back
127 comments
The idea that demand for work remains constant is also bad. Work is not a zero sum game, it expands depending on capability and needs. Ultimately our desires are impossible to satisfy because we always want more than we have.

Let's take software development as an example - in 70 years of coding we have automated many things, we have plenty of libraries and languages at our fingertips. And yet we have more devs than ever, how can that be? Computers were five orders of magnitude slower 3-4 decades ago. Where did that power go into? We can clearly do today everything we did back then with less people. In reality we invented databases, applications, internet, and cloud computing, and now we have more work than ever.

Another example - in ML we suddenly got to a point where almost everything we did pre-2020 is now trivial. Why are we hiring even more devs to do NLP tasks? Our old skills suddenly became outdated, we had to learn a new paradigm, and here we are, with lots of work. Even labeling is seeing a surge - you'd think we don't need it anymore since LLMs are so smart, but in reality we need even more benchmarks and tests now that shit got serious.

Same for medicine and other fields - they all seen massive improvements and yet humans work just as hard.

Jevons paradox happen all the time, except for when it doesn't.

People have some simplistic view where society always respond the same way to the same inputs. It's patently false, but well, the complex real world has no place on people's mind.

This phrase:

> Ultimately our desires are impossible to satisfy because we always want more than we have.

Is visibly false nowadays. A lot of people have been happy with what they have since the last century. Yet, an entire science keeps it as a fundamental.

But at some point, there is and there will be saturation.
Core Arguments:

> "Six separate impacts

In truth, AI technology will have several impacts:

1. Augmentation: AI will certainly augment certain existing jobs.

2. Job losses for non-adoption: new hires will be expected to use this augmenting AI, some who can’t adopt and adapt will lose their jobs.

3. Gradual jobs losses: new hires will diminish as productivity gains are realised: why have 5 when you only need three admin staff using AI to do the same tasks?

4. Jobs will be automated: some jobs will be automated and eliminated using AI. This has already happened with translators (Duolingo) and there are lot more candidates.

5. Legacy companies will disappear: traditional companies will fail to adapt and large numbers will lose their jobs.

6. New jobs will be created: new companies will emerge, as they did with online learning, with AI experts and knowledge managers. This is already happening."

===

Is anyone making a good faith argument that AI will "JUST" augment jobs? Do grown ups take the word "JUST" seriously anymore?

Terrible title.

I believe the argument on your point 3 is that those 5 augmented with AI would be able to do way more. But that requires the company to widen its goals, to change and augment its strategy as well, and this doesn't happen over night. The immediate goal is to lower costs because that's the simplest thing and understood by all, so they reach to AI to do that aka fire people. I know the company I describe here is "traditional" but I'd say right now all are like this. What will come next, how the new companies will augment themselves (not only augment their workers), we will see.
Think back a few hundred years when tools were all made by hand. As the Industrial Revolution kicked off it would have been easy to say, we'll you can do simple things like weave with machines but you'll never be able to build a machine that can do something complicated like build a [insert complicated tool name].

Fast forward to today where we've got the Tesla Gigafactory and you can see how far manufacturing automation has come.

Not only is it rare to find tools that are made by hand today, but most of the highest quality tools are made by machines!

Even if you argue that building tools by hand is better for the people no major economy in the world is going to go back to a less efficient and more costly model.

The Pandora's box of AI has been opened and the world is evolving. Yes jobs will be lost and yes we wont be able to predict how far this will go.

All of those examples are still just repeatable processes. The machines aren't making decisions independently and then executing complex tasks based on their judgment.

Not saying we'll never get there with AI...but it's a fundamentally different problem space.

The question of the future isn't about what jobs will be around, or what AI will be able to do.

It is if the greedy will attempt to horde the gains from the technology and set forth a path of world war and destruction in doing so.

Based on the defect rate coming out of that Tesla Gigafactory, we humans still have some runway. But it’s merely a matter of time.

Improved precision, driven by AI, is likely to squeeze even more costs out of overbuilt consumer products. The phenomenon of an appliance that dies 3 months after the warranty expires will become more and more common.

A far more recent and relevant example is that of secretarial and typing pools in offices.

Pre personal computers, this was a viable career: taking notes in meetings, typing up and sending letters, dictation, filing, etc. functions that, for 99% of folks they just do themselves.

Yet people still have to work for a living, which is the point
> Fast forward to today where we've got the Tesla Gigafactory and you can see how far manufacturing automation has come.

Fast forward to today where we don't have the equivalent of the Tesla Gigafactory for software dev or any business operations at all.

I'd argue that jobs wont be lost so much as change, just like all the previous technological jumps. You had an entire industry to cater to horse travel until cars came along. Then you needed an entire industry to cater to cars and their maintenance.
I actually hope certain jobs will be lost: Anything that involves physical, biological, chemical danger, or unhealthy work habits (e.g. night/weekend on-call). These kinds of jobs shouldn't exist, if possible.
Best practice is to talk about net job loss, or net job creation. "Augmenting" or "Replacing" jobs doesn't mean anything. Technology has been "replacing" jobs for hundreds of years, that's the nature of societal growth and change.

For instance, see Page 25 of this report which cites 25% net job creation through 2027 attributed to AI: https://www3.weforum.org/docs/WEF_Future_of_Jobs_2023.pdf

I think we can already see AI replacing junior developers. GPT-4 alone does good enough at scaffolding out semi-working code that I do think there's far less of a need for seniors to hand off tedium to juniors. This might be a partial explanation for why there seems to be significantly fewer junior developer openings today than there were a decade ago. People are going to split hairs over whether LLMs are actually AI, but the end result is all that matters. LLMs will continue to improve, especially on the code front, raising the minimal skill level just to get in to the field of programming as a human. It's already no longer enough to simply be able to build websites or mobile apps, and there will come a point soon where it will be challenging to compete against no-code/low-code platforms that use AI to tie all the pieces together or create customs at ease.

People love to cite prior automations, but the automobile only made hostlers obsolete because it required different skills to work on. AI fundamentally exists to replace skill itself. We might not be totally there yet to where it's a massive threat to wealth, but the potential is there in a way that was simply never the case with grain mills, the cotton gin, the printing press, conveyor belts, automobiles, and so forth.

I think this problem is a symptom of another problem though. I mean the fact we have to scaffold stuff out like that suggests our programming languages are shitty, wasteful and lack sufficient abstraction over common problems.

If we're paving that with junior developers or AI then we need to push the problem up the stack a bit more.

Example the other day. It took me 30 minutes to work out how to read a damn CSV in Go. Yeah I can ask ChatGPT to do it to me, but why the hell is it so damn unintuitive getting to there. Today I want to plot geographic data in R. This is also a hill to climb.

The instructions are simple. Our abstractions are just crappy and ChatGPT is just a shit crutch over a crock.

> I think we can already see AI replacing junior developers.

Even if this is the only effect we see, it still has pretty major consequences for the future. How would one become a senior developer in the future if there are almost no opportunities for junior developers?

>I do think there's far less of a need for seniors to hand off tedium to juniors.

I think that that's something that has been happening now for a long time, at least for the last 20 years.

In the beginning of 2000 if you wanted to have a streaming for your website, you'd have to pay your nose to an apache specialist tune your server so that it would be able to survive the load of streaming to a few hudrend of people. Today you just put your stuff on youtube and it's over.

More than twenty years ago I once met a "HTML developer". As I read here on hnews: nobody is interested in that kind of productivity anymore. People still have problems with HTML and CSS, but nobody is hiring someone only for that. Today you have to know both HTML and CSS among a ton of other things.

It's not this need for heightened productivity that scares me though, what really scares me is that a much, much more productive developer of today probably earns the same thing or even less than a much less productive developer from the past. The productivity gains were all collected by corporations and the capital, the workers got only stagnating wages and increasingly high pressure and scope of work.

So it's not only me who has this bad feeling. Workers and common people know that all the gains from AI are not going to stay with the working class. In the past, automation in the public discourse was associated with a richer society where people would have to work less and would have more time to leisure, now even the rocks know that this won't be the case. Companies will use AI to decrease salaries and increase even more their already immense leverage over the rest of the population.

If AI is replacing your junior developers, you’re missing the point.

Junior developers are an investment that will grow into senior developers.

I don’t think this is particularly controversial. Augmenting work means you need fewer workers and cheaper workers. “Replace” is just semantics.
Technical writing is one of those jobs where it's tempting to think that AI will indeed take humans' jobs. But only a company that has gone bananas would do that. All it takes is understanding that the deliverable is not the job; there's so much more to it. I wrote about it here for my fellow writers: https://passo.uno/ai-anxiety-tech-writer-howto/
> only a company that has gone bananas would do that.

There are plenty of companies that have, indeed, gone bananas.

There was a story on HN last year about a tech writer who lost their lost job and was being replaced by ChatGPT. They argued to their boss "but I'm 10x as good as ChatGPT, you aren't going to get writing nearly as good once I'm gone!" The boss replied "I know you're 10x as good but you're 10,000x as expensive. The company is okay with the drop in quality."

Job is something about you that others find valuable enough to pay for. So for AI to replace jobs we have to like it enough to spend our lives in separate holodecks, assuming human reproduction and young child care are somehow taken care of or we don't mind going extinct in a generation.

Seems far fetched, and also from a utilitarian point of view some will say it's Ok as long as everyone is really happy. But in every other scenario, there will be jobs. They just may not look like jobs we are used to now or require same training as required now. They may be jobs that people enjoy doing, or that people enjoy being done by people for whatever the reason. Someone who lived centuries ago would not recognize our jobs as jobs either. And maybe if someone wants to just sit in a holodeck and be fed, we will take care of it for them for the sake of global peace and quiet. For example, let criminals do whatever they want in VR, including commit crimes, cheaper than the rest of us suffering crime IRL.

> for whatever reason

I worked at a predominantly male engineering company a long time ago. They hired a girl to improve morale. This took me like 2 decades to understand.

Despite all the advances in technology and alleged productivity gains over the last century, people still somehow have to work 40 hour weeks. Stands to reason that using machine learning isn’t going to change the course of what is ultimately a deep rooted culture based on working these long hours on projects that aren’t your own.
The thing about AI is that software replaced people piecemeal. The report that took 1 person 2 weeks to generate now is a 20 minute batch job, but that report was run quarterly and was only 1 position. There is plenty more work to do.

We've already automated much of what is there to automate, most of our "bullshit" jobs are human communication or human reasoning. Once you fully automate 1 CSR or 1 compliance officer in an industry, you've effectively automated a large percentage of them. Instead of seeing a gradual displacement of labor, you're going to see large chunks of people become completely irrelevant. We're seeing this in the gaming industry now, where a huge focus is on AI art, aiding in (but not solely responsible for) mass layoffs.

I don't know if AI will eliminate more jobs than AS/400s did, but it will be a much more disruptive transition.

I would look at this somewhat differently - I think very 'narrow' jobs (with lots of people doing the same, very specific job) were previously more common. Think of the rooms of people replaced by 'mail-merge', for instance. For jobs that involve lots of disparate activities and responsibilities, it is difficult to 'automate' the job because it consists of many different jobs, none of which fully occupy a human. That type of work resists expensive solutions, since automating 1/16 of someone's job is just a slight productivity boost, but not necessarily much of a savings to the company, and you might need to develop 16 separate solutions with varying levels of reliability and effort. If a 'department' is only a handful of people, reducing the number of people has negative impacts on reliability and sustainability for the company - how vulnerable is your company now if you only have 1 compliance officer, and they leave / get sick / whatever? There will definitely be some jobs that get eliminated by this type of technology just the way mail-merge eliminated lots of workers (or a zillion other examples in the last 200 years), but it's not obvious to me it will even increase productivity enough to deal with the worsening demographics hitting pretty much every developed country these days.
I think the examples given in the article actually are in favour of the idea that AI will not be a catastrophe. Most people don't want to spend their life doing gruesome work on a farm, most people don't want to spend their life turning the same screw in a factory. If you believe David Graeber, up to 50% of jobs are actually meaningless 'bullshit jobs'. If AI replaced those, would society be worse off ?

The real issue is that people need a job to make money to survive. What we need might not be to save jobs but rather to find new ways to redistribute wealth...

I am not convinced ai tools will replace those bullshit jobs, when every other piece of technology since has not obviated those jobs, perhaps has created more even. There must be some inherent value for keeping the working population occupied on these sorts of things during the day. No time or energy left after a work week to plan revolutions perhaps, or to harden oneself against consumerist propaganda, or read into what the actual verbiage of what public policy is saying.
> This idea that AI will not impact jobs is ridiculous.

Is that a very widespread idea? I haven't really seen it, though I guess I try not to hang out places on the internet where there's too much nonsense.

I mainly see people arguing about how AI will impact jobs. (I think it's pretty obvious that AI will augment jobs, replace jobs, and create jobs, so a better way to frame the discussion is how much of each we are going to have and how can we make it so we get an overall positive outcome rather than negative one.)

He punts on the timeframe, using examples that occurred over 60, 100, even 200 years. Thus, you could apply the argument to nearly anything.
I've learned these firms and therefore their workers, who are hired to exert the firm's will, will rationalize anything and come up with any story as long as it steers them towards growth/market dominance. That is the dynamics of any and all attempts to monetize a new technology.
Like remote work vs physically proximate work in software engineering, I’m hopeful that people will remember the value of human cognition so much so that it’s considered, like it is today for some people in some cases, a competitive advantage.
I guess that's a given right? AI will augment job for some and hence lead to reduction in total available jobs since the people who are augment are more productive. But will AI lead to net new jobs?
Are these job losses directly attributable to AI, or companies simply reducing headcount from the Covid hiring spree?
I think this all remains to be seen. Someone has to truly solve the bullshit/hallucination problem before AI systems are truly useful for a wide variety of tasks.

More importantly, however, I believe that people have some seriously dumb thoughts regarding economics. Every single thing in life has some economic value despite how much I hate knowing it. I weigh things like "free time" against things like "income" all the time, and I could therefore assign a numeric value to any given aspect in my life. I do not do this because I am not a psycho. On the other hand, knowing this, I can definitely say that someone out there will notice when the price of labor falls of a cliff and then that person will come up with a job that requires labor and wouldn't be economically feasible if global labor prices hadn't fallen. More importantly, machines are expensive. Running an LLM is extremely expensive. Making modern GPUs is also extremely expensive. You won't need too sharp a fall in labor costs for humans to continue to compete against the machines. But this leads to the true problem. These AI systems, if the hallucination problem is solved, and if the profit turns out to be good, will most likely serve to further entrench the current neo-feudalist world order as the working class become poorer and the lords become ever wealthier. The scenario where a UBI is needed is ridiculous though. Should no one come up with a use for cheap human labor, the folks running the LLMs would run out of money and be forced to close down, and the jobs that used to exist would return. Every product needs a purchaser.

Finally, despite all of the above, I still think that the AI systems are a step in the correct direction. Yes, the current LLMs and spin off technologies might (most likely will) further entrench neo-feudalism. Yes, many of us (maybe even me) will be relegated to cheap manual labor. Yes, we will likely have some crazy dystopia of "post truth" in a technologically advanced dark age (I think we are actually already in that one though). But, the automation of work is a net good. The article's charts didn't show longevity, didn't show free time hours, didn't show standard of living. Taking us off of the farms saved lives. Taking us out of factories saved lives. These two things together created the modern conception of childhood. All of this was a benefit. So, despite the shifting of the labor force, humankind will likely end up in a better position despite the negatives that every new technology will inevitably spring upon us.

It's a lie which I'm frankly astonished so many people fall for. Put yourself in the shoes of the profit maximizing firm for once: Why would you pay dollars per hour for bespoke human cognition when store brand costs pennies of electricity and works [almost as well/just as well/even better]?
Because AI is nowhere close to replace a large swath of jobs? Besides content spam authors? Because even in case of artists, the 100% AI generated stuff is crap compared to the AI-supported stuff.
I think most businesses will struggle to completely eliminate the need for humans. Surely AI will continue to drive efficiency for those who remain employed and others will find new jobs and/or start their own businesses.

That, or all the jobs go away and then no one has money resulting in a non-functioning economy :).

The most hopeful answer is 'because the competition will use AI to augment instead of replace, and produce a superior product.'
Another good one that shocks me that people ignore is the long-term effect of remote work on salaries. If you can do your job from home, there's someone who can do it who has a cheaper home than you.
You say "put yourself in the shoes of the profit maximizing firm", do you expect governments to not optimize for cost, or other organizations should not optimize for cost?
We even have ample evidence from history that this did not occur with other innovations. We don’t have a field of workers anymore, we have one guy in a tractor. We don’t have a factory of workers anymore, we have one guy watching the machines.

We won’t have a field of lawyers, radiologists, programmers anymore, we will have one guy watching the machines. You can tell yourself you’ll be that one guy, but statistically you won’t.

Yes, but it is a comforting lie.
> We convinced ourselves that it was only physical working class jobs that would be automated through self-driving cars and robots. That may well turn out to be true but a more immediate fact is that their jobs are rock-solid and safe.

This was the main delusion. The reason robots don't do the jobs of "essential workers" is because they are far more expensive at the same quality. A machine getting sick or dying costs a business tens or hundreds of thousands of dollars, a person getting sick or dying is discarded into the gutter and a replacement is brought in.

A lot of people are getting paid for their opinions, though, or more specifically as the article puts it, their "text production." They are paid an amount far outstripping their usefulness for superstitious reasons: it's hard to predict the quality of the opinions of a future hire, so we just pretend that more expensive is better. To balance this out, society already made that class the most heavily taxed, and kept their existence precarious through surrealistically expensive health care, mandatory education, and homeownership (and in the US, I'd add dental care specifically.)

I don't see how they aren't all trivially replaced. Capitalists act like ownership is a job that deserves compensation, rather than a wager on what is usually a sure thing. It's not. Owners buy people to make decisions for them, in fact they overbuy and overstaff (creating bullshit jobs) and pit those people against each other to guarantee the best outcomes. The losers get sick, lose their homes, and their kids can't get into college; the winners get another day in suburban paradise with 60-hour weeks and two-hour daily commutes.

I can't imagine that 90% of high-status middle class jobs would survive properly-applied generative AI even at current quality, and techniques are getting better literally hour-by-hour.

At least we'll still need doctors, although at greatly reduced responsibility. Will they also cut hair?

> It's an uncomfortable thought but one we must face up to if it is to be managed politically, economically and socially.

Call me a doomer, but I doubt there's anything we can actually do to stop mass inequality in a future dominated by AI.

I worry political options here won't work because people misunderstand the roots of political power today. What I'm about to say is a very clear idea in my head, but I struggle to explain it well.

In a democracy power is distributed to all citizens, but the reason this system is stable is because those citizens have real collective power.

Historically this power has been physical – no single entity had more physical might than the armed resistance of its collective citizens. But today that power is mostly economic. Politicians must appeal both to businesses and citizens to remain in power and businesses must appeal to people to sell their goods and services and to attract employees. The most powerful entity in a capitalist democracy is the people since they are both voters and consumers.

The issue I have is that if 95% of your population is unemployed there's no longer much value in people economically. Additionally whatever demand they have is purely artificial and coming from the state in the form of welfare.

We need to remember true value isn't money. Money is what we use to acquire resources that are truly valuable to us. Things like food, planes, oil and gold.

If those who have access to AI can now acquire these things without trading with humans or employing them then businesses no longer need to care about humans at all. Food can be acquired without humans. Jets can be built without humans. Oil can be drilled without humans. Etc...

Similarly politicians will find that if the reason they value money is so they can acquire resources and businesses can easily provide all the resource they might want then there's really no economic gain to represent the people.

In this AI dystopian world I imagine you'll have a handful of resource creating businesses and the owners of these businesses simply trade resources with each other.

An additional problem we face with say some kind of redistribution of AI created wealth is that businesses are not tied to any nation. Assuming our political systems can avoid corruption and issue a 99% tax on companies leverage AI they might as well just leave. Then you can say, well we'll nationalise the businesses instead then, which is fair enough, but then you have the corruption issues again because how would we humans collective prevent state corruption if we're literally all just a drain on the state's resources and have no physical way to resist? Why wouldn't the state just say screw these leaches if that's what is in their economic interest?

Like I say, I can't explain this idea very well so sorry for this long rambly comment, but to try to end on something constructive here I think what we need to be thinking about now isn't what the correct rate of tax is in a world of AI, but how do we ensure people retain their collective power against businesses and the state? If we have no collective power then we should assume that with time any solution we try to implement will fail.