Without speculating here on how likely this is to happen, suppose that GPT-4 (or some other LLM or AI) speeds up, streamlines or improves quite a lot of things.
> We need to move away from this attitude that interacting with other human beings is something you should just bring a note from your doctor and get excused from.
Tiny? Most people on the spectrum (IDK what could you possibly mean by it being meaningless phrase; it's not some nebulous concept) are not functional / cognitively disabled.
The amount of functional ones is already tiny, and the amount benefiting from being forced into in-person interactions at work is almost certainly negligible. Something like 90% aspies are unemployed. That's what obsession with face-to-face communication does to people who _don't communicate reliably through body language._
You nailed it! The powers that be will definitely keep us employed, as "work" is too useful as a means of social control. The nightmare I see coming is the movie Office Space, but instead of a flesh-and-blood Bill Lumbergh, you got ChatGPT telling you, in some robotic monotone, "And if you could go ahead and come in on Sunday too, that'd be great."
"The powers that be will definitely keep us employed, as "work" is too useful as a means of social control."
Can anyone explain to me how this works in real-life examples? It's one of the big problems that I have with the bullshit jobs thesis. Every historical example I can think of works in exactly the opposite direction. Serfdom, chattel slavery in the Americas, Gilded Age factory work, South African apartheid; they are all systems in which the powers that be used social control to make sure there was a cheap and compliant source of labor.
What's an example of elites using labor to keep social control? Conceptually, it doesn't make any sense to me. Why would elites even care about social control if not for the purpose of controlling some input into the production process: land, labor, natural resources, whatever?
Social control in the sense of not wanting lots of unemployed and restless youths. Having a system where long term and steady work is required in order to "live a good life" implies control - you have to act right and follow the rules in order to keep a job, which is itself necessary in order to have enough food and other necessities. Those making this argument here I believe are also making an argument for an alternative where the productivity of society is more equally spread, without the need to make everyone work for it. I agree that it's a system of social control, but I don't think it's nefarious or bad. We really don't want to live in a society where 25-year-old men don't have meaningful work to do and roam the streets getting into trouble.
The other side of the theory is that all of those previous systems eventually failed when the workers got fed up with the severity and type of social control. In the early 20th century, almost every developed country in the world had significant social upheaval, and several had socialist or communist revolutions (many others felt like they were close to it). By implementing social safety nets, this tendency was blunted significantly (to the chagrin of dedicated leftists all over).
> I agree that it's a system of social control, but I don't think it's nefarious or bad. We really don't want to live in a society where 25-year-old men don't have meaningful work to do and roam the streets getting into trouble.
This opinion is frankly terrifying. You're essentially arguing that people need to be imprisoned, otherwise they will... what? Start killing others? Why?
Zvi writes: "ChatGPT lets students skip make-work. System responds by modifying conditions to force students to return to make-work." The underlying assumption seems to be "any school work that can be done by ChatGPT is make-work." This strikes me as extreme.
The fact that you can train a computer to do X does not imply that now humans can do X without any training. More importantly, it very much doesn't imply that humans can learn to do (Y that has, until now, been strongly correlated with X) without training in X, because there isn't a process for training for (Y but not for X). This probably means there's fertile ground for research there -- to use a subject I understand better than writing, the analogy would be "how do you train general numeracy without spending so much time on the sorts of arithmetic, algebra, and calculus that a calculator / WolframAlpha can do," -- but until results exist there, the claim "computers can do X, therefore making students to X by hand is make-work" does not follow.
(This does not apply to actual bullshit jobs, whose stated goal is the execution of the task at hand rather than any changes wrought in the mind of the worker.)
Ah, I wasn't attempting to make a claim quite that strong, although I do think that is strong evidence for a given assignment and a large % of assignments are make-work to start with. It was more 'If X is make-work and GPT then allows it to be skipped, the school will move to stop this.'
There's a nightmare I don't think you've quite articulated - the job of developing bullshit rules for others to follow is itself a bullshit job that can be automated by GPT-X.
It's obvious now - the hypothetical AI has turned all matter in the universe into paperclips because another AI has turned all energy in the universe into paperwork.
It’s possible that bullshit employers can retain a lot of the desired bullshit function of their employees by not acknowledging the reality that employees can use GPT-X to do most of their work.
I typed this comment while in a remote meeting surrounded by other people in the same meeting, all of whom were on their phones or doing something else.
1) Homework might be 'make-work' for you because you're really smart. For a kid who isn't as smart, it might (MIGHT) actually help them master the concepts they're supposed to learn. Using ChatGPT to produce the homework defeats that purpose, and so the child is worse off using ChatGPT than just mastering the material.
2) In my experience, there aren't a lot of jobs that are completely bullshit. There's almost always some element of the job that is difficult to automate (typically something involving coordination). If AI can provide automated coordination, most companies will transfer the people out of those jobs, or fire them. That's an economic win, IMO.
I used to work at a company that had a fairly large customer service department. For about 1/4 of the year the department was pretty slow (to the point that they actually had a small library among the cubicles and they would spend their days reading to fill the time). Then about half the year they were pretty busy to about the standard level you would see in most jobs. The other quarter of the year they were extremely busy, working long and stressful days. The jobs were skilled enough that we couldn't lay off employees during the slow times, but the owners would get mad about people literally doing nothing productive, so make work was found for them more often than not.
Also, completely agree about your #1 - I would even go so far as to say that even very smart kids need to practice the skills they use, including writing. You may not need to write 30 examples to get the idea of a five paragraph essay, but you should at least do some yourself!
There is something almost Randian about the concepts "bullshit work" vs what I guess is something like "real producers" -- a little normative kiss if you feel like you are part of the second group (or perhaps not that little).
Yes, people are asked to do all kinds of nonsense, but bullshit does not arrive out of nowhere. A great part of this seems like a structural feature of a capitalist economy (perhaps bound by the Protestant ethic of needing to earn) since without money circulating the whole thing bellies up.
So whatever pressure AI mounts, it mounts on the entirety to my eye (unless a society is very agile or very forward looking).
I think there are different flavors of bullshit jobs, and sometimes it's not clear what is and what isn't.
If I'm a security guard at a factory, but have almost nothing to do but occasionally help a lost visitor find where they're going, or help someone who's locked himself out of his office, that probably feels like a bullshit job. But I might be there as insurance against low-probability disasters, and it may be a globally very sensible thing for me to be there. (The backup generator at the hospital isn't a bullshit precaution just because the local power company is on the ball and you haven't had a power outage in ten years.)
If I'm the same security guard and I'm constantly having to run lowlifes off the property and stop break-ins because the city we're in has a huge problem with homelessness and low-level street crime, my job won't feel like bullshit, but it's possible that in a better-functioning society, my job wouldn't be necessary.
Yeah a *lot* of tasks have the property that someone simply needs to be present all the time due to long-tail events, and if not for some impossibility (separate rampup time for each knowledge set, different locations, or whatever) multiple of these tasks could easily be combined and done by one person. While I'm not particularly a traditionalist or naturist, I do feel some efficiency is lost here in childcare and other home-related tasks in the Western scheme by atomizing them rather than out(in?)sourcing them to the community/village, and particularly to those who most prefer to do that work.
Lots and lots of BS jobs are for long-tail events in other ways too. Quality Control is about as BS as it gets in many organizations, but it's there to prevent real problems (like shipping obviously defective product to customers). Lawyers spend like 95% of their time telling companies not to do things that would probably be fine to do, and then identifying the same three really important things over and over again. A lot of lawyers never go to court (or so rarely as to be the same), but are there in case it's needed as another long-tail event.
One thing that's weird about the whole bullshit jobs thing is that tons of apparently useful but unnecessary customer-facing jobs have been eliminated over the last decade or so, in favor of technology that approximately everyone hates--instead of being able to call the company and get a receiptionist who knows how to put you in touch with the right person, you get a voice mail menu that might eventually connect you to a call center employee with a menu they'll get fired for deviating from. My own employer got rid of our front desk entirely, so now when visitors show up in our main office, they often have to ask around (bugging people with offices close to the front door) to figure out where they're going.
I suspect the same forces that lead to expansion of bullshit jobs also lead to elimination of useful jobs--the feedback to the people making the decisions about hiring/firing/funding is broken in some way. There's a manager who gets to put "cut customer support costs by 50% by outsourcing them to a call center in India" on their performance review, but there's nobody whose performance review is dinged for the corresponding lost customers in any legible way, so useful jobs get eliminated. There's another manager whose budget and importance is determined by the size of his staff, but nobody whose incentivized in a legible way to decrease the fraction of employees whose actual job is to underline someone's importance.
Externalities exist within companies as well as between them. I don't think there is a generic solution to this in all cases, but finding some way to cause people to internalize the cost/benefit of their externalities will make a better organization/world.
Customer service is expensive. The only way I know to deal with it is by staying local.
Buy stuff at local stores, bank with a small local bank. Know your vendors and your customers. This costs more, but there are many hidden benefits, which don't appear on the bottom line. And I'm not sure the extra 'cost' is all real. I mean if more money stays in my neighborhood, doesn't that make me richer too? My neighbors are my customers and will then be able to pay more for my services.
For certain types of goods, absolutely. For other goods, banking seems a good example, George is correct that even "more expensive" can be offset by being happier with the product. My wife and I bank with the same local bank I've used for over 30 years. Even though the actual tellers have changed many times, we still feel valued by people who know us and seem to respect us more than some giant faceless bank. My wife had previously used some bigger name banks in other areas. Between hidden fees and a significant drop in responsiveness, those "cheaper" banks generally weren't even during normal times.
Yeah, I agree it needs to be a choice to buy locally. And if not enough people make that choice then the local stores fail. And we are lessened by the loss, though we do have a bit more money in our pockets.
I think there's a coordination problem here. We all benefit if there is a local shoe store where we can try on shoes, but each of us can benefit individually by trying them on at the brick-and-mortar store, finding what we want, and ordering them online from someone who doesn't pay for the overhead of brick-and-mortar stores where people can try on shoes.
Right, And since I'm one of the few shopping locally, I'm the sucker who is footing the bill. I don't have a solution to this. I mean it's been happening for ages. The loss of the corner deli to the super market, the loss of the quirky hardware store, to the national chains... I could go on and on. The homogeneity of everything makes life much less interesting, but cheaper.
Industry "Best Practices." There is a cottage industry of maturity models and consultants in every specialty that can assess efficiencies. The is a lot of room for slack and bullshit, but if a company is under pressure financially, you can bring in external experts to determine if you are using outdated practices. This is often done very poorly, or to grind some unrelated ax, but also can be useful.
Yes, what are the BS jobs? And having just spent several minutes on the phone, entering all sorts of info to get to a real person. I've now got this view of a dystopian future, where my AI is talking to their AI for those minutes.... customer service becomes even more screwed up. Let's figure out how to fix customer service. And as long as I'm bitching about work, from what I can tell more and more health care workers are doing piece work, independent contractors working at different places weekly. I was talking to a guy last night at a party who's job is to hook the nurses and such up with the hospitals. All so hospitals and other health care services don't have to hire full time workers, with the attendant union/government mandated costs. (Sorry for the rant, unloading from minutes wasted on the phone.)
But what happens when the average AI-customer-support is better than the average-person-customer support? It seems like we aren't too far from that. Not for the complex cases, but for the "average" ones. Even today, automation is generally better for basic inquiries like store hours, current account balances and the like, which genuinely saves everyone's time. They just don't work for real problems ... but GPT seems damn close, for any scenario where the system has prior experience.
And mutual AIs communicating might be a good way to solve the general basic-communication, data entry issue. If your "personal AI" can negotiate all the basic address, transaction-number, member-ID data, re-summerize your problem in the jargon of the downstream provider, and give you clear options - then than might indeed be better world.
AI is only approaching human-level CS because human-level CS has deteriorated so much in the last 20-50 years. The few places that still hire a human who knows and cares about the particulars of your situation are world apart from anything GPT can do (even if it could also pull off the warm fuzzy feeling of talking to a live person, which it can't even try to do).
Fair, enough, but my point is that AI might exceed the current crappy version of human-CS, script driven, non-native speakers, limited discretion, overworked and poor attitude. I don't think it is on-track to exceed "Quality CS" anytime soon. But there are lot of of crappy call centers in the world.
Yeah, I'll agree with that. I'll also agree with George above that CS is screwed up and GPT will not make it any better (while adding in another layer of problems).
I like this post, though I disagree with it quite a lot.
I think many bullshit jobs are soulcrushing because they require the person doing them to spend the majority of the time on the low-value-add tasks, rather than the high-value-add ones. That's because (public or private sector) bureaucracies are, basically, superhuman intelligences composed of ordinary humans, some IT, and a lot of slow-moving process rules. And for many of those bureaucrats writing the rules, "have a human do it," is the default answer because their modal IT procurement fails. They know how to yell at a human to get them to do stuff.
So what I suspect you'll see happening is that a small portion of people in bullshit jobs will liberate themselves by using new tools, but most won't and will stay stuck in the current way of working for far longer than you think. For example, when MSFT ships GPT-5 embedded in Excel, it will transform the lives of those bureaucrats who already experiment with learning new features in Excel, but a shockingly large part of the economy won't even know those features exist until it becomes unavoidable.
I mean, we have people who still run their businesses on _paper ledgers_. That's gonna last for a while, even if there's another part of the economy where we have AIs building us flying cars made of nanodiamond and aerogel.
Agreed. An enthousiastic person can probably dedicate a few hours a week for a month to figuring out how currently-available AI can improve their workflow (if they're in a job where that's even possible). For an individual, it's a low marginal cost with the potential for high returns in productivity and reclaimed time. Overhauling the internal structure and processes of a large organisation, like a federal government department, isn't comparable. Organisational agility isn't a hallmark of civil service.
The YouGove poll in the UK found ~40% self reported bullshit jobs. The real number will be higher or lower depending on whether you think there are more bs-jobs that aren't obviously bs or real jobs that look superficially bs, or visa versa. Also the UK is probably a world leader in bs.
I don’t know of anyone in the business world who takes the Bullshit Jobs hypothesis seriously. There is too much pressure on margins in *all* businesses over the long term and on *most* businesses even over the short term.
Samo’s Law of Bullshit Is nothing more than a corollary of Parkinson’s Law: “Work expands so as to fill the time available for its completion.” At a for-profit organization It is the job of managers to be ever vigilant of the threat posed by Parkinson’s Law.
Bullshit jobs created by government regulation are real to be sure. But that’s nothing more interesting than deadweight loss imposed by legislatures or bureaucrats who are generally terrible at pricing externalities, because of course they are. That’s an unfortunate reality, but to a business regulations are a cost to be managed just like any other cost - that can be done well or badly, but it’s not a source of bullshit jobs, at least not from the standpoint of the bottom line.
I have a colleague whose job is contract management for our entire company. It’s a tedious and thankless job, and like the hypothetical security guard jobs discussed in another comment here, seems like a bullshit job. The thing is those jobs seem marginally important at best, until they aren’t, at which point they can be revealed to be very, very consequential. Risk management is a crucial undertaking for any business that expects to survive forever (most do) and those who toil to make sure a firm’s contracts and insurance and safety and finances and supply chains stay in order are doing very consequential work that is only fully revealed in the fullness of time, when crisis arrives.
I was headed here to say this. Private employers are under a lot of pressure to provide goods and services using fewer employees than they currently are.
"Bullshit jobs" usually mean the employee doesn't feel that they're fulfilling a meaningful function, not that the employer feels that way.
One area where I do agree with Zvi is that if AI makes it cheaper to comply with regulations, we can expect to see more regulations. The theory for this is that an optimistic view of regulations is that government sees the regulation as having a benefit, but the costs of compliance limit government ability to impose regulations right down the the point where the marginal benefit of a regulation is just over zero - at some point, you force your whole banking industry to move to Delaware or Singapore or something.
If an AI allows a small restaurant to comply with ten times the regulations we currently have and still stay in business, then I'd expect to see the pace of new regulations increase.
Every time I read a post like this I get lost in thoughts about all the non-bullshit-absolutely-can't-be-done-by-a-robot-or-AI jobs that are going unfilled. In the last year of our family's life, we have struggled to find:
--someone to paint the house ($20,000 job, only 2 of 15 people even responded, 3 month wait from those who did)
--chimney inspector (2 month wait, $500 for 30 minute job)
--chimney repair (another 2 months, $2500 for 2 hour job)
--vet for dog (first three weren't accepting new clients)
--groomer for dog (first four weren't accepting new clients)
--mammogram (booking four months out)
--tire installer for winter tires (6 week wait)
--propane line inspection to use heater (6 week wait)
--window repair (2 month wait)
--kitchen backsplash tile job (4 month wait)
Why aren't people doing these jobs instead of bullshit jobs!?!?!?
I was going to ask if you live in a city / suburban/ rural. But then in all places, you have to get to know people. And who does what, if you can't do it yourself, handymen are in short supply always. Giving them cookies, or pizza or other food can help, be nice.
What? Why do I have to get to know people to hire them to do their work? Are painters who are too busy to take on a $20,000 really going to be swayed if I offer them $20,000 + cookies? My point was that if handymen are always in short supply (and they are), people should consider being handymen instead of doing bullshit jobs that will be lost to AI. I'm ready to hire them anytime.
Hmm OK maybe it is different in the city/ suburban/ and rural areas. I live in the country, if I need to find a new person to do something for me I'll talk to the people around me and get a few names. Then I'll call up Joe, and tell him I got his name from my neighbor Bill and ask him if he's available to do job X. Joe might say no he's too busy, or that he'll get back to me, or make a date to come out and look the job over. Now Joe will probably ask about this guy George who needs his house painted. What's he like to work for? Does he pay his bills on time? Is he always adding extra work onto the contracted job?
Anyway, my point is that with handymen in short supply, they get to pick and choose their customers. Why would they pick you? So yeah, get to know them, be nice. I have made cookies to give to my auto mechanic. The cookies are a thankyou and an investment in the future, where I have some car emergency, and my auto mechanic will go that extra distance to help me out, 'cause I'm a good customer. (For a long time I worked in a small business, you get to cherish both good vendors and good customers.)
I hate the hearts! Why is that? It's not money. I'm not here to make money.
The hearts are part of why news and bounded distrust is about more than just the ad money involved. (though I agree ad money distorts things more.) Bounded distrust is also about getting hearts. Even if we are paying for the news, the news will still be giving us what we want. Telling us the story we want to hear. And yeah, I want good stories too.
Well and finally, I don't check email so often, and silly 'someone gave you a heart email' are more of a pain than a pleasure. I'm a grumpy old man, 'get off my lawn', give me no hearts.
Turn off the hearts, it's somehow the wrong type of feedback. (Would only negative down arrows be any better? It seems unlikely.)
I actually DID initially ask Substack to turn them off, they said they could do it if I insisted. I ended up not following up. Not sure if that was right or not.
I know hearts seem good, I mean positive feed back. It's good. But now, am I posting to get hearts? Or what I think is the truth. Positive feedback, takes whatever most of your audience agrees with and amplifies it, till it takes over. (We should all build electronic feedback circuits... mostly because then we could have a common language. ) Positive feedback drives the circuit to one power supply rail or the other. (Maybe in the social sphere there are more than two power rails... power rails are the simplest one dimensional analogy.)
It says a lot about the feature that many people just use email filters to direct all like-related emails directly to trash. Comment prioritization shifts - first-mover advantage/Speed Premium increases even further, and of course the default on many Substacks is to only highlight the top two most-liked comments, hide the rest behind a Read More button. (Beware Trivial Inconveniences.)
Honestly I prefer having them off, like over at Scott's...when there's no (easy) way to like a comment, then one must write an actual reply to show favour, and that's conducive to further dialogue and generally higher standards. Pretty sure that's a big reason why ACX has perpetual issues with comment overload, compared to other blogs with similar-sized readerships...so many potential comments are nipped in the bud via the short-circuit of merely liking instead. One often finds additional thoughts to express in the process of writing a comment.
(Liking comments is a Using the One Time action for me. Used to not like posts either, until Scott expressed that that metric has actual meaningful value to him as a writer. So now I feel guilted into liking by default, unless it's a genuinely bad post. Here Be Goodharts...)
Re Scott and his liking hearts. I just want to say when Scott expressed this desire, (That he liked getting hearts on a post.) I thought it was bad. If Scott starts writing to get hearts... Anyway since then I have stopped giving hearts to Scott posts. There needs to be some sort of feedback from his audience. But that can be by the comments. "Hey Scott I loved this post." or with money, signing up for a subscription.
Agreed, it's a terrible motivation. I do still appreciate hearts on Twitter because they make more people see the Tweet in question. Here, I am mostly successful in ignoring them.
There are few actual bullshit jobs, but the ones that do exist emanate either from government regulation or costly social signalling. Both of those burdens are likely to grow to meet the increased capacity for bearing them. Costly social signals that become cheap due to AI cease to be good signals, and people will find new signals that are more costly. Social signaling is a positional good but there's no good way to tax it the way other positional goods should be taxed.
The responses I've seen from other university instructors so far have gone along two paths, both of them pretty expected.
The first is going back to more high-stakes, in-person exams. No tech. Sit and write for two hours, answer the questions, etc. This way of teaching has been really unpopular for quite a while even though (in my opinion) it's a really good way of actually determining student achievement and learning. This should have the effect of reducing the make-work for students at the high-end.
The second path is extreme amounts of scaffolding. I teach in the humanities, which is a lot of writing. One system for teaching is scaffolding assignments so they build on each other and where each new assignment is required to incorporate instructor/peer feedback and correction. ChatGPT might make this easier, but barely, for the moment. This structure should have limited reduction in make-work. It works to teach writing to the median student quite well.
I'm lucky to teach a visual subject, film and media, so my assignments for next semester involve a lot of visual description, shot analysis, formal description. I've also limited film choices that students can write about to things released in the last year.
I also want to chime in to agree with Elena. Much of what you call make-work is make-work for you, but for the median student it's called "practice." People learn to write by being forced to write over and over and over again until they at least begin to approach the limits of their natural ability. Especially if that writing is in an unfamiliar mode (journalism, science, etc.)
Here I suppose the goal could be to get students to the point where their ChatGPT inputs are sufficient to spit out what they actually want to say and that they can parse the output to determine that it's correct.
Not for people on the spectrum.
> We need to move away from this attitude that interacting with other human beings is something you should just bring a note from your doctor and get excused from.
Tiny? Most people on the spectrum (IDK what could you possibly mean by it being meaningless phrase; it's not some nebulous concept) are not functional / cognitively disabled.
The amount of functional ones is already tiny, and the amount benefiting from being forced into in-person interactions at work is almost certainly negligible. Something like 90% aspies are unemployed. That's what obsession with face-to-face communication does to people who _don't communicate reliably through body language._
You nailed it! The powers that be will definitely keep us employed, as "work" is too useful as a means of social control. The nightmare I see coming is the movie Office Space, but instead of a flesh-and-blood Bill Lumbergh, you got ChatGPT telling you, in some robotic monotone, "And if you could go ahead and come in on Sunday too, that'd be great."
"The powers that be will definitely keep us employed, as "work" is too useful as a means of social control."
Can anyone explain to me how this works in real-life examples? It's one of the big problems that I have with the bullshit jobs thesis. Every historical example I can think of works in exactly the opposite direction. Serfdom, chattel slavery in the Americas, Gilded Age factory work, South African apartheid; they are all systems in which the powers that be used social control to make sure there was a cheap and compliant source of labor.
What's an example of elites using labor to keep social control? Conceptually, it doesn't make any sense to me. Why would elites even care about social control if not for the purpose of controlling some input into the production process: land, labor, natural resources, whatever?
Social control in the sense of not wanting lots of unemployed and restless youths. Having a system where long term and steady work is required in order to "live a good life" implies control - you have to act right and follow the rules in order to keep a job, which is itself necessary in order to have enough food and other necessities. Those making this argument here I believe are also making an argument for an alternative where the productivity of society is more equally spread, without the need to make everyone work for it. I agree that it's a system of social control, but I don't think it's nefarious or bad. We really don't want to live in a society where 25-year-old men don't have meaningful work to do and roam the streets getting into trouble.
The other side of the theory is that all of those previous systems eventually failed when the workers got fed up with the severity and type of social control. In the early 20th century, almost every developed country in the world had significant social upheaval, and several had socialist or communist revolutions (many others felt like they were close to it). By implementing social safety nets, this tendency was blunted significantly (to the chagrin of dedicated leftists all over).
> I agree that it's a system of social control, but I don't think it's nefarious or bad. We really don't want to live in a society where 25-year-old men don't have meaningful work to do and roam the streets getting into trouble.
This opinion is frankly terrifying. You're essentially arguing that people need to be imprisoned, otherwise they will... what? Start killing others? Why?
the argument is senseless, that's not how evolution works
Zvi writes: "ChatGPT lets students skip make-work. System responds by modifying conditions to force students to return to make-work." The underlying assumption seems to be "any school work that can be done by ChatGPT is make-work." This strikes me as extreme.
The fact that you can train a computer to do X does not imply that now humans can do X without any training. More importantly, it very much doesn't imply that humans can learn to do (Y that has, until now, been strongly correlated with X) without training in X, because there isn't a process for training for (Y but not for X). This probably means there's fertile ground for research there -- to use a subject I understand better than writing, the analogy would be "how do you train general numeracy without spending so much time on the sorts of arithmetic, algebra, and calculus that a calculator / WolframAlpha can do," -- but until results exist there, the claim "computers can do X, therefore making students to X by hand is make-work" does not follow.
(This does not apply to actual bullshit jobs, whose stated goal is the execution of the task at hand rather than any changes wrought in the mind of the worker.)
Ah, I wasn't attempting to make a claim quite that strong, although I do think that is strong evidence for a given assignment and a large % of assignments are make-work to start with. It was more 'If X is make-work and GPT then allows it to be skipped, the school will move to stop this.'
There's a nightmare I don't think you've quite articulated - the job of developing bullshit rules for others to follow is itself a bullshit job that can be automated by GPT-X.
It's obvious now - the hypothetical AI has turned all matter in the universe into paperclips because another AI has turned all energy in the universe into paperwork.
Thanks, as a punnoisseur that gave me a chuckle.
It’s possible that bullshit employers can retain a lot of the desired bullshit function of their employees by not acknowledging the reality that employees can use GPT-X to do most of their work.
I typed this comment while in a remote meeting surrounded by other people in the same meeting, all of whom were on their phones or doing something else.
I don't really follow your arguments.
1) Homework might be 'make-work' for you because you're really smart. For a kid who isn't as smart, it might (MIGHT) actually help them master the concepts they're supposed to learn. Using ChatGPT to produce the homework defeats that purpose, and so the child is worse off using ChatGPT than just mastering the material.
2) In my experience, there aren't a lot of jobs that are completely bullshit. There's almost always some element of the job that is difficult to automate (typically something involving coordination). If AI can provide automated coordination, most companies will transfer the people out of those jobs, or fire them. That's an economic win, IMO.
I used to work at a company that had a fairly large customer service department. For about 1/4 of the year the department was pretty slow (to the point that they actually had a small library among the cubicles and they would spend their days reading to fill the time). Then about half the year they were pretty busy to about the standard level you would see in most jobs. The other quarter of the year they were extremely busy, working long and stressful days. The jobs were skilled enough that we couldn't lay off employees during the slow times, but the owners would get mad about people literally doing nothing productive, so make work was found for them more often than not.
Also, completely agree about your #1 - I would even go so far as to say that even very smart kids need to practice the skills they use, including writing. You may not need to write 30 examples to get the idea of a five paragraph essay, but you should at least do some yourself!
There is something almost Randian about the concepts "bullshit work" vs what I guess is something like "real producers" -- a little normative kiss if you feel like you are part of the second group (or perhaps not that little).
Yes, people are asked to do all kinds of nonsense, but bullshit does not arrive out of nowhere. A great part of this seems like a structural feature of a capitalist economy (perhaps bound by the Protestant ethic of needing to earn) since without money circulating the whole thing bellies up.
So whatever pressure AI mounts, it mounts on the entirety to my eye (unless a society is very agile or very forward looking).
I think there are different flavors of bullshit jobs, and sometimes it's not clear what is and what isn't.
If I'm a security guard at a factory, but have almost nothing to do but occasionally help a lost visitor find where they're going, or help someone who's locked himself out of his office, that probably feels like a bullshit job. But I might be there as insurance against low-probability disasters, and it may be a globally very sensible thing for me to be there. (The backup generator at the hospital isn't a bullshit precaution just because the local power company is on the ball and you haven't had a power outage in ten years.)
If I'm the same security guard and I'm constantly having to run lowlifes off the property and stop break-ins because the city we're in has a huge problem with homelessness and low-level street crime, my job won't feel like bullshit, but it's possible that in a better-functioning society, my job wouldn't be necessary.
Yeah a *lot* of tasks have the property that someone simply needs to be present all the time due to long-tail events, and if not for some impossibility (separate rampup time for each knowledge set, different locations, or whatever) multiple of these tasks could easily be combined and done by one person. While I'm not particularly a traditionalist or naturist, I do feel some efficiency is lost here in childcare and other home-related tasks in the Western scheme by atomizing them rather than out(in?)sourcing them to the community/village, and particularly to those who most prefer to do that work.
Lots and lots of BS jobs are for long-tail events in other ways too. Quality Control is about as BS as it gets in many organizations, but it's there to prevent real problems (like shipping obviously defective product to customers). Lawyers spend like 95% of their time telling companies not to do things that would probably be fine to do, and then identifying the same three really important things over and over again. A lot of lawyers never go to court (or so rarely as to be the same), but are there in case it's needed as another long-tail event.
Is writing articles about 'bullshit jobs' a 'bullshit job' or not? :-)
One thing that's weird about the whole bullshit jobs thing is that tons of apparently useful but unnecessary customer-facing jobs have been eliminated over the last decade or so, in favor of technology that approximately everyone hates--instead of being able to call the company and get a receiptionist who knows how to put you in touch with the right person, you get a voice mail menu that might eventually connect you to a call center employee with a menu they'll get fired for deviating from. My own employer got rid of our front desk entirely, so now when visitors show up in our main office, they often have to ask around (bugging people with offices close to the front door) to figure out where they're going.
I suspect the same forces that lead to expansion of bullshit jobs also lead to elimination of useful jobs--the feedback to the people making the decisions about hiring/firing/funding is broken in some way. There's a manager who gets to put "cut customer support costs by 50% by outsourcing them to a call center in India" on their performance review, but there's nobody whose performance review is dinged for the corresponding lost customers in any legible way, so useful jobs get eliminated. There's another manager whose budget and importance is determined by the size of his staff, but nobody whose incentivized in a legible way to decrease the fraction of employees whose actual job is to underline someone's importance.
Externalities exist within companies as well as between them. I don't think there is a generic solution to this in all cases, but finding some way to cause people to internalize the cost/benefit of their externalities will make a better organization/world.
Customer service is expensive. The only way I know to deal with it is by staying local.
Buy stuff at local stores, bank with a small local bank. Know your vendors and your customers. This costs more, but there are many hidden benefits, which don't appear on the bottom line. And I'm not sure the extra 'cost' is all real. I mean if more money stays in my neighborhood, doesn't that make me richer too? My neighbors are my customers and will then be able to pay more for my services.
In practice, a lot of your neighbors are liable to go browse at your local shop, find what they want, and then order on Amazon at 70% of the price.
For certain types of goods, absolutely. For other goods, banking seems a good example, George is correct that even "more expensive" can be offset by being happier with the product. My wife and I bank with the same local bank I've used for over 30 years. Even though the actual tellers have changed many times, we still feel valued by people who know us and seem to respect us more than some giant faceless bank. My wife had previously used some bigger name banks in other areas. Between hidden fees and a significant drop in responsiveness, those "cheaper" banks generally weren't even during normal times.
Yeah, I agree it needs to be a choice to buy locally. And if not enough people make that choice then the local stores fail. And we are lessened by the loss, though we do have a bit more money in our pockets.
I think there's a coordination problem here. We all benefit if there is a local shoe store where we can try on shoes, but each of us can benefit individually by trying them on at the brick-and-mortar store, finding what we want, and ordering them online from someone who doesn't pay for the overhead of brick-and-mortar stores where people can try on shoes.
Right, And since I'm one of the few shopping locally, I'm the sucker who is footing the bill. I don't have a solution to this. I mean it's been happening for ages. The loss of the corner deli to the super market, the loss of the quirky hardware store, to the national chains... I could go on and on. The homogeneity of everything makes life much less interesting, but cheaper.
Industry "Best Practices." There is a cottage industry of maturity models and consultants in every specialty that can assess efficiencies. The is a lot of room for slack and bullshit, but if a company is under pressure financially, you can bring in external experts to determine if you are using outdated practices. This is often done very poorly, or to grind some unrelated ax, but also can be useful.
Yes, what are the BS jobs? And having just spent several minutes on the phone, entering all sorts of info to get to a real person. I've now got this view of a dystopian future, where my AI is talking to their AI for those minutes.... customer service becomes even more screwed up. Let's figure out how to fix customer service. And as long as I'm bitching about work, from what I can tell more and more health care workers are doing piece work, independent contractors working at different places weekly. I was talking to a guy last night at a party who's job is to hook the nurses and such up with the hospitals. All so hospitals and other health care services don't have to hire full time workers, with the attendant union/government mandated costs. (Sorry for the rant, unloading from minutes wasted on the phone.)
But what happens when the average AI-customer-support is better than the average-person-customer support? It seems like we aren't too far from that. Not for the complex cases, but for the "average" ones. Even today, automation is generally better for basic inquiries like store hours, current account balances and the like, which genuinely saves everyone's time. They just don't work for real problems ... but GPT seems damn close, for any scenario where the system has prior experience.
And mutual AIs communicating might be a good way to solve the general basic-communication, data entry issue. If your "personal AI" can negotiate all the basic address, transaction-number, member-ID data, re-summerize your problem in the jargon of the downstream provider, and give you clear options - then than might indeed be better world.
AI is only approaching human-level CS because human-level CS has deteriorated so much in the last 20-50 years. The few places that still hire a human who knows and cares about the particulars of your situation are world apart from anything GPT can do (even if it could also pull off the warm fuzzy feeling of talking to a live person, which it can't even try to do).
Fair, enough, but my point is that AI might exceed the current crappy version of human-CS, script driven, non-native speakers, limited discretion, overworked and poor attitude. I don't think it is on-track to exceed "Quality CS" anytime soon. But there are lot of of crappy call centers in the world.
Yeah, I'll agree with that. I'll also agree with George above that CS is screwed up and GPT will not make it any better (while adding in another layer of problems).
I like this post, though I disagree with it quite a lot.
I think many bullshit jobs are soulcrushing because they require the person doing them to spend the majority of the time on the low-value-add tasks, rather than the high-value-add ones. That's because (public or private sector) bureaucracies are, basically, superhuman intelligences composed of ordinary humans, some IT, and a lot of slow-moving process rules. And for many of those bureaucrats writing the rules, "have a human do it," is the default answer because their modal IT procurement fails. They know how to yell at a human to get them to do stuff.
So what I suspect you'll see happening is that a small portion of people in bullshit jobs will liberate themselves by using new tools, but most won't and will stay stuck in the current way of working for far longer than you think. For example, when MSFT ships GPT-5 embedded in Excel, it will transform the lives of those bureaucrats who already experiment with learning new features in Excel, but a shockingly large part of the economy won't even know those features exist until it becomes unavoidable.
I mean, we have people who still run their businesses on _paper ledgers_. That's gonna last for a while, even if there's another part of the economy where we have AIs building us flying cars made of nanodiamond and aerogel.
Agreed. An enthousiastic person can probably dedicate a few hours a week for a month to figuring out how currently-available AI can improve their workflow (if they're in a job where that's even possible). For an individual, it's a low marginal cost with the potential for high returns in productivity and reclaimed time. Overhauling the internal structure and processes of a large organisation, like a federal government department, isn't comparable. Organisational agility isn't a hallmark of civil service.
Wait. How much of the economy do we actually think is bullshit? Is there evidence for this? A post?
The YouGove poll in the UK found ~40% self reported bullshit jobs. The real number will be higher or lower depending on whether you think there are more bs-jobs that aren't obviously bs or real jobs that look superficially bs, or visa versa. Also the UK is probably a world leader in bs.
I don’t know of anyone in the business world who takes the Bullshit Jobs hypothesis seriously. There is too much pressure on margins in *all* businesses over the long term and on *most* businesses even over the short term.
Samo’s Law of Bullshit Is nothing more than a corollary of Parkinson’s Law: “Work expands so as to fill the time available for its completion.” At a for-profit organization It is the job of managers to be ever vigilant of the threat posed by Parkinson’s Law.
Bullshit jobs created by government regulation are real to be sure. But that’s nothing more interesting than deadweight loss imposed by legislatures or bureaucrats who are generally terrible at pricing externalities, because of course they are. That’s an unfortunate reality, but to a business regulations are a cost to be managed just like any other cost - that can be done well or badly, but it’s not a source of bullshit jobs, at least not from the standpoint of the bottom line.
I have a colleague whose job is contract management for our entire company. It’s a tedious and thankless job, and like the hypothetical security guard jobs discussed in another comment here, seems like a bullshit job. The thing is those jobs seem marginally important at best, until they aren’t, at which point they can be revealed to be very, very consequential. Risk management is a crucial undertaking for any business that expects to survive forever (most do) and those who toil to make sure a firm’s contracts and insurance and safety and finances and supply chains stay in order are doing very consequential work that is only fully revealed in the fullness of time, when crisis arrives.
This sound right.
I was headed here to say this. Private employers are under a lot of pressure to provide goods and services using fewer employees than they currently are.
"Bullshit jobs" usually mean the employee doesn't feel that they're fulfilling a meaningful function, not that the employer feels that way.
One area where I do agree with Zvi is that if AI makes it cheaper to comply with regulations, we can expect to see more regulations. The theory for this is that an optimistic view of regulations is that government sees the regulation as having a benefit, but the costs of compliance limit government ability to impose regulations right down the the point where the marginal benefit of a regulation is just over zero - at some point, you force your whole banking industry to move to Delaware or Singapore or something.
If an AI allows a small restaurant to comply with ten times the regulations we currently have and still stay in business, then I'd expect to see the pace of new regulations increase.
Every time I read a post like this I get lost in thoughts about all the non-bullshit-absolutely-can't-be-done-by-a-robot-or-AI jobs that are going unfilled. In the last year of our family's life, we have struggled to find:
--someone to paint the house ($20,000 job, only 2 of 15 people even responded, 3 month wait from those who did)
--chimney inspector (2 month wait, $500 for 30 minute job)
--chimney repair (another 2 months, $2500 for 2 hour job)
--vet for dog (first three weren't accepting new clients)
--groomer for dog (first four weren't accepting new clients)
--mammogram (booking four months out)
--tire installer for winter tires (6 week wait)
--propane line inspection to use heater (6 week wait)
--window repair (2 month wait)
--kitchen backsplash tile job (4 month wait)
Why aren't people doing these jobs instead of bullshit jobs!?!?!?
Because as a society we assign low status to people who do things and high status to people who (however uselessly) delegate.
I was going to ask if you live in a city / suburban/ rural. But then in all places, you have to get to know people. And who does what, if you can't do it yourself, handymen are in short supply always. Giving them cookies, or pizza or other food can help, be nice.
What? Why do I have to get to know people to hire them to do their work? Are painters who are too busy to take on a $20,000 really going to be swayed if I offer them $20,000 + cookies? My point was that if handymen are always in short supply (and they are), people should consider being handymen instead of doing bullshit jobs that will be lost to AI. I'm ready to hire them anytime.
Hmm OK maybe it is different in the city/ suburban/ and rural areas. I live in the country, if I need to find a new person to do something for me I'll talk to the people around me and get a few names. Then I'll call up Joe, and tell him I got his name from my neighbor Bill and ask him if he's available to do job X. Joe might say no he's too busy, or that he'll get back to me, or make a date to come out and look the job over. Now Joe will probably ask about this guy George who needs his house painted. What's he like to work for? Does he pay his bills on time? Is he always adding extra work onto the contracted job?
Anyway, my point is that with handymen in short supply, they get to pick and choose their customers. Why would they pick you? So yeah, get to know them, be nice. I have made cookies to give to my auto mechanic. The cookies are a thankyou and an investment in the future, where I have some car emergency, and my auto mechanic will go that extra distance to help me out, 'cause I'm a good customer. (For a long time I worked in a small business, you get to cherish both good vendors and good customers.)
I hate the hearts! Why is that? It's not money. I'm not here to make money.
The hearts are part of why news and bounded distrust is about more than just the ad money involved. (though I agree ad money distorts things more.) Bounded distrust is also about getting hearts. Even if we are paying for the news, the news will still be giving us what we want. Telling us the story we want to hear. And yeah, I want good stories too.
Well and finally, I don't check email so often, and silly 'someone gave you a heart email' are more of a pain than a pleasure. I'm a grumpy old man, 'get off my lawn', give me no hearts.
Turn off the hearts, it's somehow the wrong type of feedback. (Would only negative down arrows be any better? It seems unlikely.)
I actually DID initially ask Substack to turn them off, they said they could do it if I insisted. I ended up not following up. Not sure if that was right or not.
I know hearts seem good, I mean positive feed back. It's good. But now, am I posting to get hearts? Or what I think is the truth. Positive feedback, takes whatever most of your audience agrees with and amplifies it, till it takes over. (We should all build electronic feedback circuits... mostly because then we could have a common language. ) Positive feedback drives the circuit to one power supply rail or the other. (Maybe in the social sphere there are more than two power rails... power rails are the simplest one dimensional analogy.)
It says a lot about the feature that many people just use email filters to direct all like-related emails directly to trash. Comment prioritization shifts - first-mover advantage/Speed Premium increases even further, and of course the default on many Substacks is to only highlight the top two most-liked comments, hide the rest behind a Read More button. (Beware Trivial Inconveniences.)
Honestly I prefer having them off, like over at Scott's...when there's no (easy) way to like a comment, then one must write an actual reply to show favour, and that's conducive to further dialogue and generally higher standards. Pretty sure that's a big reason why ACX has perpetual issues with comment overload, compared to other blogs with similar-sized readerships...so many potential comments are nipped in the bud via the short-circuit of merely liking instead. One often finds additional thoughts to express in the process of writing a comment.
(Liking comments is a Using the One Time action for me. Used to not like posts either, until Scott expressed that that metric has actual meaningful value to him as a writer. So now I feel guilted into liking by default, unless it's a genuinely bad post. Here Be Goodharts...)
> Here Be Goodharts
I physically shuddered at this.
Me too, but it is also a nice turn of phrase.
Oh, definitely. It wasn't clear, but that was me expressing appreciation for a wonderful, horrible pun.
Re Scott and his liking hearts. I just want to say when Scott expressed this desire, (That he liked getting hearts on a post.) I thought it was bad. If Scott starts writing to get hearts... Anyway since then I have stopped giving hearts to Scott posts. There needs to be some sort of feedback from his audience. But that can be by the comments. "Hey Scott I loved this post." or with money, signing up for a subscription.
Agreed, it's a terrible motivation. I do still appreciate hearts on Twitter because they make more people see the Tweet in question. Here, I am mostly successful in ignoring them.
Scott has them turned off, so it's definitely doable.
There are few actual bullshit jobs, but the ones that do exist emanate either from government regulation or costly social signalling. Both of those burdens are likely to grow to meet the increased capacity for bearing them. Costly social signals that become cheap due to AI cease to be good signals, and people will find new signals that are more costly. Social signaling is a positional good but there's no good way to tax it the way other positional goods should be taxed.
The responses I've seen from other university instructors so far have gone along two paths, both of them pretty expected.
The first is going back to more high-stakes, in-person exams. No tech. Sit and write for two hours, answer the questions, etc. This way of teaching has been really unpopular for quite a while even though (in my opinion) it's a really good way of actually determining student achievement and learning. This should have the effect of reducing the make-work for students at the high-end.
The second path is extreme amounts of scaffolding. I teach in the humanities, which is a lot of writing. One system for teaching is scaffolding assignments so they build on each other and where each new assignment is required to incorporate instructor/peer feedback and correction. ChatGPT might make this easier, but barely, for the moment. This structure should have limited reduction in make-work. It works to teach writing to the median student quite well.
I'm lucky to teach a visual subject, film and media, so my assignments for next semester involve a lot of visual description, shot analysis, formal description. I've also limited film choices that students can write about to things released in the last year.
I also want to chime in to agree with Elena. Much of what you call make-work is make-work for you, but for the median student it's called "practice." People learn to write by being forced to write over and over and over again until they at least begin to approach the limits of their natural ability. Especially if that writing is in an unfamiliar mode (journalism, science, etc.)
Here I suppose the goal could be to get students to the point where their ChatGPT inputs are sufficient to spit out what they actually want to say and that they can parse the output to determine that it's correct.