In my space AI is mostly used to help doctors upcode patients and increase Medicare revenue.
There is also an arms race with AI re-submitting claims and other AI denying claims. Like your example of lawyers creating work for each other, I could easily see competing AI creating work for each other.
There is a lot of dumb Red Queen Race Bullshit out there that people forego because it would be "too expensive" to do. If you lower the price, they can increase the quantity. This is more or less the thesis behind why we don't have Keynes 15 hour workweek.
In a longer reply to John Hall I note that if olds and poors prefer humans CMS has a lot of ways to incent using humans at the taxpayers expense. CMS would much rather have sales brokers fielding member complaints then having to do it themselves.
I'm quite sure there's an arms race between LLMs writing resumes and cover letters and LLMs screening them.
In the end, it's kind of nice: I tell the LLM my experience bullet points, an LLM encodes that as flowery corporate speak, and the receiving LLM decodes it back to bullet points. Now, a sane human would observe we could save a lot of time and money by just skipping the prose stage but there we are.
My experience is blue collar on the delivery end(USPS). I can see a robotaxi type vehicle running the routes as well or better than I do in the near future or even today but I am having problems visualizing the actual to the doorstep delivery of packages. This is going to require many robots covering all sorts of terrain in the snow and the rain to place packages(opening doors) in porches. I could care less about a million dollar lab robot dancing around some lab. A 5 to 7 year time limit means you need a factoring somewhere turning out a thousand of these robots a week at an affordable price to dent a million plus person workforce. Same with trucking, I can see the truck going from point a to b but what if only half the product is being unloaded at point b. Do you just let the customer climb on in with his pallet jack and hope that he knows he is only supposed to be taking 10 pallets before the truck pulls away.. A human even a non driving human would really be helpful in this situation.
Inroads by AI and robotics will naturally be faster in some industries, slower in others. One factor is resistance. The battle against AI / robotics will play out, e.g., Trump’s support for union dockworkers to resist port automation. Union industries will be unanimously opposed to automation of any sort. For fun, I asked Claude to rank each union industry for resistance to automation, whether it be AI or robotics, differentiating between the U.S. and E.U. Answer -
Ranking Union Industries by Resistance to Automation (from Most to Least Resistant):
United States:
1. Construction (high skill, complex physical tasks)
5. Education (AI-resistant, but facing technological pressures)
6. Postal Services (highly automatable)
European Union:
1. Manufacturing (strong worker protections)
2. Energy/Utilities (complex technical skills)
3. Automotive Industry (specialized technical labor)
4. Metalworking (precision skills)
5. Public Transportation (moderate automation risk)
6. Public Sector (administrative roles more vulnerable)
European unions generally have stronger worker protections and more gradual technological transition policies compared to U.S. unions, potentially slowing automation implementation.
Public safety is an "Information" job and I expect technology to greatly reduce and change the work humans do. In the private sector, I expect human security guards and receptionists to be greatly reduced. At the same time, "public safety" in the government realm will probably expand in employment as government will find all sorts of reasons to add humans to manage and oversee technology solutions.
I think human management of technology is going to become a huge career field. And this is the natural progression. Electronics / computers gave us the Information Technology field. Then this became the Information Systems field. And now we will have the Information Management field. I'm polishing my resume now.
I really like these semi-Sci-Fi speculations, because many will come true and none of them are now false. Tho while intelligence gathering will increase, the huge % increase will be in useful intel analysis. Like an ai looking thru 14,000 J6 videos to get all of those for each of the 1500 or so pardoned protesters. And maybe all the violent & violence advocating guys who were NOT picked up by the FBI.
It would be good to have a Rep ai be watching, weaponized against Dem & govt actions; along with a competing Dem ai watching, weaponized against Reps & govt actions. With monthly transparency releases of suspicious behaviors by govt & related officials.
In 1900 there were approximately 5 million working horses in England propper - pulling plows, carriages, firetrucks, wagons, etc. Fifty years later in 1950 that number had dropped by 99% to approximately 50 thousand. They didn't DISAPPEAR but the VAST majority were replaced by diesel tractors, gasoline engines, and electric motors. The "Great Creative Destruction" affected them FAR more harshly than it did people. They no longer had a comparative advantage in anything that was significant enough to justify continuing to feed them.
I'm not saying it's going to happen, but I CAN forsee a scenario where a lot of people, maybe even the majority, can't do ANYTHING that can't be done better, faster, and more cheaply by an AI.
Do you go to live performance music? Do you go to sporting events? Do you go to theatre? Do you like stand up comedians? Have you been brought to tears by a dramatic performance on stage? Have you been brought to mirth by an author? Tears by an author? Have you been comforted in a time of grief or sorrow by a compassionate live person?
People who can do these things, and other things like them not listed here, will enjoy rising demand and rising incomes. People who are unable to bring the human touch to what they do will suffer falling demand and falling incomes. Beyond these observations, who knows?
Will Ai increase or reduce aggregate employment? I'm going with all of history; aggregate employment will increase, even though the jobs will be for things we cannot even imagine right now, try as we might.
Comparing Dr Kling's predictions to a set of "impact studies" described on the Palantir website (down towards the bottom on the right: https://www.palantir.com/ ) it seems to me like anything that would have benefitted from good operations research in olden days is probably now something that AI can improve on. For example, there are a couple hospital cases listed and AI is claimed to be producing staffing and scheduling and patient queue volume improvements. One wonders whether broad based quantitative operations research efficiencies enabled by AI might be as significant an influence on employment as simple replacement of workers by machines?
The Ferrari impact study, however, seems to provide a perfect, and interesting, example of how AI may wind up improving human-machine interfaces:
"Palantir software integrated several types of Scuderia Ferrari data, such as sensor data from races, tests, simulations and testing equipment, and parts information. The key source was data from car sensors in the form of high frequency time series.
Palantir contextualized time series data within a data model – the Ontology - and organized it into human-readable objects. This framework allowed Scuderia Ferrari engineers to interact with and analyze the data with code, low-code, or no-code tools; and optimize transformations that sped up calculations."
On the coding front, one wonders if the really big corporate hard coded applications doing stuff like dynamic pricing in real time are already so effective and efficient that the room for AI incremental improvements will be likely to meet cost-benefit tests?
"I don’t think that chatbots will be difference-makers here [agriculture, mining, forestry]. But productivity has been going up for a long time in agriculture, and that will continue, with some of the gains coming from machine learning. Dangerous jobs in mining and oil and gas extraction will increasingly be done by robots. Utilities may find robots useful for installation and repair."
If you're just thinking about LLM-based chatbots, likely true. However, I think you're understating how much AIs might affect things like agriculture. I keep reading articles about how AI-driven autonomous drones are collecting data and might enable very precise chemical applications (e.g. spraying Roundup on just the weeds, not the crops). I have to imaging mining companies are feverishly working on AIs to direct where to dig mines. I don't think this will be a small effect.
No one knows how AI will influence the number, distribution, or the nature of employment in our soon-to-arrive brave new world. A few days ago, however, I read that something like 54% of adults in Philadelphia are illiterate. If this is true, and I have no reason to doubt it, it won’t be too difficult for employers to adopt AI in creative and mundane ways to achieve greater productivity and better customer service. If our education system doesn’t make major improvements soon, the future will indeed be bleak for many of our fellow citizens.
I can't speak with confidence to the other sectors but with respect to Finance and Insurance there is a lot to agree at it relates to some functions, but I think you downplay the impact of human involvement.
US employment in the finance and insurance (so finance ex real estate and renting) is about 6.7mn. Of that, about 3mn is insurance, and another 1.3mn is commercial banking. I'm fairly sure that the technology has existed to have "AI" handle most of the primary responsibilities here (for instance: offering insurance quotes, a chatbot that can suggest what insurance people will buy, deciding whether to approve loans or not, etc.). And yet people are still hanging around. Why? Well I'm sure there are plenty of jobs tied to regulations in this space (for instance, humans need to review the models). But if chatbots are so good, then probably the place where they would save the companies the most money is replacing salespeople (think robo-advisors on steroids). Then, the companies could avoid paying pesky commissions. So I think that would be one place to look for success of this thesis. If companies can get away with paying lower commissions to salespeople (because their chatbots are so good), then people would quit those jobs (or fewer people take them). But I think there are a lot of times when you really want to deal with a human. And the more money is at stake, the more you are willing to pay for the privilege.
I don't disagree that there will be some jobs that are under pressure, but I think it's more of a transition (starting with the ones discussed in the article) than a shock. At the end of the day, we can look at the number of people employed in these industries five years from now to really evaluate this (it may be that we lose some of these jobs and gain others, not like that hasn't happened before).
The big money in insurance is Medicare/Medicaid. These customers are the least likely to be able to interact well with a non-human salesperson or tech support.
In addition, CMS heavily weights customer satisfaction surveys in STARS payments. If you save a few pennies on a chatbot call center versus human beings, you are probably going to give it back in lower STARS payments when they give you a bad CAHPS score for the experience.
There are also sales venues that would only work for human beings.
Finally, everything a chatbot does is going to be strenuously recorded. That means people can sift through it to see if it violated (even unintentionally) one of the millions of potential regulations in the space. A human being speaking to another human being isn't recorded, and so there is less on the record to scrutinize.
For the millionth time, "This is a blog for generalists." It is not a blog for people who are insiders in Medicare/Medicaid finance. Most people won't know what STARS or CAHPS are, or perhaps even what CMS is. So spell out what the acronyms mean or explain them.
Some people will think, "he knows what those acronyms mean; he must really know his stuff." But some people will think, "what an inconsiderate asshole."
STARS = A very large bonus pool paid to Medicare Advantage carriers that hit certain "quality" metrics.
CAHPS = A survey sent to Medicare Advantage members asking them various questions that amount to "are you happy with your health insurer". Has a large impact on STARS.
"If you save a few pennies on a chatbot call center versus human beings, you are probably going to give it back in lower STARS payments when they give you a bad CAHPS score for the experience."
Maybe this will prove true but I think it is a bad assumption regarding AI.
I'd bet humans have been mostly removed from the term life process other than visits to collect blood and some health mesurements. Likewise I'd bet more complicated life insurance never moves to AI.
In my space AI is mostly used to help doctors upcode patients and increase Medicare revenue.
There is also an arms race with AI re-submitting claims and other AI denying claims. Like your example of lawyers creating work for each other, I could easily see competing AI creating work for each other.
There is a lot of dumb Red Queen Race Bullshit out there that people forego because it would be "too expensive" to do. If you lower the price, they can increase the quantity. This is more or less the thesis behind why we don't have Keynes 15 hour workweek.
In a longer reply to John Hall I note that if olds and poors prefer humans CMS has a lot of ways to incent using humans at the taxpayers expense. CMS would much rather have sales brokers fielding member complaints then having to do it themselves.
I'm quite sure there's an arms race between LLMs writing resumes and cover letters and LLMs screening them.
In the end, it's kind of nice: I tell the LLM my experience bullet points, an LLM encodes that as flowery corporate speak, and the receiving LLM decodes it back to bullet points. Now, a sane human would observe we could save a lot of time and money by just skipping the prose stage but there we are.
However much creative destruction went on in the '30's there was no reason for the Fed to allow
so many markets not to clear.
My experience is blue collar on the delivery end(USPS). I can see a robotaxi type vehicle running the routes as well or better than I do in the near future or even today but I am having problems visualizing the actual to the doorstep delivery of packages. This is going to require many robots covering all sorts of terrain in the snow and the rain to place packages(opening doors) in porches. I could care less about a million dollar lab robot dancing around some lab. A 5 to 7 year time limit means you need a factoring somewhere turning out a thousand of these robots a week at an affordable price to dent a million plus person workforce. Same with trucking, I can see the truck going from point a to b but what if only half the product is being unloaded at point b. Do you just let the customer climb on in with his pallet jack and hope that he knows he is only supposed to be taking 10 pallets before the truck pulls away.. A human even a non driving human would really be helpful in this situation.
you can have robots or you can have open borders
Inroads by AI and robotics will naturally be faster in some industries, slower in others. One factor is resistance. The battle against AI / robotics will play out, e.g., Trump’s support for union dockworkers to resist port automation. Union industries will be unanimously opposed to automation of any sort. For fun, I asked Claude to rank each union industry for resistance to automation, whether it be AI or robotics, differentiating between the U.S. and E.U. Answer -
Ranking Union Industries by Resistance to Automation (from Most to Least Resistant):
United States:
1. Construction (high skill, complex physical tasks)
2. Public Sector (bureaucratic complexity)
3. Transportation (specialized skills)
4. Manufacturing (moderate automation vulnerability)
5. Education (AI-resistant, but facing technological pressures)
6. Postal Services (highly automatable)
European Union:
1. Manufacturing (strong worker protections)
2. Energy/Utilities (complex technical skills)
3. Automotive Industry (specialized technical labor)
4. Metalworking (precision skills)
5. Public Transportation (moderate automation risk)
6. Public Sector (administrative roles more vulnerable)
European unions generally have stronger worker protections and more gradual technological transition policies compared to U.S. unions, potentially slowing automation implementation.
Public safety is an "Information" job and I expect technology to greatly reduce and change the work humans do. In the private sector, I expect human security guards and receptionists to be greatly reduced. At the same time, "public safety" in the government realm will probably expand in employment as government will find all sorts of reasons to add humans to manage and oversee technology solutions.
I think human management of technology is going to become a huge career field. And this is the natural progression. Electronics / computers gave us the Information Technology field. Then this became the Information Systems field. And now we will have the Information Management field. I'm polishing my resume now.
Technology Management will explode, I think.
I really like these semi-Sci-Fi speculations, because many will come true and none of them are now false. Tho while intelligence gathering will increase, the huge % increase will be in useful intel analysis. Like an ai looking thru 14,000 J6 videos to get all of those for each of the 1500 or so pardoned protesters. And maybe all the violent & violence advocating guys who were NOT picked up by the FBI.
It would be good to have a Rep ai be watching, weaponized against Dem & govt actions; along with a competing Dem ai watching, weaponized against Reps & govt actions. With monthly transparency releases of suspicious behaviors by govt & related officials.
In 1900 there were approximately 5 million working horses in England propper - pulling plows, carriages, firetrucks, wagons, etc. Fifty years later in 1950 that number had dropped by 99% to approximately 50 thousand. They didn't DISAPPEAR but the VAST majority were replaced by diesel tractors, gasoline engines, and electric motors. The "Great Creative Destruction" affected them FAR more harshly than it did people. They no longer had a comparative advantage in anything that was significant enough to justify continuing to feed them.
I'm not saying it's going to happen, but I CAN forsee a scenario where a lot of people, maybe even the majority, can't do ANYTHING that can't be done better, faster, and more cheaply by an AI.
Do you go to live performance music? Do you go to sporting events? Do you go to theatre? Do you like stand up comedians? Have you been brought to tears by a dramatic performance on stage? Have you been brought to mirth by an author? Tears by an author? Have you been comforted in a time of grief or sorrow by a compassionate live person?
People who can do these things, and other things like them not listed here, will enjoy rising demand and rising incomes. People who are unable to bring the human touch to what they do will suffer falling demand and falling incomes. Beyond these observations, who knows?
Will Ai increase or reduce aggregate employment? I'm going with all of history; aggregate employment will increase, even though the jobs will be for things we cannot even imagine right now, try as we might.
Comparing Dr Kling's predictions to a set of "impact studies" described on the Palantir website (down towards the bottom on the right: https://www.palantir.com/ ) it seems to me like anything that would have benefitted from good operations research in olden days is probably now something that AI can improve on. For example, there are a couple hospital cases listed and AI is claimed to be producing staffing and scheduling and patient queue volume improvements. One wonders whether broad based quantitative operations research efficiencies enabled by AI might be as significant an influence on employment as simple replacement of workers by machines?
The Ferrari impact study, however, seems to provide a perfect, and interesting, example of how AI may wind up improving human-machine interfaces:
"Palantir software integrated several types of Scuderia Ferrari data, such as sensor data from races, tests, simulations and testing equipment, and parts information. The key source was data from car sensors in the form of high frequency time series.
Palantir contextualized time series data within a data model – the Ontology - and organized it into human-readable objects. This framework allowed Scuderia Ferrari engineers to interact with and analyze the data with code, low-code, or no-code tools; and optimize transformations that sped up calculations."
(https://www.palantir.com/impact/ferrari/ )
On the coding front, one wonders if the really big corporate hard coded applications doing stuff like dynamic pricing in real time are already so effective and efficient that the room for AI incremental improvements will be likely to meet cost-benefit tests?
"I don’t think that chatbots will be difference-makers here [agriculture, mining, forestry]. But productivity has been going up for a long time in agriculture, and that will continue, with some of the gains coming from machine learning. Dangerous jobs in mining and oil and gas extraction will increasingly be done by robots. Utilities may find robots useful for installation and repair."
If you're just thinking about LLM-based chatbots, likely true. However, I think you're understating how much AIs might affect things like agriculture. I keep reading articles about how AI-driven autonomous drones are collecting data and might enable very precise chemical applications (e.g. spraying Roundup on just the weeds, not the crops). I have to imaging mining companies are feverishly working on AIs to direct where to dig mines. I don't think this will be a small effect.
“big decline in … major media ..”
https://substack.com/@cynicology/note/c-86957867
No one knows how AI will influence the number, distribution, or the nature of employment in our soon-to-arrive brave new world. A few days ago, however, I read that something like 54% of adults in Philadelphia are illiterate. If this is true, and I have no reason to doubt it, it won’t be too difficult for employers to adopt AI in creative and mundane ways to achieve greater productivity and better customer service. If our education system doesn’t make major improvements soon, the future will indeed be bleak for many of our fellow citizens.
I can't speak with confidence to the other sectors but with respect to Finance and Insurance there is a lot to agree at it relates to some functions, but I think you downplay the impact of human involvement.
US employment in the finance and insurance (so finance ex real estate and renting) is about 6.7mn. Of that, about 3mn is insurance, and another 1.3mn is commercial banking. I'm fairly sure that the technology has existed to have "AI" handle most of the primary responsibilities here (for instance: offering insurance quotes, a chatbot that can suggest what insurance people will buy, deciding whether to approve loans or not, etc.). And yet people are still hanging around. Why? Well I'm sure there are plenty of jobs tied to regulations in this space (for instance, humans need to review the models). But if chatbots are so good, then probably the place where they would save the companies the most money is replacing salespeople (think robo-advisors on steroids). Then, the companies could avoid paying pesky commissions. So I think that would be one place to look for success of this thesis. If companies can get away with paying lower commissions to salespeople (because their chatbots are so good), then people would quit those jobs (or fewer people take them). But I think there are a lot of times when you really want to deal with a human. And the more money is at stake, the more you are willing to pay for the privilege.
I don't disagree that there will be some jobs that are under pressure, but I think it's more of a transition (starting with the ones discussed in the article) than a shock. At the end of the day, we can look at the number of people employed in these industries five years from now to really evaluate this (it may be that we lose some of these jobs and gain others, not like that hasn't happened before).
The big money in insurance is Medicare/Medicaid. These customers are the least likely to be able to interact well with a non-human salesperson or tech support.
In addition, CMS heavily weights customer satisfaction surveys in STARS payments. If you save a few pennies on a chatbot call center versus human beings, you are probably going to give it back in lower STARS payments when they give you a bad CAHPS score for the experience.
There are also sales venues that would only work for human beings.
Finally, everything a chatbot does is going to be strenuously recorded. That means people can sift through it to see if it violated (even unintentionally) one of the millions of potential regulations in the space. A human being speaking to another human being isn't recorded, and so there is less on the record to scrutinize.
For the millionth time, "This is a blog for generalists." It is not a blog for people who are insiders in Medicare/Medicaid finance. Most people won't know what STARS or CAHPS are, or perhaps even what CMS is. So spell out what the acronyms mean or explain them.
Some people will think, "he knows what those acronyms mean; he must really know his stuff." But some people will think, "what an inconsiderate asshole."
CMS = Center for Medicare and Medicaid Services
STARS = A very large bonus pool paid to Medicare Advantage carriers that hit certain "quality" metrics.
CAHPS = A survey sent to Medicare Advantage members asking them various questions that amount to "are you happy with your health insurer". Has a large impact on STARS.
Thanks.
"If you save a few pennies on a chatbot call center versus human beings, you are probably going to give it back in lower STARS payments when they give you a bad CAHPS score for the experience."
Maybe this will prove true but I think it is a bad assumption regarding AI.
They record conversations in parts of the securities industry, so I wouldn't discount that as a future possibility.
On another note, I was able to find a breakdown of jobs within the insurance industry here:
https://data.bls.gov/projections/nationalMatrix?queryParams=524000&ioType=i&_csrf=projections
But unfortunately was not able to find the breakdown between life/property insurance jobs and health insurance jobs.
I'd bet humans have been mostly removed from the term life process other than visits to collect blood and some health mesurements. Likewise I'd bet more complicated life insurance never moves to AI.