
> According to an email sent by a chief resident to other residents, Stanford's leaders explained that an algorithm was used to assign its first allotment of the vaccine. The algorithm was said to have prioritized those health care workers at highest risk for COVID infections, along with factors like age and the location or unit where they work in the hospital. Residents apparently did not have an assigned location, and along with their typically young age, they were dropped low on the priority list.
> Stanford has not provided an answer to NPR's request for an explanation of its process or its algorithm.
> Neurology residents were among those who organized and protested on Friday.
> One neurology resident, who was involved in planning Friday's protest, said that a flawed algorithm is no excuse. She requested anonymity for fear of affecting her future job search.
> "[A]lgorithms are made by people and the results ... were reviewed multiple times by people," she wrote in an email to NPR. "The ones who ultimately approved the decisions are responsible. If this is an oversight, even if unintentional, it speaks volumes about how the front line staff and residents are perceived: an afterthought, only after we've protested. There's an utter disconnect between the administrators and the front line workers. This is also reflective that no departmental chair or chief resident was involved in the decision making process."
https://www.npr.org/sections/coronavirus-live-updates/2020/1...