AI and LLMs, in my opinion, are effectively the industrialization of knowledge work.
Like it or not, industrialization brought many benefits and we couldn't imagine our modern way of living without it (just think about medicines). Industrial production made so many things so much more accessible for so many people.
But an industrialist perspective doesn't care about the process of making stuff, only about the outcome. The shoemakers deep knowledge of leathers, his intuition of all sorts of different shoes, and the pride of that competence—that doesn't matter when you can make thousands of shoes a day that cost 0.5% of something artisanal. The people working at the shoe factory only carry things to and fro and occasionally fix some machine issue.
This has tons of benefits and tons of downsides. But with physical goods, a shoe is a shoe.
With knowledge work like writing, the process is often the entire point. The point of writing a paper is what you learned in the process. The point of documented software architecture is to develop an understanding deep enough to drive coherent downstream decisions. If those things are no longer true, the asset it stands for is basically worthless.
If the point of writing a given paper is learning about a subject, then it's not knowledge work, it's studying. Just like an apprentice shoe maker is making shoes mainly to learn, not to produce footwear. Acquiring more of a skill while working is incidental both when shoe making and when writing documentation. Documentation which will help others understand the system. In both cases, the main point is the finished product. I really don't see the distinction you're drawing.
We're inclined to think of them as separate, but shoemaking also requires some knowledge work. Industrialization detracts from the value of process in physical manufacturing as well, though to a lesser extent.
It's actually quite interesting here, now that I think of it, that we've been calling it "knowledge work" the whole time, rather than "thought work".
I have come to fully accept this in my line of work. My knowledge of the internals of software is similar to what a machinist knows about using a manual milling machine. It helps understand what a CNC process is doing at a bit deeper level, and some of the issues that can arise, but holding onto the idea that direct software work will continue to be valuable is similar to a machinist clinging to the idea that they need to position a piece of metal on a plate, pull a lever, and take measurements.
It's even knowledge that should be preserved and is needed for some edge customization or prototype or debugging case, but it's not how the vast majority of work will be done.
> Like it or not, industrialization brought many benefits and we couldn't imagine our modern way of living without it
Industrialization isn't unquestionable good especially the way it went. A lot of good things from industrialization should be weighted against side effects and lots of negative externalities. Early factories and infra companies externalized all the costs, dumped waste into rivers, and CO2 in atmosphere started going up right at that time in history. They didn't need to pay workers living wage or ensure safe conditions, they were pretty much all monopolies with company towns and no competitors. And those early industrial enterprises didn't displace remotely as many jobs as language model operators are threatening to.
Also people usually don't consider which of good things would've happened anyway without industrialization. Yes, it's difficult to imagine because it needs alternative history...
> Like it or not, industrialization brought many benefits and we couldn't imagine our modern way of living without it (just think about medicines). Industrial production made so many things so much more accessible for so many people.
I largely agree with your point but I see a lot of people lumping everything together.
Do we need 50 different shoes produced by borderline-slave conditions in SEA to have plumbing? Do we need to pollute be getting a car to drive to the gym to have ambulances rushing people in to the latest medical facilities?
A lot of what we are seeing is simply capitalism running rampant without much of a care with regards to actual humanity. There was no preconditions we needed to eliminate shoemakers to have a safer, more functional society. I would even argue that it's a mistake that we are paying for in human livelihoods, reduced production and over-reliance on foreign countries in the West and climate change.
Also "a shoe is a shoe" is unnecessarily reductive. Even assuming the quality stays the same (it does not and has no incentive to) and leaving all semantic arguments aside, with shoemakers you have decentralised production that boosts local economies. You have a community that can provide both (mostly locally sourced) material and ultimately knowledge. Up to a couple of decades ago, people used to go to the local shoemaker to fix their shoes, making the whole chain more sustainable. And of course you have none of the scale of the climate impact, the oversupply of cheaper goods and centralisation of production to a few major countries who have control both over the workforce and consumer.
And none of this is antithetical to technological progress. You can have shoemakers that have new tools and new techniques to aid them, without the need to build shoe factories pumping out Nikes 24/7
In software, the factory is the compiler. The code (and preceding artifacts) is the design process leading up to sending something to the factory for production.
In the shoe analogy, there is absolutely deep knowledge and a creative process that goes into designing a new shoe. This includes market research, defining the new style, selecting materials, building and testing prototypes, etc. This all happens before factory production.
Software development is the same. The factory is the compiler.
> The people working at the shoe factory only carry things to and fro and occasionally fix some machine issue.
Were this to be true, there would be no workers at all, it had all been automated decades ago. Still there’s around million shoe workers in Vietnam around, I think. Adidas tried automating shoe production at it failed miserably[1]. Seems nothing beats us pesky monkeys in hand-eye-coordination for quite some time. I think the problem was also that recalibrating the system to create different type of a sneaker was extremely hard and expensive, whereas doing same to a bunch of Vietnamese workers requires you to just, y’know, tell them.
People who never done any manual labour with their hands always think a job ω is quite simple to automatise away. And if you try it, whaddayaknow, it fails miserably. How many years now the fully automatised driving has been supposed to completely take over long-haul trucking just next year?
Coming back to your main point, with which I agree with, having an LLM doing your reading or writing for you is kinda pointless. It is like paying somebody to watch them to eat your dinner. In the end of that process is starvation. But equally we have been afraid of industrial automation taking completely over for at least hundred years, and we find that automating a making of a single sneaker is bloody expensive and bloody difficult.
That leads me to believe this automatisation of knowledge will not look quite like the C-suite prophecises and we monkeys will be needed for long time being, even in SW development.
[1] https://techcrunch.com/2019/11/11/adidas-backpedals-on-robot...
industrialized medicine requires trained consultants, and researchers who understand the underlying science and can still think.
The nuance is, industrialized material production.
Well, no, the point of writing a paper (or a Hacker News comment) is to share what you learned in the process with other people.
It only truly becomes slop if nobody reads it and learns something.
It's also how I picture things; except that LLMs are not the whole industrialization process, just the latest and most visible step (former steps include standardization of low quality in various ways - yes, the beloved "industry standards" are quite often horrendous from a purely technical standpoint).
One of the consequences that industrialization has had in the past regarding the production of physical goods, is that quantity and affordability has often been (but not always) detrimental to quality of the goods and loss of knwoledge for the people (so many crafts that have disapeared, which makes society less robust as a result). Also, a loss of social connection and peer-to-peer relatedness, as we find ourselves increasingly dependent on large scale processes that we can't even see, or feel not taking part of, instead of more local relationships.
The same thing happening to "intellectual" work is not really appealing: more slop and less education, is that where we are heading to?
> AI and LLMs, in my opinion, are effectively the industrialization of knowledge work.
I mean, sure, but it's also a lot more than that because it fully encroaches on intellectual activities outside of work. So I think it's way, way too myopic as see it as just another industrial revolution.
Frankly, if you don't actively reject LLMs in some areas of your life, I think you're gonna end up a lot dumber. Or, the kids will. If the only thinking skill we have is prompting, or not even that - there's a push for "predictive" LLM agents - then I suspect we just grow dependent on always having a subscription and stop using a good chunk of our brain. Like, if you're no longer writing, no longer reading anything that isn't ELI5, no longer formulating hypotheses, making art, learning crafts, or even comparison-shopping... what are you gonna use it for?
> Like it or not, industrialization brought many benefits and we couldn't imagine our modern way of living without it (just think about medicines). Industrial production made so many things so much more accessible for so many people.
It is extremely easy to sit in the First World and like the fruits that the Industrial Revolution brought now, some centuries later. Workers that were brought into factories had to work long days. Now we have the 40-hour workweek, and a two-day weekend. Now we have a “middle class”, i.e. wage workers with disposable income and a mortgage. We exploit the rest of the world for cheap goods. We exploit our immigrants for cheap labor.
Climate change is becoming a bit on the nose, but we don’t have to worry about the worst effects.[1]
In short, it is extremely easy to compress modernity to the headline: Progress Happened. But in the real, uncompacted world there was strife, struggle, malcontent, protestations, probably every shade of opinion and reaction. And with all that we didn’t arrive at a polished gem. We got a rich First World, for the most part (right up to the rise of China).
To bring up that compressed view of the Industrial Revolution in order to stifle discussion (let’s be real) makes no sense. We are living in it now. Of course we are struggling and debating with it now. Just like all the other historic, uncompressed events
- - -
To also compress the dimension to “progress” and “not progress” is a falsehood. People want progress that works for them. Not progress that oppresses them.
[1] https://ruralindiaonline.org/article/mosquitoes-are-ladakh-s...
> But an industrialist perspective doesn't care about the process of making stuff, only about the outcome.
It sounds nice, but in actuality the process of making stuff in the most optimal way possible is the pure essence of any industry.
At the artisanal budget, industrial version will have orders of magnitude better tolerances because the processes are perfected.
When you do things artisanally your ability to perfect the process is very limited. Only going industrial affords you the scale to perfect the processes.
An AI design doc has all the outward appearances of a -- good! excellent sometimes! -- preAI design doc, in my experience. But it also has no reason to exist. There's no social aspect, since no one has gone through the process of actually understanding the systems involved, there's no meaningful way to review it ("why do things this way?" "Because it said to do things this way"). And it's not needed for the implementation either, for which the original prompt, whatever it was, suffices. By industrialising the output of a process, we forget why it was necessary to go to the trouble in creating it by hand in the first place. Perhaps that's ultimately how things advance, and right now we're in the absurd middle, treating proofs of work like the work itself.