"An AI Job Apocalypse?" – Goldman Sachs Report [pdf](goldmansachs.com) |
"An AI Job Apocalypse?" – Goldman Sachs Report [pdf](goldmansachs.com) |
Still waiting for anyone with this viewpoint to offer a single educated guess at what any of these created jobs might look like even in very broad terms.
All I ever see is basically a religious belief that surely something will arise.
I get that previous automation waves have lead to new jobs in the past, but automating knowledge and intelligence is just fundamentally a different thing than anything we have automated in the past and what exactly stops AI from taking those new jobs too as soon as the need for them is recognized?
2. Agentic tourism - maybe you’ll be able to send your AI agent/robot to visit a place before you decide if it worth your time and investment
3. Vehicular trains on highways where vehicles going to similar destinations coordinate and move as a unit. There’ll be people optimising roads for this. E.g., exit lanes on highways are designed for a handful of cars to exit safely at a time. What if vehicular trains can be thousands of vehicles long like those going to Burning Man?
4. Hallucination historians - people whose job is to document how AI errors have led to changes in the course of history
5. Animal linguistics - people whose can finally study the correlations between how human and animal languages deliver the same information
6. People designing puzzles and challenges to keep AI engaged during routine maintenance
7. S3x workers who specialise in servicing AI bots
8. Cryptographers whose job if to encode content using non mathematical methods because math base cryptography has become a solved problem
9. Interior designers who specialise in humanised decor because everything else is now handled by AI
10. Lawyers who fight cases on AI’s right to participate in the world as equals of biologics
(7) is basically "a mechanic/service technician" which is literally a job that already exists all over the place. Is is, at least, the only one that doesn't seem like one of these "agents" could do itself today, but in the world where the agents get that good, that level of robotic ability - the ability to manipulate a screwdriver with human dexterity - doesn't seem very far off at all and is being actively invested in by many many people.
Welcome to economics! Bullshit and conjecture all the way down.
Fundamentally, economics is the belief that humans have self-interests and act on them.
Economics is obviously not 100% right - it’s not a physical science - but more often than not, you can see it in action.
Economics is closer to weather science - we can make certain inferences and predictions which can be wrong but the basic working theory is broadly correct, imho.
Another frame I’ve heard is that people will only trust humans when it comes to their kids, health, and money. That’s the simpler view.
Sample efficiency, energy efficiency and training data availability.
The current tech stack is around 6 orders of magnitude behind human capabilities. We have scaled up a paradigm that's been built out continuously for decades (transistors, matrix multiplication). To close the gap requires completely reinventing the entire stack (e.g. spintronics, physical graph traversal). The most advanced tech that is only theorized and not yet even proven is still 4 orders of magnitude less energy efficient than a human, and we have no idea how to get close to the same sample efficiency.
The question you are fretting won't matter for decades, and by then our culture, social and economic framework could be totally different than it is today.
Rather than worry about something so abstract and misunderstood, I would suggest focusing your time and attention on what is within your control here and now: who do you want to help, and how can you be of help, such that your efforts are sustainable and rewarding.
And if you really want to fret something at the edge of your control, fret about how AI is being used today to manipulate you and those you know into believing things that aren't in your best interest but are in someone else's best interest. If you aren't paying, you're the product. Most free media exists manipulate you, and if it has been designed to be addictive then it's virtually guaranteed.
The new businesses created, might look very much like the old businesses that disappeared. But they're not focused on scale and profits. They're focused on an ideal of pushing the boundaries. Maybe a single craftsman making small one off hand made pieces. Maybe a pizzeria that puts out a small quantity of the best pizza. Little individual pursuits towards perfection. Enabled because economics is not important. If physical AI and super intelligence can do "everything" (work with me here) you're free to do small scale things you love. You're also free to raise children at home, work on mega projects, explore the world.
The alternative is a small subset of an elite inherit the world's wealth, create a prison society of surveillance robots that keep the permanent underclass at bay with video games, tiktoks, and pills.
Sounds fantastic(al).
> The alternative is a small subset of an elite inherit the world's wealth, create a prison society of surveillance robots that keep the permanent underclass at bay with video games, tiktoks, and pills.
I have to be honest with you, I'm not sure how anyone could be paying any attention at all to our current socioeconomic reality* and believe that the first alternative is likely to be the way this all goes (unless the issue is eventually forced through a massive, generational, bloody revolt).
(* I am admittedly being US-biased here, some other countries may fare much better)
Which count of countries comes out higher?
Americans are fucked. European on the other hand may start bring out the guillotines again.
I’m not sure if this really is different or not. I also think there’s an enormous amount of white collar work that isn’t actually “intelligent” work. Especially in software development.
The question I’ve been pondering isn’t about “where’s the intelligent work?” It’s, “what do we do when we’ve successfully vanquished work itself?” Will we move towards utopia? Or will the power disparity only grow and broken capitalism be used against us?
There's a lot more outside of software development.
It's mystifying to me why software - where bugs or weird behaviors are often obvious to users - is considered easier to replace with these tools than, say, product management (is this accurate research about what features would drive what revenue? who knows even today! is this summary of customer feedback accurate? nobody's gonna know if it's not!), marketing, middle management, or any number of line-of-business "inspect stuff and pass it along to the next person on the chain" roles.
So much of those roles are basically "how can we turn people into process-following machines for consistency and predictability" and intentionally de-valuing intentional reasoning. Adding natural language processing of the inputs/outputs + deterministic rule systems on what to do is basically the entirety of replacing them.
>“what do we do when we’ve successfully vanquished work itself?”
How does automating white-collar paper-pushing get us over the hump here? I think that would require a combination of human-level reasoning + extremely good robotics and automation, at the self-servicing level.
Which could be coming.
But until it happens, I think the real question around the "future of work" are more around what the future of exchange is. It doesn't necessarily have to look like 40 hour work weeks, which is a pattern that arose from pushback against industrialists trying to get ever-more hours out of people to watch the automation machinery and run the factories etc.
I think the interesting, non-apocalypse options fall into two categories:
1) what if you don't need white-collar knowledge factories anymore but smaller groups of individuals can establish markets between themselves with much-more-personalized services and solutions. Can you reverse the traditional economies of scale? Running SaaS as a megacorp requires a certain amount of hard-to-iterate-in-a-try-until-the-tests-pass decisions and expertise since you're trying to be everything to everyone 24x7. If you can provide a more customized version of the same service to a smaller group of people with much less operational overhead for even just 1/10000'th of a giant SaaS vendor's current revenue, you might have something (1/10000 of Salesforce's 40B revenue, for instance, is quite non-trivial).
2) what will people want when everything they spend heavily on today is commoditized? This is the historical path, and why we've never vanquished work in the past. Either ways to get an edge over a competitor by doing more for less, instead of the same for less; or improvements in essential goods and services (you don't need million dollar cancer treatments before they exist, but once they do you don't want to opt out); or in new novelties, keeping up with the Joneses. Showing off. Etc.
That is begging the question pretty hard.
You give 2 systems 2 seemingly different problems. They start building/updating tools and theories and their internal world model starts diverging. As we already see with human specialists. Now imagine 2000 different systems reality diverging. Keeping them all in sync becomes more work than the actual prob they work on.
The burden of proof seems like it's actually on you to show why this current tech boom is any different from previous ones.
As to the answer to your question, well just look at what people are doing at their jobs today. Those are what replaced jobs from 2 years ago.
In past tech advances, there was always a long list of things that could not be automated by a machine.
For some visions of "AI", almost by definition there is nothing left in that list.
(Except some things, of course, where people want to interact with a human).
Still waiting for mesopotamian farmers to imagine what all these workers will be doing after "the plough" automates farming.
>All I ever see is basically a religious belief that surely something will arise.
I would rather see someone describe why Jevons doesnt apply to this one specific instance, without using big scifi imagery.
>I get that previous automation waves have lead to new jobs in the past, but automating knowledge and intelligence is just fundamentally a different thing than anything we have automated in the past and what exactly stops AI from taking those new jobs too as soon as the need for them is recognized?
Its yet to even be demonstrated that AI is going to take many/any jobs. Its yet to be demonstrated that it can.
There was a great post here a few months ago showing how various technologies flagged as AI have a tendency to cap out at roughly 110% human capacity. And even for 110% there's no guarantee that this will be cheaper than hiring 1.1 humans instead of paying API fees. Theres also finance considerations, capex vs opex for deploying local models etc.
The last time I felt like this, I was trying to talk other crypto speculators out of buying every hyped up shitcoin. AI at the moment feels like it is running very much on hype about future potential that has little to do with reality. I hate to use the term bubble, however.
I was dealing with a support LLM just today, and all it did was give me nonsense to do before escalating. The human it escalated to immediately asked for the same nonsense. I just dont feel like reality has been as impacted by AI as people suggest.
Please point me to what I asked for, an example of what this magical new job actually looks like. What does a person with that job do on a day to day basis, in very broad terms. I didn't spot anything like this in the paper.
> And they even close the section by acknowledging that “this time it’s different” might actually be true with AI, and that we simply don’t have enough data to say for sure!
The paper is basically a round-up of three different viewpoints. One of them (Acemoglu) I'm much more in agreement with than the others, and its his contributions you are referencing above.
The quote I pulled came from Briggs. He's the one I want more information from on what these jobs will look like and in what ways these new jobs are safe from AI displacement when the lost jobs they replace are not.
Meanwhile Meta has so much AI compute that they don't know what to do with, and they are ready to lease it out. And corporations suddenly want token austerity across the board. OpenAI is delaying their IPO until "next year."
Something is starting to give.
Hey people: not to worry, new jobs will come!
1 + 1 = 3 because growth!
Also Microsoft has been pouring money into AI and it's like not working at all. They might be in trouble.
Personally, I’ve switched back to the track I was on in my twenties: I spent the day doing moderately interesting work in a skilled trade, periodically having to haul 80lb-90lb steel machine parts up two flights of stairs in an un-air-conditioned, 100 degree manufacturing facility wearing heavy protective clothes and tools. I’ve worked in blue collar jobs before so I’m used to it, but I’ve seen a lot of developers say they’re going to be electricians or something. Good fucking luck! I’d conservatively estimate that 90% of developers I worked with are too soft, either emotionally or physically, to complete the first week of a years-long trade apprenticeship. The starting salary is only marginally better than working at Walgreens, and you’re expected to do a shitload of grunt work that is usually done by 20-year-olds. I foresee the entry-level retail, DoorDash, uber, instacart, et al scenes getting a whole lot nerdier over the next few years.
A GS report - including discussions with MIT labor economists DAren Acemoglu (Nobel), Neil Thompson
After all, you're not a "software engineer" anymore, but an "intelligence engineer"! You can do more with less! Meanwhile, the work is identical and hours are tracked more vigilantly.
1. As a somewhat self-denying but admittedly obvious "tech bro", I can't help but cringe at some of the tech bro podcasts claiming that AI is not going to mass replace jobs. After a previous round of layoffs at my company due to financial constraints. We've managed to not just replace the bulk of our developers withy Claude, but seriously outperform them and build some serious, real-world applications in very short order. These were applications that touch highly sensitive databases across multiple departments in our company. I cannot see where all these engineering jobs are going to go if our small company was able to replace our engineering team with one 4X smaller and get better output.
2. These same podcasters claim that "this is why people should be in trades and doing skilled labor, such as electricians and plumbers." We have to be fools to think that AI + robotics is not going to replace all this very shortly. For the longest time I said it would be about ten years, now, but seeing how my unbogglingly fast Claude got so good (i.e., just compared to last year), I find it hard to imagine that this would be less than three years out. I personally think that these "tech-elite" podcasters are trying to brand a message so that people won't be afraid of AI, but that none of them believe this message themselves.
3. For a tech-first forum/news site, I'm shocked by how anti-AI the sentiment seems to be here. I too am concerned for jobs, but that doesn't change how bullish and excited I am about what I am currently able to build and will be able to build going forward using AI. You may call that selfish, and I won't disagree, but I think it's also intrinsically tech-first, to have this viewpoint. I have no issue with somebody not being tech-first but complaining on Hacker News about a tech-first reality. It sounds disingenuous to me.
https://people.duke.edu/~charvey/Teaching/BA453_2004/Street_...
Well, living in an AGI world was sci-fi until arguably, we found ourselves in one post 2022.
What examples are you thinking of?
Mainly what I remember is multiple waves of previous apocalypses not materializing. Off the top of my head:
- programming will contract because of higher level languages, you won't need nearly as many people
- outsourcing will kill programming job wages because one person can write the spec and ship it off to a team that will send back the working code for 1/10th the wage
- better modeling methods will kill programming job wages because one person can write the spec and the machine will generate the code based no it
Obviously that doesn't guarantee that this time will be the same. But you seem to be making a much stronger claim about what history tells us in the opposite direction...
>(Except some things, of course, where people want to interact with a human).
And, of course, the things that the person holding the vision thinks they personally will still be doing. ;)
You’re probably right that many jobs can eventually be done by AI but not on day 1. Between Day 1 of the brave new AI world and Day N, humans will be in the loop doing stuff that other humans will try to automate away ad nauseam. On Day N+1, there’ll be yet another HN question asking what exactly humans will be doing in the Braver Newer AI World.
Every need of the common people was met and their happiness was catered for. And those who were unusually independent, intellectually curious, creative or couldn’t fit in were “exiled”, to a place with their equals, where their were free to continue their studies, art, etc.
Imagine a world where the MAGAs are happy and content instead of angry.
They threw Copilot at every single thing they could think of with no discretion and now Copilot is the worst brand in all of AI. They don't even realize this yet and are still dumping money into the Copilot hole.