Has the decline of knowledge work begun?

At least for the moment, AI still needs knowledge workers to spec and prompt and check. AI makes knowledge workers more productive, but it doesn't eliminate the need for them.

And if knowledge workers are more productive, then knowledge work is cheaper. Cheaper knowledge work increases demand for knowledge work. So the number of workers required might actually increase. It also might not, but first order analysis that assumes decreased knowledge workers is not sufficient.

C.f. garment makers. Partial automation of clothes making made clothes cheaper, so now people have closets full of hundreds of garments rather than the 2 sets our great-grandparents likely had. There are now more people making garments now than there was 100 years ago.

I wonder how the ratio of people making garments relative to the total world population has changed though in this example.

No easy answer since most garments > 100 years ago were home-made. But I can confidently assert without data that the number of man-hours of labor in the average closet is substantially up.

garment makers chosen because of this recent discussion: https://news.ycombinator.com/item?id=43450515

the more obvious recent example is that we employ more bank tellers than we did before the ATM, because the ATM reduced the amount of labor hours needed to operate a bank branch and made a lot of marginal bank branches pencil out.

Only the recent trend of online banking services is really actually turning this around.

> Cheaper knowledge work increases demand for knowledge work.

This is Jevon's paradox.

> So the number of workers required might actually increase.

The increased demand for work turning into new jobs for existing workers, that is where the question is more complex.

This has gone the other way too in matters of muscle - people who wouldn't have been employed before can now be hired to do an existing task.

When you go from pulling shopping carts to an electrical machine that pulls carts for you, now you can hire a 60 year old to pull carts in the parking lot where previously that job would be filled by teens.

This is all a toss-up right now.

In an ideal world, I will be paying less for the same amount of knowledge work in the future, but as a worker I might get paid more for the same hours I spend at work.

My hours are limited, but my output is less limited than before.

How can you be so sure?

I found this document: https://web.archive.org/web/20210126040017/https://ribevikin... It asks the question "How long would it take to make a Viborg shirt?". The answer seems to be 354 hours per their experiments. This is from seed to shirt. (linen)

I'd be surprised if we had that many man-hours, let alone 3 or 4 times that (this is a single piece of clothing), in our wardrobes. Conservatively assuming a man-hour in the wardrobe costs us $5 (while people are often paid less, their salaries are also but one expense), you'd need at least around $1500 to equal just that shirt.

According to this cheap food is good because people just start eating more. Actually cheap imports can be really bad for local businesses.

Now imagine dystopian world where AI can solve most data / engineering / science problems for cheap. Or even for free, just ask.

I'm talking 100 years ago, not 1000. We've had mechanized fabric production for > 300 years.

From your document, weaving and spinning are > 85% of the labor in your shirt. Those would be almost 0% for a shirt made 100 years ago. And those wouldn't be the only steps mechanized 100 years ago.

AI working isn't the dystopian option. Imagine AI can do none of those things, but the people who control the capital believe they can.

That's the dystopia.

100 years ago is 1925 - people were already buying cloth in stores at that time. You need to go further into history.

It’s actually the Jevons paradox, non possessive, named for William Stanley Jevons. I thought it was possessive too because many people write it that way.

Free after it took a trillion in energy to train

How much time, energy, and resources went into developing your language of choice? How much datacenter energy is burned recompiling GCC every night for the last 30 years?

Comparatively little, I'd imagine.

Yeah and not unlikely.

There are trends in both directions in manufacturing. On the one hand, we had amazing fountain pens with forged gold nibs hand-tuned by a nibmeister that can do things you can't replicate now. The trend was toward really crappy pens that could be mass-manufactured and required less skill by the consumer to operate, but are demonstrably inferior. People chose to buy the cheaper, barely passable option more often by orders of magnitude.

On the other hand, we have automotive manufacturing. Modern cars are more reliable, safer, more comfortable, more performant, more featured, capable, (and expensive?) than ever.

With knowledge work, it's clear that certain things will get "optimized" into disfunction. (Look at the complete lack of agency corporations have taken away from the people operating the public interfaces to the company.) Be it help/support centers, resolving unusual issues, or dealing with automated bans/account deactivations.

It's hard to say what direction it will go...

> Imagine AI can do none of those things, but the people who control the capital believe they can.

I don't need to imagine, that's the reality we live in right now

> There are now more people making garments now than there was 100 years ago.

Exponential population increase only began 75 years ago. This is correct but the analysis is wrong.

Current AI is just a word calculator, it hallucinates because it doesn't know what should come next, a 5 is a 5 to it. LLM will never be general intelligence

If you recall how it was just 4 years back. Since then more progress in AI than in previous 40. Which in turn better than prev 400. It's accelerating and unstoppable. It will be very different world in 10 years from now. I hope I live this long...

There are several dynamics here.

1) AIs enable knowledge workers to be more efficient. They don't replace them. But they'll get more done. So, you might need fewer of them.

2) This frees people up to do different, more valuable things. There's a scarcity of people on the job market. We have record employment, not unemployment. In short, freeing people up to do valuable things that need doing is a good thing.

3) A lot of the economy is the service economy. It's not about producing goods, or providing essential things like food, health care, etc. Instead it is about providing services to people at some value. The reason the economy has transitioned to that is the industrial revolution. A few hundred years ago, most of the economy was about scraping together enough food for people so they wouldn't starve. That's a solved problem. Farmers use machines instead of dozens of employees. Some of those machines are autonomous.

4) Economies are about value chains and upcycling low value resources to high value services and goods. AIs make certain things cheaper which just frees up spending on more valuable things. What's valuable is determined by people and what they value.

5) If you fire all the people and replace them with AIs (the dystopian view of AI) and they no longer make money. So their spending behavior changes. The economy changes and adapts. Spending just shifts to things we value. Like maybe a human touch.

6) You could argue that much of the economy is already bullshit jobs. Who needs managers, marketing experts, social media influencers, and all the other fluffy jobs that we invent? Somebody values that. That's why that stuff exists.

2) and 5) assume people let go or fired have ways to make money afterwards. That's the high question mark no one is really considering right now. People can't spend money they don't have.

I think people talk excitedly about this post-labor society without considering how we upkeep it while all the value labor is managed by billionaires and worked on by an exceedingly small labor force maintaining the real labor force of AI. Current directions don't support topic ideals like UBI.

Idk if I'd call it "progress" at this point. There were a few big steps and then companies poured billions to shove it in our face. And from what I hear we hit a plateau with current approaches.

I don't think AI is really following Moore's law here.

Our brains are pretty efficient, so magnitudes less, yes.

I'm also sure you can take an average personal computing consumption and multiply that by # of PCs to approximate your compilation example. I'd wager it pales compared to what the Last year of crypto mining has done alone.

Progressive taxes

I don't see why it can't be written possessively since he stipulated it. It makes more sense.

You would write it Jevons' paradox. Six sources on its wiki page write it as such in their title.

> I hear we hit a plateau with current approaches

Then we need new. The important last year step was distillation as mainstream. In my opinion. Now using old models to train new is normal or even necessary. That was done before, but it was sort of experimental. Creating targeted datasets is a very powerful thing. Now big models can be trained on quality data instead of internet random mix. This includes long thinking and tools use examples from the beginning and not as fine tuning.

Another way of thinking is AI is steadily getting close to human level IQ. Not approximating, it will cross the line. This distance had reduced dramatically in last few years. Then it's singularity that everybody was talking about for so long.

This year we already have google's robotic multimodal. It's closed, but likely will be reproduced. Significant step toward useful generic robots.

We do. But we're not in an innovation environment anymore. It's pump and churn what we're currently doing and hope we brute force "intelligence". Ironic situation.

There's lots of promises out there but not much action nor real world appeal to this stuff. That's pretty much textbook gifting as of now.

I don't subscribe to this narrow view of economics. The AI fallacy is that whatever they do cheaply just stops being valuable. We pay a little for it, but we spend most of what we have on other things. What that is shifts over time. Pre-industrial revolution it was food and agriculture. Then it shifted to manufacturing. Today it's people doing stuff for other people (aka. services). Over time, we work less and less and we earn more and more.

You mention UBI. You could actually argue that we already have some notion of that. It's just a terribly inefficient, poorly administered, and very costly, and not very good version of it. People don't starve, they mostly have access to health care (the US being an exception to this relative to most countries, including most developing nations). And shelter too (people freezing to death because they can't hide from the elements is pretty rare). Many people live on what they get for free. It's called charity, social security, unemployment, pension, childhood, etc. But one way or another, somebody provides for them. And being dependent on just the free stuff is something that would horrify most normal people. But it's there for pretty much everyone.

Feudalism is cyclic. It comes and goes. We had a lot of it early last century. And then we got communism, socialism, unions, etc. And a surge in economic wealth for the middle class after things were rebalanced. The post WW II economic boom in the US was powered by Roosevelt's new deal. I don't think there's much respect or loyalty to the current batch of trillionaires. They can exist only because people allow them to. Future governments could find themselves empowered and tasked to do something about the economic wealth distribution. Such things have happened before. Often after some kind of revolution.