And how will it achieve this? By eliminating jobs of course.
McKinsey’s study found that generative AI and other technologies could automate work activities that currently take up 60 to 70 percent of employees’ time.
Who knew that 60 to 70 percent of employee's time was mostly just BS that could be easily automated by a BS generator?
> And how will it achieve this? By eliminating jobs of course.
For now, probably only white collar workers. It’s probably good if those jobs go away.
Then, we need to think how to compensate those people. Capital is not evenly distributed so no compensation will just drive profits and this will not end up in the pockets of most people.
I'd say the employer has to keep paying salaries for some time to let people acquire new skills.
Compensate white collar workers? I don't think so. They get the same tough love that blue collar workers have been getting for over a century with each machine innovation that upended their lives.
I don't think it's reasonable to say that just because something can be automated via AI that it's BS. I've used AI to write some scripts that automate a bunch of tasks that involve pulling data from different sources and putting them into a Google Sheet for reporting purposes for an ecommerce business I own. Is reporting on the financial reporting of my business BS?
Also, this study is about future gains coming from AI, so even if you take the very cynical position that LLMs/generative AI are nothing but "BS generators" today, unless you have good reason to believe that they will not improve in quality, that's not really a good basis on which to judge their future applications.
unless you have good reason to believe that they will not improve in quality
I do.
These LLMs can't tell facts from BS now. With LLMs starting to generate ever more BS, going forward it will only become harder to tell the difference --- for both them and us.
What do you think will happen when LLMs start sucking in each others BS from the internet?
I don't think it's possible for this sort of global BS feedback loop to produce an improvement in quality.
I think this is severely overstated as a problem that would prevent LLMs from being useful. LLMs generate BS if you use them like a search engine, but there are lots of use cases where they don't.
Need to summarize a document or rephrase something in a different tone? Hallucination is really a non-issue.
I've used LLMs to write scripts - while hallucination does happen, it's a situation where I can easily validate that the scripts are doing what I would expect, so hallucination isn't a serious issue.
Exactly. There are many, many cases where AI significantly speeds up doing something.
I've already used it productively for writing code, understanding how to set up new tech, summarising documents and academic papers, and writing clearer text. And I'm not a power user at all.
So this will result in fewer existing jobs, and the productivity of the remaining jobs must increase. And some new jobs will be created along the way.
Once new LLMs start to consume hallucinations from older LLMs, the hallucination problem will only start to grow and expand.
With LLMs producing more content and producing it faster (this is their purpose), it will become prohibitively expensive if not impossible to check and verify everything that is fed into these models.
LLMs are multipliers --- garbage in equals orders of magnitude more garbage out for the next generation of LLMs to consume.
> Who knew that 60 to 70 percent of employee's time was mostly just BS that could be easily automated by a BS generator?
I feel like most office
work being BS has been in the zeitgeist since before Dilbert was popular for sure. And part of why The Office works so well as a sitcom environment is we don't expect anyone to do any work there.
> Who knew that 60 to 70 percent of employee's time was mostly just BS
A lot of it is software industry's fault.
Think back to all those office jobs that got eliminated by software in the past 50 years. Like secretaries, in-house art departments, ${whatever the people distributing internal memos around the office were called}, etc. The jobs may be gone, but their tasks didn't disappear. Instead, they were smeared out over the remaining workforce.
How much time a week each of us spends on trying to sync up meetings with different teams, writing reports, preparing slides, managing our mailbox and calendar - or all the other tasks that are not the actual thing we are specialized in? All those things were done by in-house specialists, and like with all specialized jobs, those people were much more efficient at it. Now, they're gone, while we're left doing their jobs on top of our own, and suffering a huge context-switching penalty on top of it.
I'm increasingly convinced that most of the savings software brought to office work is just an accounting trick. The secretaries, in-house graphics teams, etc. all took salaries, which were clearly visible on company balance sheets. In comes the software, out these people go, their responsibilities get distributed between all the remaining employees - suddenly, the salaries for those specialists disappear from the balance sheets, while the company... it feels like it experiences an unexplained, broad productivity drop. A great mystery. Is it because Feds are messing with interest rates? Is it because some "cost diseases"? Who knows? Doesn't matter - it looks like a global thing, so it doesn't show on the balance sheets.
If Large Language Models are going to automate away all those BS tasks we've been doing, what's really happening is that they're finally materializing the gains that regular office software promised to deliver, while it in fact made things worse.
(Exercise for the reader: based on the above, think about websites and apps enabling "self-service", and consider if, as individuals, we're really gaining much by being made to do for free the work that used to be someone else's job.)
Fundamentally I don’t understand how a new technology can claim to make everything more efficient but at the same time grow the economy. Shouldn’t we have loss of jobs and reduction in consumption and GDP if we become way more efficient?
Because a lot of jobs are ridiculously inefficient. The efficiency gains make things (and money) move faster. Faster movement of goods, services, and money makes GDP grow.
If you can get done twice as much because you have a robot assistant, then, your basically producing twice the amount of goods or services. So, people can consume more. To be able to produce or consume more the abstractly growth GDP.
I guess an important assumption here is that people get payed according to productivity. Otherwise this would go into the profit of the company. But still the increase in wealth would end up somewhere.
The funny part to me is that they assume this journey to uber-efficient utilization of labour will be super... efficient? How many decades will it take us to unlock the potential of all this displaced BS work?
They come up with a hypothesis like "Every hour of labor shaved off a job by AI gets converted to an hour at the highest value measure". Then they produce very complicated spreadsheets that show how many hours will be saved and multiply this by a very big per unit number. Then they report on the results with very little scrutiny of the methodology. The huge dollar $ign$ blind executive thought leaders, who will rush to adopt - at least for the next 6-9 months.
I think we can't assume the reduction in consumption. After all productivity increase is not an end itelf but a means to sell more or earn more producing stuff cheaper/faster. I expect that with increased productivity consumption will increase because people will want more and can afford more. As a result they will buy more, also due to lowering quality/durability. Low prices and efficient production lead to dispensable everything: clothes, devicesc car etc.
Real GDP is what matters in the end, not nominal GDP. Real GDP can grow via deflation, even if nominal GDP declines.
AI -> lower cost of goods -> real GDP rises
AI -> unemployment -> lower salaries -> lower nominal GDP -> workers reposition into new jobs
Somehow the USA has the lowest unemployment ever after the industrial revolution destroyed millions of jobs... how did that happen?
It's obvious from history that as real GDP per capita rises, new professions emerge or become more accessible such that displaced workers can find new jobs. Perhaps higher real GDP enables more creative professions, or more consumption of housekeeping/helper type roles.
* McKinsey told AT&T that cellphones would be a niche market in the year 2000 with only 900,000 subscribers. The firm’s estimations were off by 108 million.
* In the 1990s McKinsey advised SwissAir on the controversial “hunter strategy”. The major expansion program failed miserably and the airline was forced to declare bankruptcy in 2001. Other notable clients who ended up bankrupt include Kmart and General Motors.
* McKinsey & Company gave the go-ahead for the infamous $350 billion merger between Time Warner (TWX) and AOL. Looking back, the merger is viewed as one of the greatest company disasters of all time.
McKinsey and Jeff Skilling used Enron as a "sandbox". [2]
McKinsey is somewhat a consult cult that ruined agility with "Agile" which goes against everything agile was supposed to be. [3] Real agile and agility is now crushed due to them turning it into a micromanagement always on critical path. [4]
McKinsey margin cutting has killed off lots of research and development and innovation.
These studies aren't done wholistically and only done from a business perspective. That's why these studies are garbage they fail to account for the feedback effect.
Employers pay money to workers so that workers in turn can buy products from employers.
These studies only account for how much money Employers save when they get rid of workers and pay less money to workers.
Well when you get rid of workers or pay them less there's less money going around for workers to buy things from employers. These studies fail to account for this feedback.
It's the tragedy of the commons. If one company utilizes generative AI to reduce labor costs that company benefits. If all companies collectively do this, then everyone loses. It's the aggregate behavior of all companies acting in their own interest that ultimately causes them all to act against their own interests.
How is this any different from farm machinery 200 years ago or industrial automation 50 years ago? More production with far less people.
The same concerns that tithes would lead to massive unemployment. Instead, the economy and global living standards shot up and completely new fields opened up.
>How is this any different from farm machinery 200 years ago or industrial automation 50 years ago? More production with far less people.
The economy will for sure eventually get into some sort of equilibrium again. Things will become normal sort of like how the giant wealth inequality gap is super normal right now (likely caused by the same automation you're describing).
But there is so much different now then before. Additionally the velocity in which this replacement is occuring is much higher then industrial automation. Given the differences we cannot fully know the outcome.
It's easy and convenient to allude to examples in the past to predict the future, but that is not a data driven or logical conclusion. We don't know what will happen, and to ignore the possibility of a bad outcome is folly.
My wild guess is that there will be a temporary period of destabilization and this temporary period could last between one to two decades all the way to several generations. By then all the "problems" will be normalized; sort of like how the wealth inequality gap has been normalized and how it's basically become normal to see tons and tons of homeless people living in RVs in the bay area.
> How is this any different from farm machinery 200 years ago or industrial automation 50 years ago? More production with far less people.
The skill gap for in-demand labor was less back then. For instance, when farm machinery took off, unskilled farm labor could shift (at massive scale) into unskilled factory work.
You're not going to have a massive shift of low-end or midrange labor into high-end ML jobs. A lot (most?) people are just plain not capable of that. These technologies just kick a bunch of people down then pull of the ladder. Poverty, precarity, and inequality will increase. It'll be great for the ultra-wealthy, who will be able to keep more money (power) in their own pockets without sharing with the plebs.
But who knows, maybe that concentration of elite power will open up promising opportunities in the entertainment industry for the plebs to play squid games.
> You're not going to have a massive shift of low-end or midrange labor into high-end ML jobs. A lot (most?) people are just plain not capable of that.
People focus too much on job types. The reality it's actually worse. Even if eliminating one mid-range job[0] would create two more same-level jobs of different type - that is, the total number of jobs available would double - and even if the people automated away from the former were fully capable of retraining for the latter, they'll still be in a world of hurt, because this still means their entire career progression suddenly got reset.
In simple terms: your average Jane and Joe, 15-20 years in their mid-skill career accumulated some skills, experience and promotions, which allows them to get a mid-level salary. They built their life around it - bought a flat or a house sized right for their income, in the area sized right for their income. They started a family, and are caring and educating their kids in a way appropriate for their income. Suddenly, their entire occupation disappears, and they're forced to retrain. They manage to do that, and find new jobs in the new field. Guess what level those jobs are, and how much they pay? That's right, they're starting at junior level, with junior pay. Suddenly, their entire life is way too expensive for their income levels. The house, the area, the schools, the car - and by proxy, their social life, their kids' education - all of these need to be cut down. What didn't change, however, is their age and associated health problems.
As for the kids, they too are unlikely to benefit from the newly-opened fields, because they'll be too busy working their way out of poverty.
--
[0] - And note that unlike some earlier techology-induced job shifts, AI is threatening to displace the high-skill jobs first. Generative models won't displace your barber or the local handyman or the policemen on patrol. They are going to displace artists, clerks, possibly medical and legal techs, testers. They'll sooner displace programmers and lawyers and doctors before they'll be able to impact blue-collar jobs.
This is a nice story that omits one crucial detail of every such transition: the people who get automated away and the people who benefit from improved economy and standards of living are not the same people.
For the former, this is not a story of temporary hardships compensated in full by everything getting better few years down the line. For them, this is a story of life permanently derailed, their hopes and dreams destroyed, and the future of their kids taken away. They do not get to enjoy those promised benefits - they're too busy trying to salvage what little they can from their lives, after being suddenly dropped one or two levels down in socioeconomic class ladder.
The benefits? Sure, they come some years or decades later, and they're great for everyone else.
VRChat is similar to Metaverse and while it hasnt exploded in popularity, its growing.
I could see it being a multi-trillion dollar thing. My retired dad is obsessed, he spends hours per day in it meeting new people and his friends from around the world. He even has become more open minded to Trans people(He has been socially liberal his whole life, but that topic he was resistant on).
He tells me about parties at clubs, celebrating holiday's like St. Patrick's day, having birthday parties, even a wedding.
Soon companies will begin monetizing it more blatantly and capitalism will feed it.
Facebook failed because Facebook is creepy and we try to avoid their products.
Facebook didn't fail because it's creepy. It's product just wasn't good enough to overcome its creepiness cost. (although, going by Facebook's stock price I don't think it has failed...?)
TikTok is much creepier and more invasive, but people keep flocking to it.
Facebook released, hands down, the best video conferencing hardware and experience I have ever used. It was at an aggressive price point, and was well built hardware.
The first thing people said when they heard about it was "I don't want Mark Zuckerberg watching me". That's not rational or correct, but that's the sort of problem Facebook has made for itself.
>>"I don't want Mark Zuckerberg watching me". That's not rational or correct
I think there's a non-zero chance if it took off they would measure anonymised metrics. e.g. We noticed 1 user stand up, stop watching TV 4 times to go to the kitchen. We now have websites that record every mouse move to allow replay. We have had phone apps listening in to hear what TV programs are being played currently to gather viewing stats. If there's money to be made it's happened. The webcam gathering anonymised statistics seems likely.
Well, Portal had an E2EE implementation that was well tested and reviewed, excellent hardware security (including immutable indicators in the form of LEDs of camera activity). It was built exactly like you'd build something to be immune from this criticism.
But since Meta was/is a privacy dumpster fire, none of that mattered.
Since Meta is a privacy fire, being concerned that a product of theirs either is—or will be—used to violate users privacy seems a perfectly rational concern to me.
facebook failed is a hyperbole, as is the belief that VRchat will scale to billions. It is not the first time these things happened, Secondlife has a few million users peak.
As many people here, I program. My performance directly reduces the cost of company operations.
Previously I was responsible for $500k/yr in savings, but with ChatGPT aiding my coding, I can easily see that number 1.5x-ing. Not to mention, we could use local models for multi-millions in savings(we don't trust putting our data online).
I think as people learn when to use LLMs, it will cause a multi-trillion dollar explosion in cost reductions.
Does it increase real wealth at all though or is this like how I'm now $100,000 richer because my house appreciated, even though it makes 0% actual impact on my situation?
59 comments
[ 3.4 ms ] story [ 110 ms ] threadFor now, probably only white collar workers. It’s probably good if those jobs go away.
Then, we need to think how to compensate those people. Capital is not evenly distributed so no compensation will just drive profits and this will not end up in the pockets of most people.
I'd say the employer has to keep paying salaries for some time to let people acquire new skills.
Learn to do plumbing.
No. There’s a union to prevent that thankfully. Don’t make this the problem of plumbers or other high paid tradespeople.
Become a cop. They get high pay and they need more people desperately.
E.g. for people in academia i am sure it has removed a lot of 'bs job time' already
I think that people miss that headcount will just stop increasing first, everyone is looking for mass layoffs and miss that other component.
I don't have a problem with that. Just find it interesting that people who would have a problem with it aren't considering it.
Also, this study is about future gains coming from AI, so even if you take the very cynical position that LLMs/generative AI are nothing but "BS generators" today, unless you have good reason to believe that they will not improve in quality, that's not really a good basis on which to judge their future applications.
I do.
These LLMs can't tell facts from BS now. With LLMs starting to generate ever more BS, going forward it will only become harder to tell the difference --- for both them and us.
What do you think will happen when LLMs start sucking in each others BS from the internet?
I don't think it's possible for this sort of global BS feedback loop to produce an improvement in quality.
I think this is severely overstated as a problem that would prevent LLMs from being useful. LLMs generate BS if you use them like a search engine, but there are lots of use cases where they don't.
Need to summarize a document or rephrase something in a different tone? Hallucination is really a non-issue.
I've used LLMs to write scripts - while hallucination does happen, it's a situation where I can easily validate that the scripts are doing what I would expect, so hallucination isn't a serious issue.
I've already used it productively for writing code, understanding how to set up new tech, summarising documents and academic papers, and writing clearer text. And I'm not a power user at all.
So this will result in fewer existing jobs, and the productivity of the remaining jobs must increase. And some new jobs will be created along the way.
Once new LLMs start to consume hallucinations from older LLMs, the hallucination problem will only start to grow and expand.
With LLMs producing more content and producing it faster (this is their purpose), it will become prohibitively expensive if not impossible to check and verify everything that is fed into these models.
LLMs are multipliers --- garbage in equals orders of magnitude more garbage out for the next generation of LLMs to consume.
I feel like most office work being BS has been in the zeitgeist since before Dilbert was popular for sure. And part of why The Office works so well as a sitcom environment is we don't expect anyone to do any work there.
A lot of it is software industry's fault.
Think back to all those office jobs that got eliminated by software in the past 50 years. Like secretaries, in-house art departments, ${whatever the people distributing internal memos around the office were called}, etc. The jobs may be gone, but their tasks didn't disappear. Instead, they were smeared out over the remaining workforce.
How much time a week each of us spends on trying to sync up meetings with different teams, writing reports, preparing slides, managing our mailbox and calendar - or all the other tasks that are not the actual thing we are specialized in? All those things were done by in-house specialists, and like with all specialized jobs, those people were much more efficient at it. Now, they're gone, while we're left doing their jobs on top of our own, and suffering a huge context-switching penalty on top of it.
I'm increasingly convinced that most of the savings software brought to office work is just an accounting trick. The secretaries, in-house graphics teams, etc. all took salaries, which were clearly visible on company balance sheets. In comes the software, out these people go, their responsibilities get distributed between all the remaining employees - suddenly, the salaries for those specialists disappear from the balance sheets, while the company... it feels like it experiences an unexplained, broad productivity drop. A great mystery. Is it because Feds are messing with interest rates? Is it because some "cost diseases"? Who knows? Doesn't matter - it looks like a global thing, so it doesn't show on the balance sheets.
If Large Language Models are going to automate away all those BS tasks we've been doing, what's really happening is that they're finally materializing the gains that regular office software promised to deliver, while it in fact made things worse.
(Exercise for the reader: based on the above, think about websites and apps enabling "self-service", and consider if, as individuals, we're really gaining much by being made to do for free the work that used to be someone else's job.)
I guess an important assumption here is that people get payed according to productivity. Otherwise this would go into the profit of the company. But still the increase in wealth would end up somewhere.
source: former management consultant
AI -> lower cost of goods -> real GDP rises
AI -> unemployment -> lower salaries -> lower nominal GDP -> workers reposition into new jobs
Somehow the USA has the lowest unemployment ever after the industrial revolution destroyed millions of jobs... how did that happen?
It's obvious from history that as real GDP per capita rises, new professions emerge or become more accessible such that displaced workers can find new jobs. Perhaps higher real GDP enables more creative professions, or more consumption of housekeeping/helper type roles.
* In the 1990s McKinsey advised SwissAir on the controversial “hunter strategy”. The major expansion program failed miserably and the airline was forced to declare bankruptcy in 2001. Other notable clients who ended up bankrupt include Kmart and General Motors.
* McKinsey & Company gave the go-ahead for the infamous $350 billion merger between Time Warner (TWX) and AOL. Looking back, the merger is viewed as one of the greatest company disasters of all time.
https://www.equities.com/news/a-look-at-mckinsey-company-s-b...
Not that I want this to prevent them from whipping up some serious executive FOMO over AI. Ive got bills to pay.
McKinsey and Jeff Skilling used Enron as a "sandbox". [2]
McKinsey is somewhat a consult cult that ruined agility with "Agile" which goes against everything agile was supposed to be. [3] Real agile and agility is now crushed due to them turning it into a micromanagement always on critical path. [4]
McKinsey margin cutting has killed off lots of research and development and innovation.
[1] https://en.wikipedia.org/wiki/McKinsey_%26_Company#Controver...
[2] https://en.wikipedia.org/wiki/McKinsey_%26_Company#Enron
[3] http://agilemanifesto.org/
[4] https://www.youtube.com/watch?v=a-BOSpxYJ9M
Employers pay money to workers so that workers in turn can buy products from employers.
These studies only account for how much money Employers save when they get rid of workers and pay less money to workers.
Well when you get rid of workers or pay them less there's less money going around for workers to buy things from employers. These studies fail to account for this feedback.
It's the tragedy of the commons. If one company utilizes generative AI to reduce labor costs that company benefits. If all companies collectively do this, then everyone loses. It's the aggregate behavior of all companies acting in their own interest that ultimately causes them all to act against their own interests.
The same concerns that tithes would lead to massive unemployment. Instead, the economy and global living standards shot up and completely new fields opened up.
The economy will for sure eventually get into some sort of equilibrium again. Things will become normal sort of like how the giant wealth inequality gap is super normal right now (likely caused by the same automation you're describing).
But there is so much different now then before. Additionally the velocity in which this replacement is occuring is much higher then industrial automation. Given the differences we cannot fully know the outcome.
It's easy and convenient to allude to examples in the past to predict the future, but that is not a data driven or logical conclusion. We don't know what will happen, and to ignore the possibility of a bad outcome is folly.
My wild guess is that there will be a temporary period of destabilization and this temporary period could last between one to two decades all the way to several generations. By then all the "problems" will be normalized; sort of like how the wealth inequality gap has been normalized and how it's basically become normal to see tons and tons of homeless people living in RVs in the bay area.
The skill gap for in-demand labor was less back then. For instance, when farm machinery took off, unskilled farm labor could shift (at massive scale) into unskilled factory work.
You're not going to have a massive shift of low-end or midrange labor into high-end ML jobs. A lot (most?) people are just plain not capable of that. These technologies just kick a bunch of people down then pull of the ladder. Poverty, precarity, and inequality will increase. It'll be great for the ultra-wealthy, who will be able to keep more money (power) in their own pockets without sharing with the plebs.
But who knows, maybe that concentration of elite power will open up promising opportunities in the entertainment industry for the plebs to play squid games.
People focus too much on job types. The reality it's actually worse. Even if eliminating one mid-range job[0] would create two more same-level jobs of different type - that is, the total number of jobs available would double - and even if the people automated away from the former were fully capable of retraining for the latter, they'll still be in a world of hurt, because this still means their entire career progression suddenly got reset.
In simple terms: your average Jane and Joe, 15-20 years in their mid-skill career accumulated some skills, experience and promotions, which allows them to get a mid-level salary. They built their life around it - bought a flat or a house sized right for their income, in the area sized right for their income. They started a family, and are caring and educating their kids in a way appropriate for their income. Suddenly, their entire occupation disappears, and they're forced to retrain. They manage to do that, and find new jobs in the new field. Guess what level those jobs are, and how much they pay? That's right, they're starting at junior level, with junior pay. Suddenly, their entire life is way too expensive for their income levels. The house, the area, the schools, the car - and by proxy, their social life, their kids' education - all of these need to be cut down. What didn't change, however, is their age and associated health problems.
As for the kids, they too are unlikely to benefit from the newly-opened fields, because they'll be too busy working their way out of poverty.
--
[0] - And note that unlike some earlier techology-induced job shifts, AI is threatening to displace the high-skill jobs first. Generative models won't displace your barber or the local handyman or the policemen on patrol. They are going to displace artists, clerks, possibly medical and legal techs, testers. They'll sooner displace programmers and lawyers and doctors before they'll be able to impact blue-collar jobs.
For the former, this is not a story of temporary hardships compensated in full by everything getting better few years down the line. For them, this is a story of life permanently derailed, their hopes and dreams destroyed, and the future of their kids taken away. They do not get to enjoy those promised benefits - they're too busy trying to salvage what little they can from their lives, after being suddenly dropped one or two levels down in socioeconomic class ladder.
The benefits? Sure, they come some years or decades later, and they're great for everyone else.
[1] https://www.globenewswire.com/en/news-release/2022/06/15/246...
I could see it being a multi-trillion dollar thing. My retired dad is obsessed, he spends hours per day in it meeting new people and his friends from around the world. He even has become more open minded to Trans people(He has been socially liberal his whole life, but that topic he was resistant on).
He tells me about parties at clubs, celebrating holiday's like St. Patrick's day, having birthday parties, even a wedding.
Soon companies will begin monetizing it more blatantly and capitalism will feed it.
Facebook failed because Facebook is creepy and we try to avoid their products.
TikTok is much creepier and more invasive, but people keep flocking to it.
The first thing people said when they heard about it was "I don't want Mark Zuckerberg watching me". That's not rational or correct, but that's the sort of problem Facebook has made for itself.
What's not rational about it?
But since Meta was/is a privacy dumpster fire, none of that mattered.
https://www.upf.edu/web/angel-lozano/innovation/-/asset_publ....
Previously I was responsible for $500k/yr in savings, but with ChatGPT aiding my coding, I can easily see that number 1.5x-ing. Not to mention, we could use local models for multi-millions in savings(we don't trust putting our data online).
I think as people learn when to use LLMs, it will cause a multi-trillion dollar explosion in cost reductions.
This IS the greatest threat of "AI"