Eliezer Yudkowsky asks about automation

From the MR comments:


(Tyler, have you read that?)

I don’t actually get Brynjolfsson and McAfee. I read the original book and it seemed very unoriginal and not to address at all the basic question of “Why did Ricardian reemployment work fine when agricultural jobs went from 95% to 3%, work fine when automobiles put the whole horse-and-buggy industry out of existence, work fine when women entered the workforce during WWII, and then suddenly stop working?”

The Ricardian comparison is the technology-relevant one here, and I would challenge the notion that it went fine.  Think of the machines of the industrial revolution as getting underway sometime in the 1770s or 1780s.  The big wage gains for British workers don’t really come until the 1840s.  Depending on your exact starting point, that is over fifty years of labor market problems from automation, and yes it is correct to also blame various bad laws, mobility restrictions, wars and taxes, and the like.  Even Ricardo, very much a free market economist, worried about the machinery question in his day and rightly so.  The industrial revolution was a wonderful development with huge ongoing gains, but still it did bring some very real adjustment issues.

A second point is that now we have a much more extensive network of government benefits and also regulations which increase the fixed cost of hiring labor.  Insofar as automation creates short-run adjustment problems, those problems are more likely to show up in the form of decreased labor force participation than they did in previous eras.  We are living in a time where the long-run trend is for labor force participation to fall in any case, and that was not in general the case during those earlier episodes.

You also might try to run with a “back then machines substituted for brawn, now they are substituting for brains” argument.  Maybe so, but you don’t even need to make that work to have a substantial (non-Luddite) worry.


Another way to look at the effect of mechanization is to look at how it affected the other living employees of farmers. The U.S. horse population peaked at 26.5 million in 1915. It declined rapidly after that, hitting a low of just over 3 million in 1960. While it is about 9 million now, that's because of increased ownership as pets.

I'm not saying humans will be destroyed like horses, but it raises some questions about the ease of transition.

Wow. Great comment.

(sorry if my tone is not apparent: I intend sincerity)

Farmers had to allocate up to a quarter of their farmland to feeding draft animals.

Shifting this land to cash crops contributed significantly to farm productivity.

Of course, Spencer, but that's not my point.

What about all the people that were replaced by horses?

If machines take over and decide to keep 30% of the current human population as pampered pets, I would count that as a stunning success. Eliezer probably would too.

So I am very much in the camp of "one day, automation will make human work superfluous" (http://theredbanker.blogspot.com/2013/01/robots-labour-and-sciences-fiction.html). BUT.

But, right now, before we say this is what is happening, exactly, I'd like to run a counter-factual where people salaries grow in line with productivity gains.

Because, see, I do think we have an AD issue above all else. And, if that issue was solved via salary increases, I would not want to bet that unemployment would stay nearly that high (or labour participation that low). See http://theredbanker.blogspot.com/2013/03/refreshing-macro-part-4-profit-matters.html for my own take on the profit/wages split...

Is it a coincidence that US gains that TC talks about in the 1840s happened about the time the USA adopted better patent laws? See the below, the US Patent Act of 1836. I think there's no coincidence. People respond to incentives and innovation is not on a vertical, inelastic supply curve, but slopes upwardly. You can train a hamburger flipper to become a decent, average, rocket scientist. You can teach a chess patzer to improve to expert level (grandmaster is another matter, but still my point is valid). Same with science. Nobody but an immigrant seeking US citizenship does science in the USA anymore: for good reason. It pays better to be a rent seeker in DC, NYC, etc. There's no money in science as I can attest: I got out, made my fortune rent seeking, albeit I work with engineers and scientists.

TC says: The Ricardian comparison is the technology-relevant one here, and I would challenge the notion that it went fine. Think of the machines of the industrial revolution as getting underway sometime in the 1770s or 1780s. The big wage gains for British workers don’t really come until the 1840s. -

http://en.wikipedia.org/wiki/History_of_United_States_patent_law#Patent_Act_of_1836 (Patent Act of 1836

In 1836, another federal Patent Act was passed to reform problems in the previous acts. The Patent Act of 1836 was significant in a number of ways. Firstly, the act created an official Patent Office, which was still a part of the Department of State, but no longer under the duties of the Secretary of State. It freed the Secretary of State from the enormous duty of granting patents, when he had many other significant tasks to tend to. Instead of the Secretary of State, the Commissioner of Patents chaired the Patent Office. This restructuring of the patent office allowed for more efficient process of patent applications. Secondly, the act prevented the applications of already patented inventions by requiring information on newly granted patents to be made publicly accessible at libraries throughout the country. Anyone could consult this information to check whether his or her invention was truly original before filing for a patent. This greatly improved the quality of patents granted. [20] Thirdly, the act resolved a long standing dissatisfaction on patent terms by allowing, for the first time, a possible extension of 7 years of protection in addition to the original 14-year term. With the permission of the Commissioner of Patent, upon appropriate reasons, patentees could appeal to have their protection extended. And lastly, it finally removed the US nationality and residency requirements, making it possible for foreigners to file for US patents. [21]

You can train a hamburger flipper to become a decent, average, rocket scientist.

You can? Give me a single example.

No rocket scientist ever worked at McDonalds as a teen?

I am not talking about AP-level high schoolers working part-time for car money, who then go on to get engineering degrees.

Ray Lopez's statement implies I can go to the nearest McDonald's, pick out a random burger flipper, and train him or her to be an aerospace engineer. I would like one example where this has been done.

Reality lies between those extremes. It's fair to think some significant percent of burger flippers remain those due to hard circumstances or lack of opportunities / motivation.

People don't end in dead end jobs only because of a lack of aptitude.

Spoken like someone who has been on only one side of the counter. Sorry. Other than the ones who are there a short time in transition.

No one doubts that any rocket scientist can flip burgers. No one with a shred of awareness thinks that it also works the other way around.

I'm no rocket scientist, but neither would any fast food restaurants hire me in high school.

All people do not have equivilent intelectual potential even when raised in an enviornment that encourages such things.

The brain is an organ like any other and just as some people are more gifted in strength and speed, others are more gifted in mind related tasks.

You can improve people with good pedadgogy and a good society but those average folks cannot be changed into a genius and more than they can play in the NBA. The potential isn't there.

As a society we are finding we need less and less of them and if we don't want a permenent population decline or some kind of rebelllion would do well to accomdate them.

So, if you go with “back then machines substituted for brawn, now they are substituting for brains”; and the odds are that somewhere in the future we have AI robots who can do all of the manual and mental labor to produce the necessities of life; then why shouldn’t we construct a society where we all retire and let them do the grubby work using a Kickstarter type mechanism to determine what they produce?

Isn’t the alternative our current winner take all type system where the first group of people who are lucky and skilled enough to stumble across the next Bill Gates or Mark Zuckerberg idea get to collect all the rent?


And Marx is not the sort of person talked about here - especially when his predictions are quite correct. 'Consider, for example, Marx’s prediction of how the inherent conflict between capital and labor would manifest itself. As he wrote in “Das Kapital,” companies’ pursuit of profits and productivity would naturally lead them to need fewer and fewer workers, creating an “industrial reserve army” of the poor and unemployed: “Accumulation of wealth at one pole is, therefore, at the same time accumulation of misery.”' - http://www.bloomberg.com/news/2011-08-29/give-marx-a-chance-to-save-the-world-economy-commentary-by-george-magnus.html

Automation is simply the latest step in this capitalist drama, one merely of degree, not of kind.

The Industrial Reserve Army is, unfortunately, a reserve army. It requires a high degree of class consciousness to keep the wages to suppressed, or else random interlopers can utilize the Reserve Army and make new companies and products. Or even worse, an actual army!

Thought: no one ever asks to me start a business or leverage my skills for myself, only for the benefit of a Big Business. On the other hand, I get non-stop requests for new credit cards, auto loans, home loans, etc. Maybe the problem at hand is that with all this excess capital, the only thing we can think of doing with it is spending it on more stuff?????
Using scarce resources in new ways is simply not on the radar of anyone I know, other than economists.

Why would you need to "construct" such a society at all? Given strong AI, it seems like it would happen on its own.

The question that doesn't get addressed is what does it have to do with the Great Recession? And why would automation replacement hurt our employment rate in 5 years rather than more slowly? And what about the fact that there are plenty of jobs that haven't been replaced with automation?

We currently have an indifference curve between machines and labor. So we fully expect that increases in technology shift to a new curve in which labor is more valuable (in the long run). But these curves all assume that there are diminishing returns to substituting machines for labor. What happens when computers are smarter (and cheaper) than we are, and when robots are more dexterous than we are? The indifference curve will no longer exist, except possibly in entertainment. What will happen to the jobs then?


Isn't the labor participation ratio higher than it was 40 - 60 years ago? Matt Yglesias once joked that the 2008 Great Recession popped the the labor participation ratio bubble. Considering both China and India have massively increased the non-agriculture labor force so the last 25 years we experienced a great labor glut and this glut is going to effect the growth of India (who is having an old fashion Volcker recession) and later China.

I suspect that a lot of middle class couples will return to a single income if they have more than one child.

Yes, past adjustments were painful, but the most plausible explanation is that being a peasant really, really sucks, not that people were hoodwinked from a rural idyll into urban squalor.

The peasant issue has largely been addressed in the West. Curiously, life remains a struggle - a bitch really, sometimes - for most of us, and we continue to suffer hardships, though our hardships are lost on peasants.

As for the future, I accuse you of a simple failure of imagination and a lack of faith in all those newborns today, another cohort entering the planet as Undefeated All-Time Winners of a Mind-blowingly Improbable and Ruthless Competition.

Actually, I was quite surprised to learn that a lot of 19C philosophers, economists and leading lights thought that the peasants were having it way too good and thus were lazy/unmotivated to go and 'enjoy' the squalor and low pay promised by the Industrial Revolution (a bit of analysis was showing that a pair of shoes was more afforable o the peasant than to the factory workers, even if factory work made shoes far more common) and were thus spending quite a bit of time decrying peasants' morals and thinking up ways to make their lives radically more miserable so that they would finally get on with the program and move to the cities...

I am trying to Google up the stuff but, so far, I can't find it. It was an awesome article full of actual quotes from the period, priceless.

Today, I think it's worth thinking about this when we say "oh Thai peasant girls, they actually prefer prostitution. I mean, it's awful, they know it, we know it but, since it's a free choice, being a peasant must be even worst".

I think it's worth wondering if there is not something in the background going on, making sure that being a peasant is utterly impossible...

It may be a failure for a different sort of market to clear. Call it "total cost of employment including hassle and liability, capped by cognitive limits of employers"

If I buy some machine and put it in my building, the general rule is that (a) it cannot sue me and (b) it does what I tell it and never gets bored and never wanders off on some kind of misadventure.

If I hire a human, I get all sorts of liability. And I have to use up mental bandwidth to manage the human.

In discussions in comments on places like Caplan's blog, I and others have pointed out that hiring staff might be monetarily easy, but it presents a whole new huge burden of interacting with people. And that burden is much more difficult than interacting with the dish washing machine.

So we might ask "To what degree is current unemployment caused by perceived bandwidth costs of employing those people, compared to the bandwidth cost of doing something else?"

I suggest there's a lot to that - who is having the hardest time finding a job? Young folks lacking experience, and older folks who have been out of work for some time. In other words, the sorts of people employers feel might require a lot of bandwidth, people who cannot just be set to foot and turn your back on them.

Automation only affects this by offering a substitute resource that CAN have low bandwidth requirements.

and yet there are managers who specialize in micro-management...

When does the robot you hire go shopping to buy whatever you and all the other robot employers produce?

Or do you depend on the government paying the unemployed enough to keep you in business?

Shifting a farm worker to another industry consistently meant that that workers and over-all productivity rose.

But now, shifting a manufacturing workers into another industry does not improve productivity the way shifting farm workers did. It is just as likely to reduce productivity growth.

Baumol's Disease

It's worth pointing out that the spread of steam-powered machinery and industrialization from textiles/mining/steel to all manner of British industries didn't really get going until the 1830s and 1840s. Before that, it was mostly piece-meal, with some areas picking up the technology faster than others, while the overall economy didn't change that drastically (hence the minimal changes in overall wages).

Machines are substituting for brains, but they are also going where brains could not efficiently go.

Back in the mid-80s, one of my clients was the leader in a new field called software computational chemistry. A chemist could compute likely reactions, or so could a computer, and both could screen for toxicity for a new drug. But, the supercomputer and the software could cover more, and eventually go where, because of complexity and dynamics, the chemist couldn't go.

It's not either or. It's also where no human could go, so it is output expanding without loss of human input because human input was never a substitute.

> "You also might try to run with a “back then machines substituted for brawn, now they are substituting for brains” argument. Maybe so, but you don’t even need to make that work to have a substantial (non-Luddite) worry."

I can't see how the argument "We are now automating more cognitively difficult work than before" is a good answer to "What changed?" in terms of any particular background theory of macroeconomics I can think of. It should only make a difference whether there is still some remaining form of valuable human labor the displaced workers can do, and how hard it is to obtain that work, and how much money is available to pay them, and employer downsides, and BATNA for the unemployed, and whether retraining is available, etc. It should not make a difference how 'intelligent' the lost/displaced jobs originally were, I don't see how that factor has any causal power over the search for new jobs.

Unless we think that 'automation is displacing some jobs which required more intelligence' translates into 'there is no remaining labor of any kind, which would be valuable to anyone if done for free, which those people can do'. But one would expect the less cognitive displaced jobs, to be the ones whose workers had the hardest time retraining!

Also I doubt that skilled weavers would have told you their work was cognitively easy, or accountants replaced by spreadsheets. It's only the rise of automation that leads people to think, post facto, of those jobs as cognitively easy.

I hear the "Now automation displacing smarter jobs" line a lot. But it doesn't make sense to me in logical, causal terms. As far as I can tell, this is persuasive only because it makes AI sound like a more formidable, powerful force and thereby sound on some pure intuitive level like a plausible culprit for more powerful and lasting unemployment curses. I can't see how the actual causality would play out like that, on any particular background theory of economics.


Though, you might want to consider that the economic theories we have aren't good enough to tell that kind of micro story...

A more sophisticated version of that argument is 'hollowing out' i.e. automation lead to a rise in the amount of people willing and able to do menial jobs for low pay. Again, Agatha Christie - People used to have house help and people who can afford it today (or decide to devote some of their income to hiring some level of house help) usually find it quite a big positive.

But I am unconvinced as well. I think the problem is simple. Automation is good for productivity growth but wage growth does not track productivity gains by magic. It's a willful exercise societies engage in.

Think Roman society and the rise of slavery. What did it do for the poor/non-land owner but nonetheless free Romans?

It's not really a theory. It's what has actually been happening for several decades now. Or am I the only one that sees the divergence in wages/productivity, the decline in the labor force participation rate, or the fact that most employment growth is on the low end.

What evidence do we have to support the idea that new high end jobs are replacing the ones we lost?

The largest difference is agriculture was small scale and competitive meaning prices fell with productivity. Industry enjoys such massive economies of scale, it is an oligarchy with administered pricing preventing prices from falling, leading only to branding.

The largest difference is agriculture was small scale and highly competitive, leading to falling prices with productivity. Industry enjoys such massive economies of scale, it is an oligarchy with administered pricing with the only real competition being branding.

I wonder if you looked at world wide employment you would see a different picture.

How many more jobs are there now world wide then there were 5 years ago? Is the lack of a rebound in jobs just happening in developed countries? If so, then maybe the underlying cause is cheap labor around the globe rather than robots too smart and too skilled for humans to compete against.

Yudkowski is a blithering idiot if he thinks the failure of biotech is FDA regulation.
If anything, FDA regulation has *helped* the industry; without the FDA, people would have been killed - or more people would have been killed - and the industry would have been sued out of exisitence
anyone who says that the FDA is big hindrance dont' know squat; the *real* problem is that despite all of our knowledge, we still know very little about how the human body works (not even dealing with variations due to genetics, diet, etc)

Has anyone applied something of the Public Choice analysis of technological impact on distribution?

While I agree that labor markets influence corporate pay scales it's not a tight relationship and within a world where multiple equilibria rather than the "econ 101" single equilibrium is the setting one can image a number of equally efficient alternatives. The question then is what processes are selecting the alternative that will obtain.

Has anyone taken a look at the impact of technology from a more Public Choice type of analysis?

I would not deny that labor markets have influence on corporate pay policies but the links are not tight -- meaning there's going to a wide range of policies across similar corporation for any given external labor market. Seems that increased technology that substitutes for labor (and even some management) will increase concentration of decision power leading to a larger range of corporate pay policies. In other words, the larger external labor markets will increasingly have less direct influence over the corporate pay policies that determine the internal distribution of income among executives, management, rank & file and stockholders.

Comments for this post are closed