The author non-ironically cites Klarna's year-old announcement around their pivot to letting AI do their customer support. Unmentioned is Klarna's announcement[1] from last month, after a full year the AI-first support, where they realize that people are the secret ingredient.
There's definitely some "there" there, but there is also a reason nobody is replacing their most expensive layer of staff (senior management) with much cheaper AIs.
Why do you think senior managers would replace themselves if they could? Given the choice between saving a company and giving a senior manager slightly more money or power, the senior manager will always win.
The ultimate authority in any public company is the board.
The majority of employees will never have to think about it, but everyone from the CEO down is ultimately hired at their discretion.
To see how things normally operate read up on the OpenAI drama. If Sam Altman wasn't as politically savvy as he was the board would have fired him in a Friday announcement and by Monday a new CEO would have been in charge.
And boards are largely made up of senior management of other companies, scratching each other's backs.
If it ever became clear that an LLM or other AI of some sort could genuinely replace senior management, I guarantee you'd see two things:
1) The price for access to that level of AI (assuming it wasn't possible with an open model on commodity hardware, which seems like a safe assumption for the foreseeable future) would quickly rise to be merely competitive with the cost of hiring senior management.
2) Boards would (nearly) all quickly agree that the human factor is very important, and regardless of what statistics you might have seen on the decisions made by these AIs in completely fabricated scenarios, or even a few _obviously_ rigged "real-life tests", there's no way these so-called "AIs" could ever really replace a human where it matters (ie, at the senior management level).
>And boards are largely made up of senior management of other companies, scratching each other's backs.
My experience with boards is that everyone on them would rather be doing something else. If they could offload 90% of their work to AI they would.
I know I did when I was a secretary of (several) boards and used a stt model + Claude to write minutes and agendas for the meetings I was presiding over.
Everyone rather enjoyed having a full and (near) real time overview of everything that had been said over the last three years.
I'm still hoping someone would come up with a good diarization model so we can have in person board meetings with as good transcripts as the zoom ones.
> Boards would (nearly) all quickly agree that the human factor is very important, and regardless of what statistics you might have seen on the decisions made by these AIs in completely fabricated scenarios, or even a few _obviously_ rigged "real-life tests", there's no way these so-called "AIs" could ever really replace a human where it matters (ie, at the senior management level).
Exactly how AI researchers have felt since they AI started doing better than humans in the first tasks. From the ridiculous stat that more chess champions have been disqualified for using AI than we have chess champions. Right now in every chess tournament everyone is "fighting" TWO chess opponents. Their actual opponent ... and an AI chess player (and yes, it's an AI now, IBM and their "not-quite-AI" program is out). If they come even remotely close to beating the AI player, on a move by move basis, they are disqualified and the match is halted.
Humans aren't going to play fair in competing with AIs.
Think roughly L7-L10 at Google, M2-D2 at Meta, 66-70 at Microsoft, etc. People that make "law partner" money but are not C-level executives mentioned in SEC filings.
Decisions would be made 1-2 levels up, so they would not be replacing themselves.
"When I was a child, researching a new topic meant walking to the local library with a list of subjects, an effort that could consume half a day. Back then, knowledge wasn’t cheap or easily accessible."
It was literally a free library, open to every member of the public.
Maybe it was better in big cities but when I was a kid researching papers on niche topics was tough because the local library had few books, mostly older ones, and I had to have my parents drive me there because it was too far to bike...
When we got 24/7 internet it was a game changer, it was like turning on a tap of knowledge on every topic. Of course the signal to noise ratio was higher back then...
I get what you're saying, but later in life I went to university, where I had access to better libraries, for which the quality and quantity of information is (to this day) greater than you can obtain on the internet [1].
Things like arxiv are a step toward that "tap of knowledge on every topic", but the overwhelming feeling I get from most internet content these days (and that includes LLM output) is that it's an inch deep and a mile wide [2]. A good book -- which can usually be found at your local library -- will vastly exceed the quality of what you can find on your own, if only because of intelligent curation.
Maybe I'm just a romantic (or just old), but for me, libraries are still where you do serious research.
[1] yes, I know that many university libraries have a selection of their content on the internet now, but usually this is behind a login, and it certainly isn't comprehensive.
[2] recent example: I've been working through McCullough's "The Great Bridge", and there are frequent occasions when I want to find out more information or see diagrams of what he's describing. The internet is, nearly always, completely useless for this. Tons of content on the Brooklyn Bridge, but it all says the same superficial stuff. Though I will grant that you at least can find some stuff like this now, if you try hard enough:
Going into university libraries fills me with awe and purpose.
Seeing a shelf of curated textbooks on a topic (eg non Newtonian fluid flow), gives me a feeling of depth. Each book represents years or decades of an authors life. And they condensed that wisdom into a textbook that could be consumed in a semester.
I would love to be a Knuth type character working in library and writing textbooks.
Yeah, there's a physical copy of my dissertation somewhere in the library system of my old school [1]. On a whim, I went to find it a few years after graduating, and it was (is?) in a special section of the library where there's nothing but shelves and shelves of old dissertations, going back for ~30 years (older than that, and they only have microfilm, IIRC). Thousands of black-bound volumes, with obscure titles on every possible subject.
Being in that space gave me exactly the feeling you describe. Each one is an artifact of years of a person's life, in book form. I guess that's true of any book, really, but it feels particularly acute for something like a dissertation. Every one of those unread volumes was a moment of long-awaited celebration (literally commencement) for a person. Humbling.
[1] I had to pay for that copy, but I digress. It doesn't kill the romance. I'm a sucker.
Unfortunately many universities are storing or downsizing their collections.
On one hand it seems logical that public domain books can readily be kept in digital form without taking up shelf space. On the other hand, the experience of browsing and serendipity that many find beneficial will be lost.
> because the local library had few books, mostly older ones,
I think this is an important observation. In the age of the Web, people might not appreciate what it means when your access to information on a topic is sometimes only through outdated sources, which is a bit different than not having access at all.
I was lucky that my mom frequently brought us kids around to multiple public library branches, but I definitely remember a lot of older books.
For example, a book explaining an electric circuits would involve photos of some kind of common household item large cylindrical battery cell with screw terminals, which I have never seen, not before, nor since. And pretty much any book on computers, robots, or electronics would be at least a few years old, usually several years or a decade or more. (This has improved, at least at my current local library.)
Another effect of being exposed to lots of older books is that, although my hair is still not gray, I'll sometimes inadvertently speak with anachronistic, old-timey language that predates me.
With the old public library books, I also got a dose of vintage subtle American propaganda (e.g., freedom and justice are good, and are American; dictators and secret police are bad, and are Nazi/Soviet), though the Cold War already seemed to be ending. I'm programmed to believe that that propaganda was a positive influence, but it's sometimes uncomfortable values, when one sees an overall citizenry that doesn't always seem to have been marinated as much in quite the same mix of programming.
Your public library didn’t have inter-library loan? Most public libraries have always been able to get you books not in their collection, you just have to ask.
Adding to the GP & sibling: Even in Los Angeles with some of the bigger libraries the IT/programming books were on the older (outdated?) side.. The more modern material was at the book stores like Barnes, Borders, or another one at the mall (I don't remember if it was B Dalton, Walden, or something else).. thankfully I could skateboard there after school or work and read them there for the evening since I couldn't afford them. Getting the internet and SNR of content back then was a game changer to me too.
I think it's fair to say that, particularly through the '80s and '90s, the IT/programming books were outliers in how fast they became outdated, because the field was just moving so very quickly. (It still is, to a large extent, but now we have the internet to disseminate that information.)
For research on most topics, from history to social sciences to particle physics, the books at the library (or available through interlibrary loan) would be plenty recent enough for anyone not already specializing in the field, and such people would likely already have access to at least a college/university library, and likely a variety of academic journal subscriptions (often through said college/university library).
Yup, I had no free resources on programming anything more complex than BASIC until I had reliable internet. I bought my first copy of Linux and a beginner's guide at B&N, for example.
Mine in a deeply rural area did not. Going to a library as an urban dweller would imagine was an all day affair. We generally did it a few times a year but that was part of a larger trip to go to "the better hospital" or what not.
> Your public library didn’t have inter-library loan?
Growing up before libraries had computerized databases, sure, they had ILL, but the books that weren't in their collection weren’t in the card catalog either, so you had no easy way of discovering that a book you might want outside of a library’s own collection (1) existed, and (2) was available through ILL. It wasn't until after the internet was generally available to the public that most public libraries I encountered had computerized catalogs with ILL availability.
I had the experience of needing to scan analog, historical original sources on microfiche for my history degree. You get really good at a kind of gestalt linear search through materials. Tricks like glancing at the first / last sentence of paragraphs for the general topic, then anticipating what the author is _probably_ saying in the next few pages, really helps you get through a lot of content fast, almost to the point of reading on an intuitive, unconscious level.
Also that the child who spent half a day at the library researching a subject probably learned something. Having an AI simply do all that for you teaches you nothing.
I still spend half the day at the library sometimes because I just love the library.
To say language models teach a person nothing is just completely absurd.
All learning materials really work in conjunction. The books I have read give me better questions to ask the models that lead to more books that lead to more questions.
Then layer on top of that the deep research models. It is just an incredible time for learning.
Our brains are built to learn and never stop doing it. It’s always “what” are we learning, not “if” we are. The use of AI is absolutely teaching something.
I don’t think this is the wrong take but I like the idea of moving dependence towards something that could be run locally vs. a search engine which can not.
> It was literally a free library, open to every member of the public
It is (or was?) possible for libraries to build reasonably durable collections of physical books that they can manage as they choose, shelf space permitting (and if not, books can be sold to make space). But as publishing shifts from physical to digital there is no first sale doctrine for ebooks, so they are controlled by license agreements from publishers.
Someone recently tweeted that we use so much software because we’re now expected to do more work than is reasonably to be expected of humans and I’d have to agree.
There’s too much information to manage and too many tasks to juggle to keep up without good tools. AI just extends the range of what a person can take on.
The article assumes the future of work will remain digital, but AI won’t just change how we work, it will orchestrate everything transparently in the background, eliminating the need for digital workflows and automating knowledge work entirely. The real economic shift will be toward non-digital, real-world human experiences, things AI cannot replicate, like hands-on service, emotional connection, and physical interaction, making them the most valuable assets in the AI-driven era.
A younger engineer at a place where I worked observed: "All of our processes are held together by human glue." This was after he discovered that there were individual people within the organization who possessed exclusive knowledge about how the business's products and processes worked, and even of who those people were. This is despite the fact that the organization has clear procedures for every operation.
I think it was John Gall who pointed out that all sufficiently complex systems operate in failure mode 100% of the time, meaning that some of the controls have been bypassed, or are being operated manually.
Because the automatic controls operate invisibly, the only visible work is that of humans overriding the manual controls, which will always seem chaotic and inefficient. In fact this could already be the status quo today in many organizations such as bureaucracies.
If this remains the case, then the AI-orchestrated workplace of tomorrow will look like the highly computer orchestrated workplace of today, with the computers operating invisibly in the background, and humans manually correcting the errors. To overcome this stage in our development as a civilization, the controls don't just have to be automatic, but error-proof.
All the names and places have been changed to protect the guilty:
When I was a young data engineer in 2014 at a very large telco in Canada the real time resource allocation of network bandwidth would work flawlessly from 6am to 11pm, but would have a random panic every day at 11.25 pm, where the cluster would enter a degraded mode until just after 6am the next day.
This was very weird, but I trace the issue to a bunch of NFS drives which started being used after 11pm. We figured some job somewhere needed to run then and there was nothing to do about it but migrate the hardware. After a couple of months, a lot of planning and tens of millions we moved the data from those drives to some shiny new beefy servers and the problems went away.
Years later I worked with someone else from that telco. Turns out he was in a team that did nothing but monitor the allocation of jobs on the compute cluster in real time and decide which machine each job ended up on. Their work hours were from 6am to 11pm.
This is a really interesting take. I find the argument that AI will lead to mass unemployment both historically and economically illiterate. There will always be some things a person can do that a machine can not (yet) do that is worth some economic value. I have had a hard time identifying what those things are. But you're probably right. It's probably non-digital stuff.
I read up till the "realisation moment" of something he could have easily found by using a notepad and some google searches. No, that wouldn't take weeks, it would likely take the same 2 to 3 hours that they felt "exhausted by" typing things into a textbox and reading.
Knowing what question to ask is always the hardest part of research. Getting there take a bit of creativity and it’s hard to compare one creative process with another. It can be very beneficial to swap between them when you get stuck though. Knowing that there is this Google alternative that produces similar results in similar times is a boon.
Good stuff. No one asked farmers to do less work after tractors were invented. I agree with this author after my own experiments with AI tools over the last three years. They're here to help people execute more work, faster. The human work will become deciding and controlling what they do.
Yeah, and back then they didn't face the same barriers to entry that we do for most jobs today. If automation is really hitting hard, there won't be jobs to run to in the cities either.
You mean the big cos that develop all those LLMs and robots that we give free pass for profiting by slowly killing programming/intellectual/creative jobs with ML trained on our work, you will be totally safe from them if you do farming? Why? If farming is safe from it then why intellectual/creative work isn't?
One benefit of owning a lot is that we see the big players using self driving tractors, bigger machines, etc. The point was that farming has consolidated over and over during the past 100 or so years. The point was that "no one asked farmer to do less work", but now you need fewer farmers to do it. That means those jobs need to go somewhere. That's going to be true of the jobs supplemented by AI - they aren't going away, but you'll need fewer and they will likely be concentrated with the big players.
> The problem is that there are less farmers and we don't know what to do with the former farmers that we don't need anymore
The US is way past that point, being down to under 2% farmers. China is mostly past that point, at 17% farmers as of 2017 and dropping. India is being hit hard by that, with 43% farmers. Their farmers are fighting farm mechanization.
Keep in mind we're not actually talking about farmers right now
We are talking about every industry that will have "productivity increases" due to AI tools
Which it seems might be every industry
It sure seems like we might be looking at a massive "productivity gain" across a ton of industries that is going to leave a whole shitload on people unemployed and franky
I'm a little worried about what that is going to mean
...And also leave the people who own companies that see those "productivity gains" massively richer, explicitly at the expense of the people who are now unemployed.
We are seeing in real time right now what happens when inequality grows too stark. The people at the top abuse their disproportionate financial power to capture the government, and then start ensuring that the government can never threaten their economic dominance, regardless of the consequences for the 90% of people who do not share in it.
There are a number of potential solutions to this. Many of them involve violence. Among the least violent are various forms of UBI.
> One surprising feature about all this tech is that I crunch through more work, more quickly than before — with a startling and happy result that I spend less time in front of a screen than I have in years.
I sit back in my chair with my vending machine cocoa. The 100K word report is done. Five years ago it would have been a 20K report. But still my AI assistants helped me get this work done in half the time that the 20K report would have taken.
The research justifying the narrative that AI will free up time instead of causing more work to fill up the time (if it works; if not it doesn’t matter) is done. Meaning I can take the rest of the day off.
Tomorrow I have 2M words to research. Five years ago that would have been 300K. But my AI assistants will summarize it for me. Meaning it still only will take half the time.
It's a little amusing to me that we ended up in a loop of (i) write bullet points, (ii) ask AI to expand bullet points into a long text, (iii) reader gets long text, (iv) reader uses AI to condense the long text into bullet points
Azeem Azhar never has to worry about employment in an AI-obsessed world -- no executive worth his salt will ever be satisfied with an AI-based lickspittle. You need the genuine human article for that.
I'm of two minds about all this. As someone who has become obsessed with seeing what these models are capable of, I can confirm that they can be used to achieve unprecedented things. But you get out what you put in. The most interesting results of these models are human-AI collaborations. If the "knowledge economy" just becomes bots passing generated outputs back and forth between each other, I think we're in for a rude awakening.
I can think of many AI generated outputs, perhaps with many quality suggestions, that I skimmed over or didn't fully appreciate. Did the sum of knowledge increase in the world? No, because knowledge has to be received by a mind. It's only the ones that I engaged with, influenced and ultimately made my own, that left an impression and added to the sum of knowledge in the world, often though a kind of alchemy of me and the model interacted in a "mind-meld" greater than the sum of its parts.
Human minds-minds in general, which these models don't have-stand for something. They're the glue that stitches things together, that makes important decisions and judgement calls, that forms the integration point where everything comes together. We can't remove that from the equation without losing something in return.
I've also caught myself making embarrassingly lazy queries for coding problems that would have taken me 10 minutes to fix, but I prompt for them anyways because I've already been through that song and dance and want those 10 minutes back. If I kept doing that, if that was all I ever did, my brain would wither to the size of a walnut. I genuinely worry how LLMs are going to deprive an entire generation of youths of the use of their minds.
The author basically tries to argue that not only corporate talent with low pay (e.g. customer service), but high impact talent pools will go from being scarce to being replaced by AI agents (e.g. PhD level researchers discovering new pharmaceuticals).
I don't think this has played out thus far. It still takes a qualified individual to ask the right questions of an LLM or Agent and then refine the results.
What I do agree with is that it's made talented individuals more productive, but that's less eye catching than claiming that it will cause companies to entirely restructure because educated labor will be replaced by LLMs.
AI is a big change, which will permeate life. This kind of change can be scary. Stories trying to predict how things will change is both meaningless and also thought provoking.
History is full of big changes. We've all grown up with grandma's stories of indoor plumbing or electricity coming, or TV or whatever. Many of us now would sound insane describing a rotary dial phone to our grandkids.
So yeah, change is coming, but to be fair it's always been here. And the effects of the change can be surprising or predictable. But we'll figure it out as we go.
When I went to school there were no mobile phones. Computers at home were very niche. There was no internet. No email. Crumbs, when away from home we'd write, you know, letters! Being a typist was a job. Secretaries took dictation and wrote in shorthand. If we wanted a meeting we got together in the same room.
Will AI change things? Surely it will. Welcome to my world.
Im not worried with AI as much as who will be controlling it. It could be a powerful tool for the wealthy to get wealthier and supress discontent if they’re the ones who leverage it. Yeah, of course nobody knows how things will eventually turn out but we ought to keep a tab on it.
For most of history, hiring a dozen PhDs meant a massive budget and months of lead time. Today, a few keystrokes in a chatbot summon that brainpower in seconds.
Lol. Poor Azeem Azhar who is a Freelance Contributor.
There's definitely some "there" there, but there is also a reason nobody is replacing their most expensive layer of staff (senior management) with much cheaper AIs.
1 - https://www.conversationalainews.com/klarnas-ceo-just-had-an...