I don't agree with most all of it.
I read it down to the conclusion.
"We explore the effects of AI on the skill premium. To this end, we develop a nested CES production function in which industrial robots predominantly substitute for low-skill workers, whereas AI predominantly substitutes for high-skill workers. We show analytically and numerically that AI has the potential to reduce the skill premium and thereby mitigate or even reverse increases in inequality that have been observed in recent decades."
Why I don't agree; robots, CNC, AI all still greatly enhance the abilities of high skill workers way more than low skill workers.
While to they raise the ability of low skill workers, they similarly raise the ability of high skill workers, not scientifically, by the same amount.
They raise the salary of low skill workers because they can now do things they could not do, supported by the paper. But this is also true of high skill workers.
There is not a cap on ability or productivity. For example, similar to AI, a CNC machine be more effective in the hands of a high skill worker than a low skill worker. What AI might do is lower the entry of them to low skill workers.
In the end all the wages may go up, but the disparity will be the same.
I'm an economist and I think just posting the paper misses a bit of context:
There's a bit of an arms race between researchers to write what they hope will be influential papers about how ChatGPT will affect (the economy|inflation|gdp|productivity|inequality|whatever).
I would say that at least 90% of all the new papers are quite useless, so not surprised that you find that
My loose thought model uses an AI assistant on a PC or similar thats constantly trying to predict and help you, and that someone actually makes this desirable and not just Clippy 3.0. This AI mimicry will advance high skilled productivity quite a bit, but when an AI has been well trained on a high skilled person, it can then be backwards applied to a low skilled person as a decent bump in productivity, and also potentially be run solo by software thieves in the third world for like 1/4 of the benefit of hiring someone. And that this sort of creates a marketplace for trained assistant AIs, "This one trained on an electrical engineer for 3 years" "This one trained on an entire call centre for 10 years, no longer needs guidance" etc etc.
if you're of the best story or essay writers, chatgpt won't be able to help you much. If you're one of the very top programmers, you will find chatgpt full of all kinds of stupid mistakes in the work you doest (which is hard). If you are not a native speaker (like me), chatgpt is magical for helping me write. If you're an average developer, chatgpt would offer tremendous help as most problems you encountered are common problmes that chatgpt is likely to solve (or contain fewer mistakes in doing so). So no it is not equal. High skill workers certainly lose more comparatively.
There is zero value, or negative value, from an employee that just copied and pastes the output from ChatGPT and runs it. The employer can just do that herself and save the cost of the “developer” entirely.
"Why I don't agree; robots, CNC, AI all still greatly enhance the abilities of high skill workers way more than low skill workers."
At somepoint CNC won't need humans making things in CAD. You'll just tell and AI what to great and it'll model it and plug it all in.
I believe there will be job loss in both categories too. Obviously that leads to inequality as well. Sure, some productivity increase will be helpful, but after a certain point if demand doesn't also increase it will result in headcount cuts.
I feel bad for the next generation. I'm not sure what ro tell my kid to study when they get older. It seems the only places with longer-term stability will be with artificial or natural moats. Maybe law (litigation), law enforcement, or medicine. Maybe a business degree since that could be applied to many different things. I don't know. The choice seemed easy for me - I was good with computers and logic, and IT paid well.
> You'll just tell and AI what to great and it'll model it and plug it all in.
That's the same as writing a book. Anyone can patch random amounts of text together and call it a book. Will it be good enough for people to read? Only if the person doing it is good enough to create an appealing story. The same is still valid for AI: some people know how to use it better and will get better results while 99% will be trash and ignored. This won't ever change, imo.
I generally agree with you except that all the wages may go up: there are tools that directly replace people and their wages go down or just disappear from the scene.
AI is just one powerful example but there are a lot of recent example that sounds more basic: before [Twitter] Bootstrap was created companies of all sizes hired graphic designers to create their web page. Most of this work now is compressed in time and the same graphic designer do a smaller part.
At the same time, in the software development field we invent all kind of tool that increase the size of the team for other kind of people (e.g. React developers). Regarding AI, Low Code, or whatever concepts we use in the following year I think the role of software development will also decrease.
People touting automation rarely follow up after the initial effort in my opinion
For example, i work on heavy diesel trucks. We've had tire machines to mount and unmount tires for 40 years now. Machines still can't so the entire process though.
Proposition 1. The growing use of AI, ceteris paribus, reduces wage inequality between high-skill and low-skill workers, as long as AI is more substitutable for high-skill workers than low-skill workers are for high-skill workers.
This makes sense within the confines of the model and assumptions. But:
- perhaps AI is less like industrial automation (which substitutes for labour) and more like assembly line machinery (which complements labour)
- the static 'low skill' / 'high skill' divide assumed by the model, seems like a major limitation; it doesn't allow for the idea that AI will split the 'high skill' group into two groups: those who can be mostly replaced with AI, and those who can do stuff AI isn't good at yet.
Either of these alone (or both together) would suggest increased wage inequality.
It also only addresses the substitution of high skill labor as it pertains to generating text/image/etc. and not the assessment nor integration of that output.
It's been my experience that GPT can help with the generation and can try to help with the assessment. But it is not consistent on assessment and is very lacking in integration of the output.
This varies some based on the task and the kind of high skill being considered. Generating copy or stock images? Re-rolling is easy enough and integration is already heavily assisted by existing tooling. Writing business logic for a low risk backend system? Integration is a little harder but overall I could see it lifting a low skilled worker to nearer high skill. Designing a large system with many moving parts? Prepare to re-ask your questions a lot and your assessment skills had better be already in the high-skill area for that task, I wouldn't trust GPTs beyond a certain scale or complexity.
I think that we'll see a lot more solopreneurs and we'll see people use it to help learning things that they would have otherwise thought were beyond them. And the bifurcation of high skill workers will be more about those who use it in their work and those who can't or don't. But I'm not as confident as this article is about high skill jobs like programming entirely replaced by GPTs.
>> perhaps AI is less like industrial automation (which substitutes for labour) and more like assembly line machinery (which complements labour)
These are really rather similar from a bird's eye view. In the second case, you complement labor and increase the productivity of an individual worker. However, in relative terms, you reduce the number of workers employed and/or increase the amount of capital used. This marginally decreases wages and increases returns to capital.
Once the dust settles - the famous "long run" - you've increased the size of the economy by increasing productivity. This results in a slightly higher wage and a relatively even higher increase in the capital rate of return, thus increasing inequality.
But then - in the even longer run - the capital stock grows faster than population, so you get the opposite trend. Capital becomes relatively more plentiful, favoring more capital-intensive production functions. This helps wages on the teeter totter.
I think the most obvious changes are going to be the increases in the relative wages (and social prestige) of jobs that are neither low-skilled nor AI-augmentable, i.e. the "stuff AI isn't good at" and might never be. An illustrative edge case might be masseur/masseuse.
> How will the emergence of ChatGPT and other forms of artificial intelligence (AI) affect the skill premium? To address this question, we propose a nested constant elasticity of substitution production function that distinguishes among three types of capital: traditional physical capital (machines, assembly lines), industrial robots, and AI. Following the literature, we assume that industrial robots predominantly substitute for low-skill workers, whereas AI mainly helps to perform the tasks of high-skill workers. We show that AI reduces the skill premium as long as it is more substitutable for high-skill workers than low-skill workers are for high-skill workers.
I don't know if I agree with the "literature" as represented here. It seems to thoughtlessly conflate low-skill with blue collar and high-skill with white collar. I would expect the impacts of AI on wages to vary across skill levels of white collar workers.
Yeah I was going to say they seem to be equating anybody doing "office work" as being "high skill" and there are tons of mediocre hacks in office jobs doing menial, brain-dead work. I think those people will continue to do menial, brain-dead work, except now it will leverage AI.
95% of people or some ridiculously high number like that used to be involved directly in agriculture in some way. Now a small percentage of americans do farm labor. If you told someone in 1845 that in 2024, only 10% of americans were working on farms, they would probably naively assume that the vast majority of us would be living lives of idle leisure, when what actually happened is that automation just made different, more advanced work economical to engage in.
There is not a limited amount of work that can be done, and if we relieve people from having to do some kinds of labor, they'll figure out how to spend their time doing stuff that would have been unimaginable before.
The scenario that I bring up a lot is that there is not a limited amount of movies and books that could be produced a year. We only produce the amount we do because that's how much we can produce _economically_. But if you had a super powerful AI that could actually cheaply produce movies on demand near the quality of professional movie makers (something that I don't imagine will happen any time soon, but it's the "worst-case-scenario" for creative workers), people will be spending their time making their own personal streaming service -- movies just for them, on demand. TV shows that never end, tailored for the individual's taste. There could literally be 10s of billions of hours of media produced a year, all of it at the quality of a modern block buster -- all of it that might be viewed by only a single person. And there will still be plenty of jobs available at the companies that are producing that media.
I think what realistically will happen is high skill workers will be making far more complex and valuable things with the help of AI than they ever could without. Same thing happened after the Industrial Revolution. Eg in software engineering AI could enable design and maintenance of much more complex software than the stuff we can design and maintain today. And it will happen, because frankly software as it exists today is barely adequate, after almost 70 years of running up against the limits of human cognition and stopping there.
I responded to you in earnest (no AGI needed for the conclusions to hold in this paper) but I’ve realized that you’ve also posted the same link under two other submissions. It is not great practice to plug your own stuff in places where they are clearly off topic.
"We explore the effects of AI on the skill premium. To this end, we develop a nested CES production function in which industrial robots predominantly substitute for low-skill workers, whereas AI predominantly substitutes for high-skill workers. We show analytically and numerically that AI has the potential to reduce the skill premium and thereby mitigate or even reverse increases in inequality that have been observed in recent decades."
Why I don't agree; robots, CNC, AI all still greatly enhance the abilities of high skill workers way more than low skill workers.
While to they raise the ability of low skill workers, they similarly raise the ability of high skill workers, not scientifically, by the same amount.
They raise the salary of low skill workers because they can now do things they could not do, supported by the paper. But this is also true of high skill workers.
There is not a cap on ability or productivity. For example, similar to AI, a CNC machine be more effective in the hands of a high skill worker than a low skill worker. What AI might do is lower the entry of them to low skill workers.
In the end all the wages may go up, but the disparity will be the same.