AI won't take our jobs and it might even save the middle class The future described in OpenAI's mission statement, in which autonomous systems "outperform humans at most economically valuable work," sounds like a hellscape to MIT economics professor David Autor. A world where humans supply only generic, undifferentiated labor and wealth flows to AI system owners and rights holders would …
Autor does sound a bit like an arch-optimist, living in some alternative reality (never been to a Post Office?), a kind of Voltaire's Candide. I do however like his statement:
"Artificial Intelligence is this inversion technology"
as it does evoke that tech's potential for topsy-turvy upside-downedness, which many expect from it. On the other hand, his pastel watercolor suggestion that:
"we should ask not what AI will do to us, but what we want it to do for us"
when we should obviously ask both questions, and more, simultaneously, tells me that his spinal tap is turned nowhere near that 11-overdrive limit, more like 0.8, or 1.3 (IMHO).
It won't, he says, let untrained people perform skilled tasks like catheterization. Thank goodness for that. I would not want some AI-trained clutz trying to shove a tube up my mary-jane. But surely this only serves to illustrate the previous "train as a plumber" guidance?
What AI can do for or to us depends on the data set and rules the AI is trained on. My suspicion is that I or you are not going to be the ones given the training task. Thus AI will be a tool which does not empower the average person but just becomes another tool to control us. AIs will be promoted based on their ability to push power to the employer or large corporation. Expect egalitarian AIs to be made illegal.
Beer icon because Valentines Day.
We know what it's trained on. Everything that can be scraped uncritically from the internet and the contents of every account on cloud services run but the usual suspects. Given that will already include its own output, and increasingly so as time goes on it's clearly going to develop by eating its own dog-shit.
> We know what it's trained on. Everything that can be scraped uncritically from the internet and the contents of every account on cloud services
Er, no. The article wasn't talking about Large Language Models trained on scraped text from the internet. The systems under discussion would be trained on data pertinent to the task at hand. An ML system for reducing the search space for new pharmaceuticals would be trained on a data set of existing pharmaceuticals, for example. ML for spotting cancer cells would be trained on medical scans, etc.
Sooo, the idea would be that someone who's not a doctor could make doctor-level decisions, provided they have a LLM helping them?
Sure. Go ahead. Give it a try. Seriously.
There's so much hype around "AI" that I doubt we could persuade the people who came up with this that it's a bad idea. So, run a trial or three, see it fail spectacularly, and then maybe we can move on.
Just, please, during the trials, have a real doctor double-check everything before doing anything on the patients.
This line of thinking makes it feel like we're already approaching the backside of the boom cycle, where the true limitations of the technology are starting to show and those invested heavily in selling it have started pitching more practical sounding but still highly improbable use cases to keep interest high.
Saying that an LLM could be used as an instruction book that can enable non-skilled people to performed highly skilled tasks sounds silly to people who have been watching this closely over the last couple years, but there are just as many who weren't scrutinizing it as much that will hear that and think "Oh, it could do that? Wow that sounds great, let's go ahead and look into that then. Here, have a research budget." And thus, another year or so on the payroll while they try to stretch out that dead avenue as long as they can.
>Sooo, the idea would be that someone who's not a doctor could make doctor-level decisions,
>Sure. Go ahead. Give it a try. Seriously.
In europe your smear test is processed by an ultra high quality scanning microscope, that images the whole sample in several visible and invisible wavelengths, at a stack of focus steps and automatically adapts to any flatness variation in the slide substrate. The image is digitally enhanced and fed to a machine learning algorithm which has been trained on billions of previous images.
In the USA a single field of your sample is looked at for under a minute by a real doctor using a Mk 1 eyeball and an unserviced uncleaned uncalibrated microscope they bought decades ago when they got the job.
One of these produces diagnostic outputs indistinguishable from random chance.
"someone who's not a doctor could make doctor-level decisions"
I don't think that was what was being said with the example of nurse-practitioners. The idea seems to me to be, rather, that you don't need to go to a consultant for something that an MD can handle and you shouldn't need to go to an MD for something that a nurse can handle. And furthermore, that at each step of the competence / experience ladder, that "AI" (which is not really AI but something more like a decision support system (remember those??)) can extend the range of cases that can be handled by any competent practitioner at that step without needing to escalate to higher steps.
If that is the way that it pans out, then people at lower steps can offer higher value (which theoretically could then reflect in their salaries)
Of course it will be a doctor that will make the decisions but he will receive the same low-pay as a non-doctor. That's where AI shines at, to allow employers to pay less (if at all) for the work done.
Why do you think AI is spreading like wildfire even though it's far from being ready.
In the time of Elizabeth I (say in 1600AD) there were about six million people living in England. Most of them living a subsistence existence.
In the time of Elizabeth II (say in 2000AD) there were about fifty million people living in England. Most of them living a much better existence than their forebears!
Technology has never reduced either jobs or wealth .... in the long run. Economics is not a zero-sum game!
Of course, CHANGES in technology have often been hellish........
There's been a new book published about the Luddites... If I understand Tim Hartford's presentation of it correctly (on his Cautionary Tales podcast), the new textiles technology DID create more new new jobs than it displaced, BUT the new jobs came decades too late to benefit the skilled workers who were displaced by the new machines.
Of course it's not a zero-sum game. Technology has always facilitated the transfer of wealth. It has always been about taking from those who have less and giving to those who already have more. With the exception of a Bolshevik revolution of course.
"But it doesn't have to be that way. In fact, AI can improve the lot of the middle class."
Sure, just as megacorps can pay their fair share of tax. And just as there is enough food in the world to strongly reduce hunger or eliminate it if we'd put enough effort in it. Or like rich companies making record braking profits can afford to give workers a break to pee.
Technological improvement is an enabler. It can enable to do more things, produce more goods and services. It can be used for both improving society or used in a way that prioritizes even more profit at the cost of harmful side effects. If it is unable to do any of that, it's a useless tool to start with. In order to try and predict what choices are likely to be made, let us look to the past.
He points to several studies of the impact that GitHub's Copilot and OpenAI's ChatGPT have had on computer programming and writing tasks respectively. Neither eliminated the need for expertise but both helped make less workers more productive.
No they didn't. They merely replaced incompetent workers with incompetent computers.
AI will not replace skilled programmers. They will always have work fixing garbage code written by incompetent ones. The only question is if they care whether said garbage code was written by a human or by a bot.
In the time of Elizabeth II (say in 2000AD) there were about fifty million people living in England. Most of them living a much better existence than their forebears!
The average rent for a semidetached house in London is now over 100% of the average net income of a Londoner.
The existence of the bourgeoisie today is now as precarious as that of the proletariat back in the days of Karl Marx.
"The existence of the bourgeoisie today is now as precarious as that of the proletariat back in the days of Karl Marx."
There are a vast amount of middle class families who are effectively living from paycheck to paycheck, just like the proletariat back in the days lived with just enough from each paypacket. The difference is that back then the proletariat were having just enough for a roof over their heads and food on the table, while most bourgeoisie families living from paycheck to paycheck are doing so because that paycheck extends beyond housing and food to clothes, gadgets, entertainment, holidays and a bunch of other stuff. The quality of life has gone up, but for many people they are still in precariousness because they never learned (possibly also were never taught) about managing money.
I notice that when I read articles on the potential of AI as seen by people who don't have actual expertise in how LLMs work, they become a kind of inkblot test for what that person wants to be true about AI.
In this case we hear about the potential of AI with "appropriate guardrails" but there isn't clear evidence from anybody with expertise in the field that such guardrails are even possible with LLMs. Perhaps we'll develop a future iteration that can reliably behave safely, but it will not look like what we have now and assuming that it's a problem we will solve quickly is more of a leap than it appears at first glance.
Right now, LLMs are great for generating junk data or vast amounts of information which will be readable but of doubtful quality. That's lovely, but it doesn't help people who care about reliable information, which is most of us with a job to do.
"Autor argues that fears of a future in which AI will leave humans with nothing to do are misplaced and in fact, AI can improve the lot of the middle class."
Pull the other one it has bells on it. I have read this garbage from the start of the internet many decades ago how it was going to improve humans lot in this world. Well in case any of these geniuses have not noticed it has done nothing but make it worse day after day, year after year, decade after decade and shows no signs of any improvement in the future as it progresses onward in the same way at all times.
"I have read this garbage from the start of the internet many decades ago how it was going to improve humans lot in this world. Well in case any of these geniuses have not noticed it has done nothing but make it worse day after day, year after year, decade after decade "
What bullshit. By almost every available measure, the lot of humans in the world is far better than it was 20 years ago.... life expectancy, infant mortality, extreme poverty, availability of water and sanitation, and many many more are far better, if you are talking about " humans lot in this world". If what you really mean is, as from the original quote, "AI can improve the lot of the middle class", that's a different kettle of fish altogether (specifically, for the middle class in wealthy 'western' nations). For these people, their gains in the last 20 years have been minimal or marginal, especially compared to the gains of poor people around the world which have been spectacular, in many countries actually creating a new middle class. Western middle-class gains have also been minimal, financially, compared to those of the 1% (or, more particularly, the 0.01%)
Sure, there are specific things that are worse, plenty to gripe about without being so generally fatalistic!!
"What bullshit. By almost every available measure, the lot of humans in the world is far better than it was 20 years ago.... life expectancy, infant mortality, extreme poverty, availability of water and sanitation, and many many more are far better, if you are talking about " humans lot in this world". "
Indeed it is more bullshit from you. That is why people in the UK and elsewhere are swimming in feces at the beaches, people are getting poorer by the day as the parasite taker rich and their greedy corporations grab an ever increasing share of the pie. Many people in all kinds of other places would have a bone to pick with you on that trash you spout off on improving at all as well. You can tell your lies elsewhere I for one will not believe them when any object look at the state of the world shows no such improvements to be seen.
Been here before, in the late 70's my first job was offset Lithographic printing, was a great job involved photography, chemistry, and took some skill to do well, then along came Xerox and a number of the printing machine companies introduced electronic control too (obviously way more primitive than today).
So a job you could take pride in became mundane button pushing, boring, dull and wages began to slide as previously needed skills were not needed..
Next door they had a pool of around 20 women doing copy typing, audio typing, shorthand etc., I went back 5 years later and the same work was being done by just 2 women using Word Processors
So yes jobs will go and skilled jobs will get augmented, but how will the people using the new aids feel about them?, it's going to turn a challenging job you can take pride in and turn it into brainless box ticking.
So you were top kid in your class at school?, wait till the flashy new AI tells you exactly why you are wrong in 15 different places in the paper you so carefully prepared (citing information that became available 2.5 seconds ago).
AI probably wont kill all jobs but it will kill the fun and sense of achievement out of many of them.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
