The data is very interesting but the timing covaries with so many other things -- 22-25 is the cohort most affected by pandemic, the rise in online classes, and automatic job search platforms. A young person could go through college without meeting a single person who could "vouch" for them or mentor them as they move from school to the workforce. I don't see generative AI as the key variable here. My full take here: https://hollisrobbinsanecdotal.substack.com/p/the-canary-in-the-classroom
There's also the impact of higher interest rates starting in March 2022 that slows down hiring. It seems logical this would be more likely to impact people newer to an industry because they have fewer people who can vouch for them and are likely to take longer to get up to speed and become profitable employees. This would be more likely in tech than things like health aides or construction where the work is more immediate.
I don't doubt that AI is having some effect, but like you, there's a lot of factors. There had just been a wild hiring spree for tech companies prior as well.
This is purely anecdotal but for me the impact of AI on the hiring process for junior engineers has been a *nightmare*. We got thousands of resumes for a listing that would have gotten less than a hundred a couple years ago. Then 90% of the candidates we reach out to don't even respond - that would be under 10% a couple years ago.
I'm sure there are savvy recruiting teams out there that are doing a better job here, so I do want to stress the anecdotal nature of my comment. But for us, "Friend-of-a-friend" is the only thing that is working.
p.s. if you happen to be a newly-graduated physics phd with no experience who is nevertheless good at programming and wants to work on a cool large scale radio astronomy project, let me know.
Yes, "friend of a friend" is the best approach. The problem here is the hiring system and AI is clogging all ends, which is I suppose a form of taking jobs...
For the future of the labor market, I take heart from Ricardo's concept of comparative advantage. It gives us a very powerful argument that even someone who is outclassed in every specific way by someone else will nonetheless have some kind of job available to them that the other is willing to pay for. The basic idea is that even if you can do every single thing better than me, you'd still rather do your own best thing and then pay me to do something else, to free yourself up.
Young people are a shining example of that kind of thing. It can help to consider why a teenager gets hired at all, even before AI. It's not because of their stellar skillset. Now imagine we are all the teenager in this thought experiment.
> It gives us a very powerful argument that even someone who is outclassed in every specific way by someone else will nonetheless have some kind of job available to them that the other is willing to pay for.
Assuming zero transaction costs, which is not remotely realistic here. And even if it were, the fixed costs of maintaining an adult human in good health creates a natural price floor for unskilled labor.
I agree on both points about how the theory works out.
It feels like a different question whether people can do *anything* that is more valuable than the food and sunlight we take in to survive. As a first stab at that question, it seems like humans were able to do enough useful things in our old Darwinian era that we could make the things we need to propagate and survive. It seems like that should still be the case in the most extreme AI-heavy future. Now, those AIs can probably do it cheaper than humans, but that's back at the original goalpost. The new one you specified is if keeping the human alive is net positive compared to the resources.
There's an implicit assumption of scarcity in your analysis. How does it change when "better than teenager" AIs can work for arbitrarily close to nothing, in whatever number you need? As many as you want, whenever you want.
What happens to the teenagers then? They cease to participate in the market.
...and become a pool of recruitable talent and muscle for a burgeoning fascist political movement that feeds off their resentment, disconnection and lack of better things to do...
Where they are either harassed or goaded/recruited via in-game text and voice chat(This is not me handwaving or pearl-clutc hing, it's a well-known & well-studied phenomenon).
Honestly, it feels likely that for things like writing a novel or developing software to spec or painting a picture or developing math theory, AIs are likely to just plain out win. These things will not be jobs in the future.
There are two areas humans seem likely to be competitive enough to beat out the scarcity element.
One is that AIs are not free, and they are especially not free for anything that requires manual labor in the physical world. For these less thinky tasks, an AI in a robot may well do it better than a human, but there wouldn't be some level of scarity for the AI in that case.
Two is that, so long as there are any humans at all, many things we find valuable require that the other person be as human-like as possible. So, so long as there's a kernel of *any* reason for humans to be around, then it cascades into more reason due to we ourselves wanting things from each other.
With regards to your point one, even without embodiment AI could reduce much skilled manual labor to unskilled labor. i.e. instead of calling in a plumber to fix your problem, you would show GPT-7 the situation (using your phone camera, say), and it will tell you `here is what's wrong, apply a wrench exactly here, turn it in this direction, now stop' etc.
Nothing in comparative advantage theory guarantees that your market clearing wage in this scenario will be high enough to allow you to purchase 2000 calories of food a day (never mind a house or anything else).
1. I don’t buy the Stanford study at first blush: how in the world could AI explain large shifts in employment THREE YEARS AGO? Three years ago AI was still science fiction for the vast majority of the population.
2. One way to test your first “yes definitely” impact would be to look at what routes each successful hire in those categories took. Were the younger groups disproportionately using the ubiquitous “spray and pray” approach while the older groups alternatively more likely to have found positions based on prior relationships? If that hypothesis is true then if you adjust for method of job search, the decline may actually be identical across age groups.
Another hypothesis is that younger workers, especially those whose educational experience was directly affected by COVID are less likely to possess the social skills that are needed to ace an interview. This would point to the big societal shifts that are happening in youths who are digital natives, as you’ve referenced in your education and dating posts.
The partially AI-driven job application spam issue seems likely heavily reduced by introducing sufficient friction in the form of a low cost - a few £/$ etc - to the transaction. It would stop many mass-spammers/scammers and also induce genuine applicants to put more effort into their CV and tighten the focus on which jobs are being applied for.
Though, I suggested this on a forum for IT contractors (usually 90th percentile in terms of salary/rates) and was surprised to find this highly-remunerated population dismissing the idea and recoiling at the thought of paying to apply for a job. The way I see it is if a £5 charge on an application for a £500/day gig increases my chances of landing it from, say, 1% to 2% then that's a bargain, the EV is huge.
It sucks, I get that, but the alternative is either do nothing but moan about the situation, or perhaps some forms of automated testing - but that seems easily gameable by AI too. Plus, how fundamentally different is it really compared to buying a nice suit and tie for a bunch of interviews? The cost of the hairgel and aftershave/perfume/makeup/transport in the morning that would otherwise not have been expended?
I think to stop scammers, either the money would need to go to the job site, with the caveat that they're audited for not creating fake companies/jobs themselves, or the company/agency gets it but is charged a sufficiently high cost to place the ad that it's not profitable to create fake jobs (possibly already the case).
Seems like there has to be way to solve the problem by everyone involved having sufficient skin in the game.
It’s a shame that the article stating a significant impact of AI doesn’t get the same scrutiny as the one stating the opposite (see the previous post from Zvi).
*Predictions are hard, especially about the future, but I would as strongly as always disagree with this advice from Derek Thompson:*
Is there any other kind of prediction?
Also, when do I get an AI that washes the dishes and sorts the laundry?
Yes, history. Not always correct, but usually slightly better than about the future.
It’s a joke.
The data is very interesting but the timing covaries with so many other things -- 22-25 is the cohort most affected by pandemic, the rise in online classes, and automatic job search platforms. A young person could go through college without meeting a single person who could "vouch" for them or mentor them as they move from school to the workforce. I don't see generative AI as the key variable here. My full take here: https://hollisrobbinsanecdotal.substack.com/p/the-canary-in-the-classroom
There's also the impact of higher interest rates starting in March 2022 that slows down hiring. It seems logical this would be more likely to impact people newer to an industry because they have fewer people who can vouch for them and are likely to take longer to get up to speed and become profitable employees. This would be more likely in tech than things like health aides or construction where the work is more immediate.
I don't doubt that AI is having some effect, but like you, there's a lot of factors. There had just been a wild hiring spree for tech companies prior as well.
Yes that's a good point about the interest rates too.
This is purely anecdotal but for me the impact of AI on the hiring process for junior engineers has been a *nightmare*. We got thousands of resumes for a listing that would have gotten less than a hundred a couple years ago. Then 90% of the candidates we reach out to don't even respond - that would be under 10% a couple years ago.
I'm sure there are savvy recruiting teams out there that are doing a better job here, so I do want to stress the anecdotal nature of my comment. But for us, "Friend-of-a-friend" is the only thing that is working.
p.s. if you happen to be a newly-graduated physics phd with no experience who is nevertheless good at programming and wants to work on a cool large scale radio astronomy project, let me know.
Yes, "friend of a friend" is the best approach. The problem here is the hiring system and AI is clogging all ends, which is I suppose a form of taking jobs...
What location for the project? I have a friend matching that description who might be a great fit.
About half the eng team is in Los Angeles (caltech) but remote is okay. The telescope itself is in the middle of nowhere. Contact Lacker at gmail
For the future of the labor market, I take heart from Ricardo's concept of comparative advantage. It gives us a very powerful argument that even someone who is outclassed in every specific way by someone else will nonetheless have some kind of job available to them that the other is willing to pay for. The basic idea is that even if you can do every single thing better than me, you'd still rather do your own best thing and then pay me to do something else, to free yourself up.
Young people are a shining example of that kind of thing. It can help to consider why a teenager gets hired at all, even before AI. It's not because of their stellar skillset. Now imagine we are all the teenager in this thought experiment.
> It gives us a very powerful argument that even someone who is outclassed in every specific way by someone else will nonetheless have some kind of job available to them that the other is willing to pay for.
Assuming zero transaction costs, which is not remotely realistic here. And even if it were, the fixed costs of maintaining an adult human in good health creates a natural price floor for unskilled labor.
I agree on both points about how the theory works out.
It feels like a different question whether people can do *anything* that is more valuable than the food and sunlight we take in to survive. As a first stab at that question, it seems like humans were able to do enough useful things in our old Darwinian era that we could make the things we need to propagate and survive. It seems like that should still be the case in the most extreme AI-heavy future. Now, those AIs can probably do it cheaper than humans, but that's back at the original goalpost. The new one you specified is if keeping the human alive is net positive compared to the resources.
There's an implicit assumption of scarcity in your analysis. How does it change when "better than teenager" AIs can work for arbitrarily close to nothing, in whatever number you need? As many as you want, whenever you want.
What happens to the teenagers then? They cease to participate in the market.
...and become a pool of recruitable talent and muscle for a burgeoning fascist political movement that feeds off their resentment, disconnection and lack of better things to do...
That is how things were, yes. These days? Play video games in their mom's basement.
Where they are either harassed or goaded/recruited via in-game text and voice chat(This is not me handwaving or pearl-clutc hing, it's a well-known & well-studied phenomenon).
Different century, same tactics.
But by that point they're still more expensive to maintain than drones and automated forces that are better than they are at violence too.
We use humans to deploy automated forces to other countries...
True.
Honestly, it feels likely that for things like writing a novel or developing software to spec or painting a picture or developing math theory, AIs are likely to just plain out win. These things will not be jobs in the future.
There are two areas humans seem likely to be competitive enough to beat out the scarcity element.
One is that AIs are not free, and they are especially not free for anything that requires manual labor in the physical world. For these less thinky tasks, an AI in a robot may well do it better than a human, but there wouldn't be some level of scarity for the AI in that case.
Two is that, so long as there are any humans at all, many things we find valuable require that the other person be as human-like as possible. So, so long as there's a kernel of *any* reason for humans to be around, then it cascades into more reason due to we ourselves wanting things from each other.
With regards to your point one, even without embodiment AI could reduce much skilled manual labor to unskilled labor. i.e. instead of calling in a plumber to fix your problem, you would show GPT-7 the situation (using your phone camera, say), and it will tell you `here is what's wrong, apply a wrench exactly here, turn it in this direction, now stop' etc.
Nothing in comparative advantage theory guarantees that your market clearing wage in this scenario will be high enough to allow you to purchase 2000 calories of food a day (never mind a house or anything else).
That perfectly lines up with interest rates shooting up, which makes very solid, core economic sense. Does the paper account for that?
Three unrelated thoughts:
1. I don’t buy the Stanford study at first blush: how in the world could AI explain large shifts in employment THREE YEARS AGO? Three years ago AI was still science fiction for the vast majority of the population.
2. One way to test your first “yes definitely” impact would be to look at what routes each successful hire in those categories took. Were the younger groups disproportionately using the ubiquitous “spray and pray” approach while the older groups alternatively more likely to have found positions based on prior relationships? If that hypothesis is true then if you adjust for method of job search, the decline may actually be identical across age groups.
Another hypothesis is that younger workers, especially those whose educational experience was directly affected by COVID are less likely to possess the social skills that are needed to ace an interview. This would point to the big societal shifts that are happening in youths who are digital natives, as you’ve referenced in your education and dating posts.
The partially AI-driven job application spam issue seems likely heavily reduced by introducing sufficient friction in the form of a low cost - a few £/$ etc - to the transaction. It would stop many mass-spammers/scammers and also induce genuine applicants to put more effort into their CV and tighten the focus on which jobs are being applied for.
Though, I suggested this on a forum for IT contractors (usually 90th percentile in terms of salary/rates) and was surprised to find this highly-remunerated population dismissing the idea and recoiling at the thought of paying to apply for a job. The way I see it is if a £5 charge on an application for a £500/day gig increases my chances of landing it from, say, 1% to 2% then that's a bargain, the EV is huge.
It sucks, I get that, but the alternative is either do nothing but moan about the situation, or perhaps some forms of automated testing - but that seems easily gameable by AI too. Plus, how fundamentally different is it really compared to buying a nice suit and tie for a bunch of interviews? The cost of the hairgel and aftershave/perfume/makeup/transport in the morning that would otherwise not have been expended?
If the money goes to the poster, this creates huge incentive to post jobs without intent to fill them just to collect application fees.
I think to stop scammers, either the money would need to go to the job site, with the caveat that they're audited for not creating fake companies/jobs themselves, or the company/agency gets it but is charged a sufficiently high cost to place the ad that it's not profitable to create fake jobs (possibly already the case).
Seems like there has to be way to solve the problem by everyone involved having sufficient skin in the game.
Give the money to a charity designated by the applicant?
It’s a shame that the article stating a significant impact of AI doesn’t get the same scrutiny as the one stating the opposite (see the previous post from Zvi).