> for AI or productivity growth or increased trade or immigration or graduating students to increase unemployment, instead of resulting in more hot dogs and buns for everyone, you must be doing something terribly wrong that you weren't doing wrong in 1950.
Not necessarily. There's going to eventually come a point where any job that can be performed by a low-skilled worker can be performed by automation.
The example used to combat this notion is that servants used to be more plentiful than they are now, and low-skilled workers would be ripe for these types of careers. Well, what did servants do? The same thing my robot vacuum cleaner and dishwasher and microwave and Amazon Echo do now.
Right. He doesn't seem to address how fully general AI can entirely obviate (at least some class) of human labor. It's true that comparative advantage means you can still profitably produce something when others can do it better. But it doesn't guarantee that the (positive) market price will be enough to pay for the worker's sustenance.
This isn't idle speculation: it's exactly what happened to horses once technology obsoleted them, which caused them to die off except for some very narrow cases (rough terrain) and vanity competitions. I know, "humans aren't horses", but they're identical in terms of "what happens when technology can replace all their economic use cases much more cheaply".
So yes: human workers would be able to sell computation services at the market price of $0.50/PFLOPS-hour ... they just won't be able to feed themselves at that wage. (This basic dynamic is arguably what's happening with disability overclassification.)
With that said, Yudkowsky does make a strong case that AI unemployment is not the problem now and that we aren't in that world yet although IMHO he's bought too uncritically into the mentality of "historically NGDP solved everything, just like inflation always cured unemployment, so NGDP must be a cureall".
"This isn't idle speculation: it's exactly what happened to horses once technology obsoleted them which caused them to die off except for some very narrow cases and vanity competitions."
The article does explicitly address this. Do you think the response is unsatisfying, and if so, could you explain why?
"Q. How about timescales longer than ten years? There was one class of laborers permanently unemployed by the automobile revolution, namely horses. There are a lot fewer horses nowadays because there is literally nothing left for horses to do that machines can't do better; horses' marginal labor productivity dropped below their cost of living. Could that happen to humans too, if AI advanced far enough that it could do all the labor?
A. If we imagine that in future decades machine intelligence is slowly going past the equivalent of IQ 70, 80, 90, eating up more and more jobs along the way... then I defer to Robin Hanson's analysis in Economic Growth Given Machine Intelligence, in which, as the abstract says, "Machines complement human labor when [humans] become more productive at the jobs they perform, but machines also substitute for human labor by taking over human jobs. At first, complementary effects dominate, and human wages rise with computer productivity. But eventually substitution can dominate, making wages fall as fast as computer prices now do."
Q. Could we already be in this substitution regime -
A. No, no, a dozen times no, for the dozen reasons already mentioned. That sentence in Hanson's paper has nothing to do with what is going on right now. The future cannot be a cause of the past. (etc.)"
This is the failed assumption of the author; that computers need human intelligence to displace jobs but they don't. They've been displacing jobs since their invention and they're getting better at it every year. They don't need to have an IQ of 70 or even an IQ at all.
They created jobs too, a lot of which people assume are done by robots because they're done in other countries.
Apple may only employ 100,000 people in the US but 1.3 million not-robots work for the company that actually makes those phones.
The argument that it automation is "naturally" creating fewer jobs that it eliminates is refuted simply by looking overseas where all those new jobs actually are.
Or are they just temporary?
Foxconn threatened in 2011 that within three years their 1.2 million strong workforce would be largely replaced by robots:
I don't think the argument is that automation creates no jobs, the argument is that (at least in the current instance of automation) it creates dramatically fewer jobs than it eliminates, and the balance is not being made up for over time or in other industries, resulting in a substantial net decrease in employment or wages.
I'm not even talking about robots; I'm just talking about computers. The majority of people developing software aren't creating new jobs, they're deliberately and explicitly reducing the need for human labor.
"Robin Hanson totally thinks that won't happen because human and computer labor are complementary" is not a sufficient response. It's equally true of human and horse labor. Still had the horse die-off. In fact Hanson isn't even disagreeing -- he says that this complement is temporary and will eventually become displacement instead, just like I was saying.
That agrees with my point that Yudkowsky has made a case for AI not causing unemployment now but still being a future risk, even with the economic analysis holding true.
I am entirely unsatisfied with his address of that problem. Mainly because he doesn't address it. He just kinda hand waves it away. And that's the only interesting thing that there is to address.
We all know that automation is coming, and that it's going to replace a lot of workers. Augmentation is also going to come, and it's going to displace many more workers, by allowing one worker to do the work of three. So what do we do with those that are displaced? So far no one has any kind of answer. Most don't even want to think about it, assuming that things will tie up nicely by themselves like on a sitcom episode.
>The example used to combat this notion is that servants used to be more plentiful than they are now, and low-skilled workers would be ripe for these types of careers. Well, what did servants do? The same thing my robot vacuum cleaner and dishwasher and microwave and Amazon Echo do now.
Where I used to live (Singapore), having a live-in servant was very common. It is hardly a primitive society either - a lot of them are probably getting fancy new Amazon Echos for Christmas.
Those servants, of course, aren't local. The labor pool is drawn from relatively more impoverished countries (usually Philippines). The women would accept very, very low wages and usually deal with a large amount of abuse.
The job wasn't automated away in England in the 19th century (no dishwashers or roombas back then), rather it was industrialization - automation, in fact - which provided better, higher paid opportunities that led to the decline of domestic work.
The key to its decline in 19th century England and the key to its continued existence today is simply about the relative exploitability of the workforce.
> There's going to eventually come a point where any job that can be performed by a low-skilled worker can be performed by automation.
A large proportion of low skilled service jobs in which being people, in the service area is 80% of the job and the actual tasks performed have in many cases have been possible to automate for the past century, if anyone thought it worthwhile. Even fast food chains, which aren't exactly renowned for joyous and welcoming customer service and are renowned for process improvement, still use people to dispense food, a century after the invention of the automat.
> being people, in the service area is 80% of the job
> Even fast food chains
Those are also jobs for which, for many customers, replacing people with automation would provide strictly better service. Ignoring the actual preparation and serving of food for the moment, an automated interface for ordering (as in "push button for what you want, pay, order shows up in the kitchen with a number") would make fewer mistakes, move customers through quicker, and cost less.
That doesn't mean automation could easily replace things that require (for instance) precise physical manipulation; it doesn't seem nearly as likely that automation could go around refilling beverages in a restaurant any time soon, and if it could, it'd cost far more than a human would. But taking orders, sending them to the kitchen, and processing payment? Automation would provide strictly better service there.
McDonalds, KFC etc who invest millions in researching process improvements, beg to differ on the strictly better service.
Part of that is down to the fact the person that takes your order also helps clean when not busy, smiles at your kid, ejects unruly customers and upsells to supersize more persuasively than a flashing red box (which was a large part of my original point)
But part of that is also down to the fact that the average person is really bad at using a human-machine interface (and that ain't gonna be improved in a busy commercial environment by voice-recognition technology any time soon)
I mean, I see a lot more automated kiosks in supermarkets than I used to. Usually I use them in preference to human kiosks. Many other people have precisely the opposite preference, and they have money to spend too. Overall, the kiosks increase the throughput of the store, but that's because there are more of them due to their smaller footprint than the checkout counters they replace. In the hands of the average person trying to use the system they're certainly not more efficient than a competent checkout operative, and require a human for every few machines to deal with their idiosyncrasies and frequent failings, and manage the queue.
And the automats, which are a far better human/machine interface than some talking touch screen connected to the kitchen because you literally see the freshly-made fast food, put money in the slot and open to receive it? They've been around for a century now, long before Ray Kroc and Colonel Sanders got into the industry, and they've repeatedly failed.
McDonalds specifically is starting to replace its cashiers with kiosks, though right now it looks like the former cashiers are switching to bussing tables instead.
> an automated interface for ordering (as in "push button for what you want, pay, order shows up in the kitchen with a number") would make fewer mistakes, move customers through quicker, and cost less.
Having sat behind people trying to figure out how to exit automated parking garages on multiple occasions, I'm not convinced this would move people through any quicker. Also, what happens when the button wears out or malfunctions? Are the line cooks also skilled in IT diagnostics or do you just begin hemorrhaging cash till a tech comes out? I've seen skeleton crews continue to make money even through complete power outages.
>Also, what happens when the button wears out or malfunctions?
You have multiple buttons, and have marginally longer lines until someone can come to fix it. Or if that would cost you too much in sales, you have a regular maintenance schedule that reduces the chance of failure to acceptable levels. (And still involves far less labour than paying for a cashier all day.)
Would? It already does, at least for a few well know fast food chains in my country. Just a couple of days ago I used one of those kiosks for ordering and paying.
Suppose you have two workers - one an average joe and the other who learns faster, works faster, produces better work, takes less vacation, never gets sick, never complains, and is willing to accept less pay. Which would you hire?
The end of jobs (as in Humans Need Not Apply) occurs when the latter employee happens to be a robot.
If the examples in the article aren't convincing, there are plenty of other ones. Eg., construction involves lots of manual labor, and lots of machines that are designed only for human operators. If automation reduces the demand for low-skill labor in other fields, the result should be a construction boom, as that sector reduces costs and snaps up more workers. But that's not what we see. In fact, in many ways it's the opposite - in NYC, subways now cost ~30x as much per track-mile (inflation-adjusted) as in the 1920s.
Unemployment in other sectors "should" result in a subway construction boom/plummeting costs only if low-skill labor were the largest cost component of subway construction in the recent past. Was it? I tried searching for a few minutes on Google Scholar to find the share of subway project costs going to labor but didn't turn up anything; maybe you can find something.
I did find articles that say labor costs have increased over time, but none that were quantitative about it. Without quantification it's not helpful.
Now extend that thinking. What if low-skill labour isn't the largest cost component in anything much worthwhile that we're not already doing? How would that affect the reasoning in this FAQ?
The extra cost of modern construction projects can easily be attributed to the increased compliance costs. Any building work that impacts the flow of traffic now requires full time traffic control (manual); Safety inspectors are now full time; Continuous testing and inspection of building materials and practices requires trained inspectors.
We have vastly reduced the injury and death rates associated with construction at the expense of increased number of people involved in assuring that safety and building practices are complied with.
Not necessarily. There's going to eventually come a point where any job that can be performed by a low-skilled worker can be performed by automation.
The example used to combat this notion is that servants used to be more plentiful than they are now, and low-skilled workers would be ripe for these types of careers. Well, what did servants do? The same thing my robot vacuum cleaner and dishwasher and microwave and Amazon Echo do now.