"If demand for a particular good or service is highly elastic and AI does not completely automate the production of that good or service, then technical change would create jobs rather destroy them."
This is the fundamental problem with this article IMO. It assumes that human labour will be required to fill the gaps in an automated production cycle. But the ultimate premise of automation is that there are no gaps.
If you look at something like car production, first we automate the factory line, which creates more demand for cars, so we need more designers, etc. But ultimately the design becomes automated. Well, maybe we need more factory managers or other high level executives. Leaving aside that the number of those roles is inherently limited, even those roles could eventually be automated.
There's a trick, oft forgotten in these discussions. You don't have to just automate your existing workflow and hope not too many gaps remain. You can always build a new workflow around automation, playing to its strength, and eliminating human factor at the system level - or even business model level. Hell, iff no machine can handle a particular task, find a way to not require it at all. If your marketing team is good, it'll even sell it as a feature.
Car manufacturing is actually a great example of that, with Elon Musk saying "Yes, excessive automation at Tesla was a mistake." Manufacturing and "pick and place" are ongoing areas of research - for now, humans are unrivaled in them.
The great thing about computers is that they do what you tell them to do. The bad thing about computers is that they do what you tell them to do.
The world is very complex, and it's often hard for the programmer or even the person training the AI to foresee every corner case of a complex system. The good thing about humans is that they can react on very small data, by knowing things from experience. They can think about what they're doing.
If we make advances toward a general AI this gap could close. Except in the media hype cycle I haven't seen any evidence that we're any closer than 50 years ago.
I always viewed the singularity-utopia / terminator-dystopia with a certain disbelief. In the end what drives all the progress and innovation is the demand. There will be new demands once the old ones are met.
Saddlers, Water-Carriers, Coal delivery men all went out of business, but productivity, income, quality of life is up compared to then and no one wants to have those jobs back.
Disruption will occur, but most of humanity will profit. If you want to help, concentrate on helping those that need to make a transition, instead of spreading fear of progress. Because then you'll just increase the amount of people in need to transition and reduce the available capital to help them out.
The problem with this line of thinking is that the cost of the transition is pushed onto the employee. Automation is a huge gain for the company while the employees whose work got automated need to pick up new skills themselves on their own time and cost. No one in their right mind wants to spend their senior years getting retrained and then having to compete with new grads. Ask yourself - would your boss be willing to hire a 60 year old programmer that spend most of his life doing something completely unrelated?
I'm okay with the singularity as well, though I gotta admit I have some reservations about it. Like, what if I get sacked because AI is taking over? Yes, they said AI could boost employment and will create new jobs but what if it doesn't? It will create unemployment problems for sure and when it does, other problems might follow.
AI actually significantly effective in practical business would be strongly deflationary and put a lot of people out of work. The deflation would hurt people who have loans on real assets.
Then the good news is, the Fed just prints enough money to counter the deflation, and with the cheap money can hire people, if not for anything very practical then for things that are fun, e.g., high end home decorating, art, pure math and science, far out projects in computing, gorgeous roads, bridges, buildings, gardens. Then also for more things that are practical, e.g., medical research, curing cancer.
Everything the AI driven robots do will get done for less money, maybe a lot less. AI used for ad targeting will let advertisers get more for their money, spend less on ads per dollar of revenue, make advertising more efficient and, thus, lower the costs. E.g., some of the old print publishers learned a very hard lesson the hard way, "paid full tuition", about how much computing, and in particular, early Google style ad targeting, could be for ads.
If an AI controlled farm tractor can plow, fertilize, seed, cultivate, and harvest the corn or wheat crop on the North 40 acres, or the south 2000 acres, the farmer can do other things, corn and wheat prices can fall, will fall due to competition, etc.
Again, automation has made big, huge improvements in economic productivity, e.g., let what used to be mostly for royalty at huge prices available to nearly everyone at low to essentially free prices. So the old prices have fallen enormously, and that's definitely deflation.
E.g., we got our first electronic digital computers by people putting together vacuum tubes, resistors, capacitors, inductors. "If you have to ask what it costs, you can't afford it." So, maybe the price would ruin a few $million. Now, with massive automation of microelectronics, hurry up, buy before the price goes down farther, can get the processor in my startup's first server, and AMD FX-8350, 64 bit addressing, 8 cores, standard clock speed 4.0 GHz, quantity 1, retail, for less than $100, with money now worth ballpark only 10% of then. So, automation has produced several factors of 10 less in cost. Similarly for agriculture versus food gathering, ocean sailing versus oars, steam ships versus sailing ships, light from electricity versus whale oil, coal gas, kerosene, candles.
Apparently some applied math called AI has helped reduce some costs. Deflation.
If advertising costs go down because of AI, then advertising prices will go down too. That money then goes somewhere else entirely, but anyhow gets spent.
If there seems to be more money around for various causes than before, that looks to the market as if money supply has increased. Which definitely will cause inflation.
Business seems to be getting more profitable because of that applied math, so investing in it seems smart. True deflation would see value of money to shoot up so quickly, that you are better of if you just stuff cash under your mattress.
The article basically says that it could reduce employment for certain industries depending on the market it's automating. Whereas reading the title made me think it was going to be an article solely focusing on the positives of AI on employment.
And it doesn't even go into whether the boost that does come is better. It's very possible that a dozen or so low paying part time jobs are worse than having one high paying job.
Its thesis is that automation of an industry first increases employment in it, and finally decreases it.
This is because early stages of automation reduce very high costs to a manageable level, increasing demand and increasing employment despite increased efficiency.
Later stages of automation reduce already low costs, so demand does not increase much and the efficiency gains translate to reduced employment.
It’s also possible to reach market saturation regardless of price point. Even if a meal costs less than 1 cent and even if that’s all anyone eats, you’re going to have a really tough time getting fast food sales to exceed around 3-4 meals per day for every person on the planet. How many mattresses can a person use? How many pairs of scissors do you keep in your home, and how often do you replace them? How many haircuts would you get per month if they were free? How many root canals? Lowering prices may increase demand, but at some point, the concept of demand increase becomes ludicrous.
I have yet to see a single example of "AI automation" causing anyone to lose their jobs. In fact, I have yet to see a convincing example of "AI automation" of anything.
Please HN school me if there is an obvious example I am overlooking.
* specifically designed and implemented algorithms (ie. tax prep)
But all of these are in some ways very primitive AI algorithms (or you might say they're all advanced heuristic-enhanced breadth-first-search algorithms, including AI algorithms ... certainly true). And about the specific algorithms, these days they all use some form of AI (at the very least for expense classification).
The initial phases of AI will actually involve more jobs. Why ? The short of it is that what used to be middle management is much easier to automate than "individual contributor" roles.
I guess pagerank counts as machine learning; totally not AI though.
Maybe translators is a good example. Machine translations actually are pretty crude, and IMO they got worse when they started using "advanced" stuff like dweeb learning; KNN was better. KNN is usually better, as it just uses the data, rather han a model of the data.
The other examples are not even in the ballpark. They're just algorithms.
See, when you say "in the end AI will kill most jobs" -people keep saying that, and it keeps not happening. I mean, you might be totally right, but it sure as hell isn't happening real fast or in any obvious way. If I look at the 8086 world of the late 1980s, it isn't super different from now. What awesome human superpower do I have thanks to 'AI?'
> What awesome human superpower do I have thanks to 'AI?'
Ok, let's name a few:
You can filter through 400 hours of video every minute (youtube's upload rate) and find, perhaps not THE most interesting videos to you, but at least videos interesting enough to watch a lot (Youtube).
Those videos can be subtitled. Again not perfect (not that human translators of videos were -on average- very good)
They can collect not just bucketloads, but absurd amounts of news and facts and write short summaries of what happened (ie. most of the newspaper articles you read these days)
You can instruct machines to do things just by asking them (alexa, cortana, google).
You can talk, over the phone, to thousands of people at once (using google cloud or amazon service, or you could hire a few machine learning engineers and give them a year to write stuff)
You can "program" robots to walk, crawl, and twirl pencils without so much as knowing what a feedback loop is.
You can generate infinite amounts (and lengths) of well-composed music.
You can generate infinite amounts of beautiful paintings or even pictures (and even video, really) of whatever you want in whatever style you want.
You can program a car to drive itself (getting a basic DNN algorithm up is not even that hard).
You can have a drone fly itself through dense foliage and not hit stuff, and more generally many tricks with drone flying that no human, or human programmed non-learning algorithm can do (and yet we still insist on model-based autopilots)
You can watch incredible amounts of timeseries and find things acting weird (very useful that one)
I didn't think I needed any of these when I was I running around schools in 80s...also, most things above seem like those that we could collectively live without and be happier.
Video filter: not AI, doesn't work as a video filter (aka searching youtube isn't actually searching video content), also nobody but NSA cares
Video subtitle: granted; pretty cool on the rare occasion it works in a usable way
News summaries: no, you can't do that
Talking machines: HMM existed in useful form in the 80s
Programming robots with 'AI' -no you absolutely cannot do this; anyone who tries to sell you this is a charlatan
Generating 'well composed' music/art: total nonsense -name one such piece of music or art. FWIIW LZW also generates canned music, and isn't 'AI'
Programming car to drive itself: no, you can't do this, despite the best efforts of Waymo's PR department to assert otherwise.
Drone flight; collision avoidance tasks & etc: nope total nonsense
"Watching timeseries:" this improves people's lives in what way? I'm aware of algorithms like SAX and so on; I'm also aware that moving average crossover is used a lot more often than ML based TS anomaly detection.
You have a big fat nothing burger above. 'AI' hasn't changed anyone's life in an appreciable way. The things you list above are ... very far from the singularity of mass unemployment that people seem to envision. Most of them actually seem to create more jobs. Most of the things you cite above visibly do not exist in the physical world, outside of PR press releases. The only stuff which does work, basically all it does is increase the level of totalitarian surveillance possible.
> You have a big fat nothing burger above. 'AI' hasn't changed anyone's life in an appreciable way.
I don't agree, obviously.
> The things you list above are ... very far from the singularity of mass unemployment that people seem to envision. Most of them actually seem to create more jobs.
A lot of things that people "show to be possible" really haven't been shown to be possible; autonomous vehicles being the most obvious elephant in the room.
Your argument seems to be "we have gradient boosting and HMM and they are useful for things" which I certainly don't disagree with; I mostly make my living using such tools.
What I do disagree with is the idea that it meaningfully increases human power over nature, or changes people's day to day lives. A list of things which meaningfully increased human power over nature: flight, radio, computers, networked computers, electrification, antibiotics, the microchip, nuclear power, the green revolution, steam engines; maybe even stuff like Pagerank/sparse SVD.
The subject known as 'AI' seems to be down around ... I dunno, electric lawn mowers in its impact on human life. You can assert that the difference is gonna be real big one of these days; people made such assertions as far back as the 1960s, but so far, it totally hasn't made much of a difference. Yet your prediction, completely without any basis in the corporeal world, is that AI will eventually eliminate most jobs. Why do you think this?
Disclosure; I attempted to patent this idea before they were public with it, because I know the autonomous claims are baloney, and there was an obvious way to look like it wasnt baloney.
I also just spent two weeks driving an Audi A-8 around Germany; most advanced 'autonomous' technology there is in a production car supposedly.
I would expect nothing less, actually. This seems a very reasonable to provide an "out" for the AI.
Are you really going to assert that autonomous control doesn't exist unless there is no option at all for it to call for help ?
Your link says it itself:
"This is a backup system for moments when the car isn’t sure what to do, so it comes to a safe stop on its own and calls for help."
The article likens it to elevators. Because elevators all of them "have operators". Really. Behind a phone line. Would you really argue elevators don't move by themselves at the direction of a customer, but only move at the oversight of a remote operator ? Because that's one of those things that's technically true ... and yet so wrong it's almost ridiculous.
Now of course self-driving tech isn't quite up to elevator level of reliability yet.
Also I'd like to point out that even under those circumstances, people would still lose their jobs. Because it'll allow 1000 trucks to be controlled by, first 200, then 100, then 10, then 1 drivers. Even if 80% lose their jobs, truck driver will stop being a significant percentage of employment, and if it was one company it is now the biggest employer in more than 70% of all US states.
As in, I hear you, and very interesting article, but it doesn't seem to change anything in the discussion.
You're really persistent in your beliefs, I'll give you that.
I suppose I am too; filling out those patent forms was a lot of work.
Query: why do you think the military doesn't have any autonomous ground (or air) vehicles? Military virtually always has the most advanced tech; unlimited budget -lots of motivation for supply lines which run themselves and can't be shot at, lots of reasons not to expose expensive soldiers to fire, and very little care for stuff like safety. Yet, somehow they have no such things. I wonder why that is?
They don't have remote control supply lines either, and those we can certainly make, which you seem to agree with ... so I must say I don't know.
I have consulted for the military, and frankly I was amazed that there seems to be no correlation between intelligence and rank. Especially some lieutenants were incredibly smart but several colonels ... I wouldn't trust them with tying their own shoelaces. This resulted in extremely smart and well organized groups trying to work with complete morons.
And I must say, that results in a pretty stupid whole. The damage the idiot groups inflict is not compensated by the better parts of the organization. Not in logistics. Not in radar operation (where I consulted). Not in organizing the cafeteria. It seems to me that it is utterly impossible to get fired for incompetence in the military. So they keep working with total idiots, even promote them.
What I'm saying is that it's the military. They have awe-inspiring gear. They have some very smart people (they pay consultants in some areas very well, so the truly do have incredible people that are such a pleasure and inspiration to work with). They also have some incredible idiots at every point on the rank ladder. And occasionally they decide on policies so incredibly stupid they are equally awe-inspiring, because the idiots do much more damage than smart ones can fix.
I don't believe it is reasonably possible to predict what the military will do. Especially when it comes to policy.
The article does not go far enough in expanding the scope of consequences. When consumers start buying more stuff at lower prices, they are diverting money that would probably be spent on other things. Hence there is no net change in consumer spending, only a shift in where they spend their money.
Automation always results in labor cost reduction for the one implementing the automation. If it did not, they wouldn't do it.
That's a misunderstanding. For each unity of a good, production costs go down, as do labor time. But because the goods are cheaper, more people may want to buy them, or people may want more of them, so the total labor time goes up.
That effect become known by the name of Jevon's paradox. If you want to, wikipedia has a large and well explained article. It has been spotted on practice enough times so that one should not question its existence. It's also the reason why software development salaries have kept increasing while the tooling allows for larger and larger productivity.
That said, it only happens some of the time, when demand has the correct properties.
The hardest part would be understanding how much people will adjust their long term behaviors. Grandma might continue wearing that one pair of clothes out of habit. But the younger generation will more quickly adapt to the new norm.
This is the fundamental problem with this article IMO. It assumes that human labour will be required to fill the gaps in an automated production cycle. But the ultimate premise of automation is that there are no gaps.
If you look at something like car production, first we automate the factory line, which creates more demand for cars, so we need more designers, etc. But ultimately the design becomes automated. Well, maybe we need more factory managers or other high level executives. Leaving aside that the number of those roles is inherently limited, even those roles could eventually be automated.