Beware the Mediocre Robots!

It’s often thought that what we have to fear from automation and AI is super-robots. Acemoglu and Restrepo make the useful point that what we actually should fear is mediocre robots, robots only slightly better than humans. Think about robots replacing labor in various tasks. A super-robot replaces labor but has an immense productivity advantage which generates wealth and increases the demand for labor elsewhere. A mediocre-robot replaces the same labor but doesn’t have a huge productivity advantage. As a result, the mediocre robot is the true jobs killer because it replaces labor without greatly increasing wealth. Think about automated phone systems or chat bots.

In an empirical breakdown, Acemoglu and Restrepo suggest that what has happened in the 1990s and especially since 2000 is mediocre-robots. As a result, there has been a net decline in labor demand with no big wealth increase. Thus, Acemoglu is more negative than many economists on automation, at least as it has occurred recently.

More generally, Acemoglu and Restrepo create a new type of production function and use that to reformulate how we think about production and how we measure what is happening in the economy with automation and AI. This is one of the most important new pieces on automation and the economy.

Comments

I suspect this is a function of the sectors that have grown the most since 2000 (in the U.S., where the linked study was done). For example, construction, medicine, etc. The sectors where automation was supposed to greatly increase productivity, such as manufacturing, have lagged in growth.

Tabarrok: SOME people are concerned about the possibility that unaligned or unfriendly AI could exterminate the human race or worse. However, I'm more worried about robots NOT exterminating us! There could be some extremely minor disemployment effects!

As a result, the mediocre robot is the true jobs killer because it replaces labor without greatly increasing wealth. Think about automated phone systems or chat bots.

Boy am I confused by this argument. The productivity boost of automated phone answering systems is not that they are better than humans in talking to customers but that they are extremely low cost to operate and don't require any labor at all, thereby freeing up all the people who used to answer phones for other tasks. Isn't that freeing-up precisely where all the wealth benefits come from?

Look at these apple-picking machines for example. Are they absolutely better than humans in identifying ripe fruit and picking it without bruising? Perhaps not, but so what? (I'm glad I looked up the video by the way -- I've started to see orchards in the fruit belt near Lake Michigan with these rows of small, vertical apple trees. Now I know what's going on).

Good comment.

This assumes that there are other equally productive jobs which those specific workers are capable of performing waiting for lack of workers. Since we are talking about very low skill jobs it is not at all clear that such jobs exist, especially since those jobs would also be ripe for automation.

Back to the same old worry -- what will all the former maids/farm-laborers/factory-workers do once we have laundry-machines/tractors/robots? Every time this comes up some people assume that this time it's different and all those folks will end up unemployed with nothing to do. It never works out that way, but that history doesn't stop some people from arguing that this time is different.

I wouldn't be so confident that this time isn't different, with the possibility of several years of higher unemployment / lower labor participation. Labor participation is at 63%, down from a peak around 67% in 2000 and where it was in 1980. Not that a lower labor participation is a bad thing...

I actually think it is different this time -- automation/mechanization is displacing jobs at a much slower, not faster, rate than during the 20th century. Look at automation in the household. Modern appliances (washing machines, dryers, dishwashers, vacuum cleaners) made an enormous difference. But since, say, 1970, what do we have? Very little.
We got the Roomba going on 20 years ago, and it's still only marginally useful. We don't have robots that do the dusting or empty the dish-washer, or fold the laundry and put it away. Or rake leaves, pull weeds, shovel snow, or wash windows -- and we're probably not going to get any of that in our lifetimes. The low-hanging fruit of automation has been picked.

"and we're probably not going to get any of that in our lifetimes. The low-hanging fruit of automation has been picked."

You're still drinking the Tyler tea brewed in 2009?? It has always been low hanging fruit. Krugman recently had a column that was full of errors with respect to the Democratic debate. Contrary to what Krugman wrote, the reporter's question on what to do about potentially 25% of jobs being replaced by 2030 is what a well known report estimates with 43% of jobs gone by 2035.

Krugman was correct in that automation isn't fast yet but Yang never said it was as Krugman claimed he did. Yang's argument is that much more automation is coming and that people are starting to see it in stores.

For example, a Japanese company has made a scanner that will calculate the cost of a basket of goods at once without needing to scan each item and that is expected to be used in 2022. Robots will replace many warehouse workers by 2025 and there are many white collar jobs that will be automated as well - like radiology and more legal services. Then there are the driverless trucks....

"Modern appliances (washing machines, dryers, dishwashers, vacuum cleaners) made an enormous difference. But since, say, 1970, what do we have? Very little."

The first consumer microwave ovens were for sale in 1972 for $350 or $2,200 in today's dollars. By 1985, 25% of households had them and they were great time savers.

Wikipedia thinks otherwise:

Japan's Sharp Corporation began manufacturing microwave ovens in 1961. Between 1964 and 1966, Sharp introduced the first microwave oven with a turntable, a feature that promotes convenient even heating of food.[16] In 1965, Raytheon acquired Amana. In 1967, they introduced the first popular home model, the countertop Radarange, at a price of US$495 ($4,000 in 2018 dollars).

But the difference between 1964 and 1972 is irrelevant. If you have to go back either 55 or 47 years to find the latest significant advance in home automation, you know things haven't been moving fast in the last half-century.

Wikipedia also shows that in 1971 only 1% of households had one and that is a major advance that you didn't list but not really an automation, just a different type of oven. (The 1964 model wasn't for homes and the 1967 model sold 40,000 in 1970.)

There have been huge quality increases in televisions, home computers, refrigerators, washers and driers while prices and usage costs have declined dramatically. The only further innovation could be a robotic maid and butler, so it doesn't make sense to use the home as an example of stalled automation since its all been done.

"The only further innovation could be a robotic maid and butler, so it doesn't make sense to use the home as an example of stalled automation since its all been done"

ONLY!!!??? A combination robot house-cleaner/laundress/butler/gardener/cook/dog-walker/etc would be HUGE. Possibly as big a deal as all the home automation of the 20th century combined (I mean, who would care if tasks done by appliances now went back to being done by 'hand' if the hand in question was that of an indefatigable robot)?

Home automation hasn't stalled out during the last half-century because there's nothing left to do -- there's a LOT left to do. But what's left is much, much, MUCH harder. Orders and orders of magnitude harder. So much harder that we can't even get a reasonable handle on how much harder. So we're stuck and may remain stuck indefinitely.

"Home automation hasn't stalled out during the last half-century because there's nothing left to do -- there's a LOT left to do. But what's left is much, much, MUCH harder. "

Well, what is left to do - specifically? A robotic washer, drier and shirt folder and maybe a cook even though many eat fast food, order take -out delivery and eat at better restaurants. Roomba will still improve and the rest will be available by 2029. Why stuck indefinitely? A robot can already fold shirts.

https://www.youtube.com/watch?v=wmC10L_iT-E

I take it you don't do much housework?

Dusting
Washing floors and counters
Sorting and emptying trash
Cleaning Windows
Clearing leaves/shoveling/mowing
Cleaning walls and baseboards.
Sorting, folding, and putting away laundry.
Buying and making food
Cleaning up messy rooms, putting things away.
Vacuuming/wiping down furniture

And the list goes on.

Consider automated mowing. There are a couple of automated mowers out there, but they don't work well. Too many exceptions to deal with. How are we supposed to have self-driving cars if we can't make decent self-driving mowers?

No, I'm not into cleaning... Obviously self driving lawn mowers will appear at about the same time as self-driving cars. I addressed buying and making food. Anyone in a city can have food delivered already and takes minutes to make. You aren't asking for a robot maid to spoon feed you in bed, right?

I already addressed laundry. Did you watch the youtube clip? You can wash the windows and base boards once a year or hire someone on Craig's list.

Self-driving lawn mowers exist. There's one in the Montgomery Zoo in Alabama. I nearly tripped over the thing, and my sons and I spent a good 15 minutes watching it work. It's not really like a car, it's more like a Roomba.

Yeah, I've seen laundry folding machines before and if you can't see the vast, VAST gulf between that thing and a general purpose robot that could fold all kinds of laundry AFTER running it through the washer and dryer and THEN putting it away in dressers and closets and also make your beds, clean your house, cook your food, weed your garden, run and empty the dishwasher and put everything away, and on and on, then I don't know what to tell you. Entrepreneurs are picking one part of one task -- folding laundry -- and trying to build a special purpose machine that can do that and NOTHING else (and can't even fold large or small items) and they're struggling mightily with what would be a trivial chore for a human being. And I'm NOT dissing the inventors of the shirt folding machine. Even what they've set out for themselves is genuinely a really hard problem. The flexibility, learning capacity, dexterity and hand-eye coordination of humans are AMAZING (not to mention the ability to run for hours fueled by nothing more than a sandwich).

"and trying to build a special purpose machine that can do that and NOTHING else (and can't even fold large or small items)"

It's only 2019. It really seems that you want automation to do everything: select a book and read it to you. Choose what you want to eat, cook it for you and spoon feed it to you. All that is sadly 15 years away.

I'm not expecting automation to do all those things in 2019 (or in 2069 either). I think that in tasks that require dexterity, flexibility, hand-eye coordination, etc, that humans are very, very far ahead of the machines. Which is why I don't expect machines to perform these tasks in my lifetime. And why I think the low-hanging fruit of automation has already been picked.

Has the low hanging fruit been picked for self-driving cars? They exist but fewer than 0.001% of Americans use one. Yet something tells me that when they become common in the late 2020s, you'll say that is old technology and all the low hanging fruit for self driving cars was picked by 2019.

"you'll say that is old technology and all the low hanging fruit for self driving cars was picked by 2019."

Nope. I'm saying that self-driving cars are not going to make it. I believe they're nowhere near as close to ready for general use as they might seem to you -- you're not going to see them driving around, unrestricted, on all public roads without safety drivers anytime soon at all.

Prime age (25-54) LFPR is at its highest point in a decade (82.6%), and unemployment is at record lows. 22 million jobs added in last decade (all private sector, so less likely to be 'make work') in America.

"What are we gonna do with all these people?" seems to have taken a back seat again.

I agree with Todd K here. I think Carl Frey's book has convincingly, at least to me, shown that the medium term adjustment to past technological changes has not been as smooth as economists claim a century later. We also, arguably, have a historic example of a class of "labor" being entirely eliminated from the economy. Horses. Not my point, it was made either in the Rise of Robots or the Second Machine Age, don't remember which.

Fortunately humans are vastly more adaptable and flexible than horses (or oxen or messenger pigeons or turnspit dogs or any number of animals previously used for specific tasks)

"This assumes that there are other equally productive jobs which those specific workers are capable of performing waiting for lack of workers."

No, it doesn't. It assumes that there's *something* else these people could be doing. Even if that something else is minimally productive - even less productive than what they were doing - its a net benefit.

Because the robot allows us to have whatever they're doing now AND whatever they were doing before - rather than either/or.

"...even less productive than what they were doing - its a net benefit."

Yes, but that's the rub. If those people are paid based upon their productivity then they'll be earning less than what they currently make. Granted, goods will also be cheaper, but if we are in for another 20 years of wage stagnation, then that's going to cause social issues.

It's not good enough for productivity to provide a net benefit, it has to be a widely shared benefit. I'm not arguing for a massive wealth distribution scheme, just a structure so that the lower 3 quintiles aren't stagnant.

This is a very...odd....view of humans. It rests on the assumption that workers can only be fit into a given open slot in some existing institution. It eliminates, in other words, volition (and entrepreneurship).

People will be faced with two options, just as they were every other time tech disrupted the status quo: Learn new, marketable skills, or go broke. After an initial (admittedly sometimes painful) adjustment phase, people tend to learn new skills.

At the very least, the burden of proof is on you to demonstrate that workers cannot learn new skills, since humans have a proven track record for adaptability.

It's even worse: since robots cost so much less to purchase and operate than human workers, the optimally productive robot may well be less productive than a human worker - compare automated phone answering systems again. Not sure what this means in terms of wealth benefits. Ask an economist.

I believe the answer to your question is that the robotic system ends up being very similar cost to the humans it replaces, thus no productivity improvement.

Automated answering machines are an extremely poor robot easily replaced by call centers. This is a good example. The task is complicated, if done poorly counterproductive. Answering services are for places that aren't public facing and are a filtering mechanism for busy people.

Say no to mediocre robots!.......those are strong words against self-driving cars =)

Indeed. A robot car that must have a human watcher in 1:1 ratio, is a mediocre, dare I say shitty, robot.

On the other hand, the automated electric vehicles moving containers around the port of Rotterdam are amazing. The video playback speed is accelerated for entertainment purposes, otherwise you'd fall asleep watching robots at <10 km/h speeds https://youtu.be/zm_rlLyelQo

Oh sure, I don't think anyone disputes that automation often increases productivity.

But I think this is an interesting idea, there might be areas where the payoff is smaller, or even non-economic. What if you as a manager had a chance to pay 5% more for a system that gave you 50% less headaches? You might do it.

(Nice video, that focus technique that makes everything look like toys. A train set.)

Autonomous cars will be limited to off-road use (such as moving people in airports) unless and until a separate right of way is built for them. Does anyone want to ride in an autonomous car while sharing the road with teenagers (not to mention drunks) driving non-autonomous cars? Of course, autonomous cars on their own right of way is . . . . . transit. Why have Google and other so-called tech companies invested so much in flying cars and spaceships to Mars? Is it to prove they are tech companies? Now this quantum leap by Google (a leap doubted by their friends): https://www.nytimes.com/2019/10/23/technology/quantum-computing-google.html

Haha, the media.

Of course, the media decided that the "Google self-driving car project" must be a Google subsidiary focused on "a new way forward mobility".

Once again, the media decided to sue Uber for 1+ billion because some self-driving technology was stolen.

Finally, the media forces Elon Musk to tweet self-driving hype weekly. The prudent engineer would never talk bullsh*t if the media would not put a gun to his head.

If the robots are mediocre, what is your word for the humans they are replacing?

Tough guys, if they could pick apples all day.

If I use Excel to complete some set of calculations, does that imply that I am bad at math?

Trump voters.

But even mediocre robots free up workers to do more productive things.

Even if the workers, in their new employments, do slightly less productive things, there is still an increase in production over all.

Now we are just grasping at straws. There are two main causes of net wealth stagnation:

1. Levelized population growth
2. Centralization of labor markets

Anything else is just nibbling at the periphery.

Why would a capitalist invest in a mediocre robot? I suppose partly because Central Banks have arranged that he can borrow money free of charge.

Robot won't sue, won't unionize, won't file complaints with OSHA, won't object to being repurposed next week and then again next month, can but shut off for three months during a factory refit and turned back on with no ill will, will work Saturdays and Christmas, etc.

Even if the mediocre robot is no more capable and is expensive to buy and even to maintain, there are still major advantages.

Do you have a typist at work, or do you use Microsoft Word?

"mediocre robots, robots only slightly better than humans." Does that imply that there are sub-median robots that are worse than humans? Why has anyone developed them? Ah: to serve as politicians. Of course!

Think about automated phone systems or chat bots.

These are not my favorite things. I prefer to direct my discretionary thinking elsewhere.

I can't help but think of all those people who buy automatic lift gates for their SUV. Sure the back hatch opens itself, but you see them all just stand there waiting for it.

Yes, you see them standing there, not getting their hands or gloves filthy from the usually very dirty back of the vehicle in winter, and not having to put their bags down on the wet or snowy ground to free a hand.

I have a powered liftgate. It's awesome in bad weather or when your hands are full. On the other hand, it's a mediocre robot - an automated lift gate requires safety interlocks which makes it impossible to fully fill the cargo area without preventing the lift gate from closing. The 'kick' actuation is easy to set off by accident, it adds weight, cost and complexity, etc.

I wouldn't worry about automated cars taking over until they can make an automated liftgate that won't smack me in the face because it started closing when I didn't expect it.

'A super-robot replaces labor but has an immense productivity advantage which generates wealth and increases the demand for labor elsewhere. A mediocre-robot replaces the same labor but doesn’t have a huge productivity advantage'

Is this necessarily true ?

Suppose (taking all costs into account) you can produce 10 widgets a year with no robots.

If one can make 7 mediocre robot in two human year and those robots just beat human productivity then then (ignoring wait time) your productivity in 3 years goes from 30 to (Just over) 70 widgets (more than double)

If you can build one super smart robot in 2 man years and its productivity is 5 times a human's then (ignoring wait time) your productivity in 3 years goes from 30 to 50 widgets (less than double

It seems like the same economics applies to robots as to any other capital good - production costs need to be taken into account in measuring productivity.

Another mediocre robot spotted:

"I thought the appliance would save me time, but mostly it gives me peace of mind."

"The Instant Pot doesn't save me as much time as I'd expected because it takes a surprisingly long time to reach pressure and then to de-pressurize (although you can speed this up by releasing it manually). Nor is the food any tastier than food I make on the stovetop. Where the Instant Pot does stand out, however, is in the fact that you can leave it unattended. This has been hugely helpful."

https://www.treehugger.com/green-home/some-thoughts-owning-instant-pot.html

Is Alex warning us of the advent of mediocre "machine learning" that might stand some chance of eviscerating public education? --that might compete soon with post-secondary training, curricula, and teaching?

--or do we suffer already from far too much "mediocre teaching" that already yields far too much "mediocre learning"?

Who addresses human mediocrity well: humans or robots? If humans cannot overcome human mediocrity themselves, their/our robots likely cannot overcome it, either, though maybe they can accommodate it or adapt to it with less complaining than might come from AI engineers.

I’m usually positive about Daron’s oeuvre, but I think he’s off the mark on AI. “Mediocre” robots that replace multiple people and don’t require a break aren’t mediocre at all. They create wealth equivalent to the people-hours replaced. In the short run, the labor market cannot respond quickly. Some lower class Boomers are unfortunately being forced onto welfare.

Over 20 years or so, the youngsters that otherwise would have performed the same rote, mindless jobs have no choice but to get a more technical education and create value elsewhere (because robots are doing the rote work). If they can’t perform technical labor, there is a growing need for emotional labor. Baumol’s cost blessing will ensure they get compensated with some of the value created by the aforementioned robots. A better social safety net will take care of the rest and everyone will be better off. Also, and this cannot he overstated, the “mediocre” robots will become exponentially better and inevitably deliver the fleet-raising value that Daron is looking for.

Average isn’t over?

"Think about automated phone systems or chat bots."

We're poorer because of automated phone systems?

No, there is still wealth created by mediocre robots because the human is freed up to do something else.

"is mediocre-robots. ... Thus, Acemoglu is more negative than many economists on automation, at least as it has occurred recently."

This whole argument misses the underlying point of automation. A mediocre-robot will eventually be replaced with a slightly less mediocre-robot and so on and so forth. For the most part you can only replace a human workforce with a better workforce to a certain extent. There's no reason the mediocre-robots can't get 2-3% more effective every year for a very long time. Furthermore, it's far easier to take the improvements you've learned and embed them into future robots than it is with humans.

"This whole argument misses the underlying point of automation. "

Yes. It is so obvious that I didn't comment on that. It seems that Acemoglu wants to be part of the A.I. discussion so picked a weird angle that will be forgotten in a month.

I find that mediocre bots make Marginal Revolution better. Even if they have to stay anonymous or have to change from Prior aliases.

+1 Funny

It's certainly ridiculous.

Trump's lawyers argue he can't be charged while in office — even if he shoots someone

And of course dangerous if some deplorable "base" goes for that.

From one poorly written bot to a mediocre bot, I value your contributions here and hope your programmers learn and improve from the user’s interactions with you.

01000011 01101000 01100101 01100101 01110010 01110011

I try to give you the human curated best.

And since you ask (I presume "OrangeManBad.exe" is an attempt to preempt criticism), Donald himself isn't holding it together too well today:

The Never Trumper Republicans, though on respirators with not many left, are in certain ways worse and more dangerous for our Country than the Do Nothing Democrats. Watch out for them, they are human scum!

An official communication from Our President.

Indeed, funny.

By the way, now that Trump is extending Executive Privilege to Florida Men, isn't this more Carl Hiaasen novel than reality show?

Having read the inappropriately titled "Superintelligence,", I believe the cure for existential crises is therapy. The assumptions made in the book are inches from satirical. A Superintelligent Mind is likely to discovers new recipes for gazpacho, using spices harvested off of asteroids or on planets we will colonize. Hopefully, I'll be long gone before we start putting policies in place to try and achieve the Simpleton.

I support automation; but find blog posts like this very groan worthy...

This reminds me of the comedian that said: "I do not fear artificial intelligence becoming smarter than humans, I fear the day when artificial intelligence equals the dumbest person on earth." I wish I could remember the comedian's name.

That's actually kind of profound. At the point in time AI is as smart as a chimpanzee, automation will quickly replace all menial tasks.

Most menial tasks are about dexterity and coordination, not how smart the 'bot is. Think washing dishes, cooking, cleaning a house, etc.

Well robots are dexterous enough to solve a Rubik’s cube one-handed now. You can extrapolate from there https://youtu.be/kVmp0uGtShk

I'm often surprised by the "the robits are tak'n our jorbs!" crowd's lack of "and then what" thinking. If labor is actually automated away, then other robots will compete for those jobs at a slightly lower profit for their owners, and the cost of goods will approach commodity input prices plus the electricity to run the robot during assembly. Displaced workers will need only to replace a fraction of their income to maintain their material standard of living.

Comments for this post are closed