AI Jobs Warning: 36 Hours Later, Author Interviewed, Paper Analysed in Full, and Why I am Still Somewhat Optimistic
Added 2024-03-28 23:25:52 +0000 UTCYesterday’s dramatic Bloomberg headlines showcased an ‘AI Jobs Apocalypse’, warnings of ‘millions of jobs lost in next 3-4 years’, triggered by a new 44-page paper from London. I interview the IPPR lead author Carsten Jung and get to the bottom of it all, giving my critiques of the paper, where I agree, why I am optimistic and hopefully show you that you should always look deeper than the headlines. The future of jobs in the age of AI is too important a topic not to.
Jobs Report: https://ippr-org.files.svdcdn.com/production/Downloads/Transformed_by_AI_March24_2024-03-25-172800_bvyk.pdf
Guardian Article: https://www.theguardian.com/technology/2024/mar/27/ai-apocalypse-could-take-away-almost-8m-jobs-in-uk-says-report
MIT AI Job Automation Analysis: https://www.imf.org/en/Publications/fandd/issues/2023/12/Rebalancing-AI-Acemoglu-Johnson
Hume AI Demo: https://www.youtube.com/channel/UCMr20KsVhsYBar3AsCvBcsw
Altman Inequality Clip: https://www.youtube.com/watch?v=WHoWGNQRXb0
Bold Predictions Time Magazine: https://www.reddit.com/media?url=https%3A%2F%2Fpreview.redd.it%2Fbold-prediction-in-the-60s-makes-you-wonder-what-bold-v0-apeizv4ue3rc1.png%3Fauto%3Dwebp%26s%3Df056dea85ebdbfd9db69eff4e2e90628e7599b01
Comments
Wow, thanks. Another example of the case of corrections spreading slower than misinformation. I only heard about the judges and lunch. On the other hand, people are not as objective as we want them to be, that part holds true.
Exponential Age
2024-06-03 07:02:11 +0000 UTCI like the care credit!
Joshua Davis
2024-05-14 18:34:20 +0000 UTCCosts of labor decreasing also decreases profits in the long term
Alexander Berezin
2024-04-05 14:06:09 +0000 UTCTax breaks for using humans instead of AI in certain jobs is a very good idea and no doubt we'll see it soon enough. One concern: Who will pay for the tax breaks? If all work is ultimately done by a handful of companies in California/ Seattle, where does the tax revenue come for everyone else?
Martin Percy
2024-04-03 17:43:27 +0000 UTCWhat do you mean Benjamin? Alternative to what?
Adeoxymus
2024-04-03 04:57:23 +0000 UTCWhy would we want to encourage and subsidize use of inefficient resources? Even if one did, at what rate do we imagine that labor would be able to earn? Under future assumptions, there would be limitless availability of human labor, and that inefficient resource would be subsidized by a tax (effectively, as the tax is just directly routed to its endpoint instead of going through the government first). Why not skip the inefficiency and just provide the subsidy directly to the human? Instead of leading to a dystopian future where the rich who own all the substrates of the new wealth (e.g. Compute, land, energy, robots) also accruing ALL the wealth generated from that capital, with something like Universal Basic Income we can provide the subsidy but still avoid using an inefficient resource, potentially leading to a new renaissance. Why should people HAVE to work? In a world where we have machines that are far more efficient than us at almost every type of work, why not accept the benefits of that technology in the form of putting everyone out of a job, and then allowing them all to pursue whatever avenues interest them. If you WANT to have a job as a hobby, by all means go ahead, or if you want to paint or sing or read and learn, do that. In short, do whatever your heart fancies, instead of being required to do a thing simply for the right to live. Historically, this has not happened. We continuously accrue an increasing percentage of the wealth to the wealthy. Even still, the poor have become richer in an absolute sense, even as they have a relatively smaller and smaller share of the ever increasing pie. This paradigm shift will require a new model however, because eventually, for the first time in human history it will be uneconomically sound to employ a vast percentage of the population. The problem will be too large to sweep under the rug. "Go get a job" will ring even more hollow than it does today. I haven't thought of a better solution than UBI, but I'm open to hearing one.
Ali Eslami
2024-04-03 03:31:35 +0000 UTC"The problem with x% of jobs are at risk of automation is that no job just consists of an accumulation of tasks some of which can be automated." - What else then? Sounds like a working mental model this far at least, whats your alternative?
Benjamin Eidam
2024-04-03 00:10:45 +0000 UTCSubsidizing worthless things is usually a bad idea. In the limit leads to an eternal high-school where the “proletariat” does make-work for eternity. Just give people some capital. Governments own a lot of land. If property rights are enforced long term, just give every human an equal share of the moon or something and that should be enough to compute billions of experience-machine years for each citizen.
Edward Huff
2024-04-02 19:55:24 +0000 UTCHow to prove human labor though? Is one security guard enough to count as human labor?
Adeoxymus
2024-04-02 10:11:18 +0000 UTCThat's an interesting idea. I would love to see a detailed proposal. I am not an economist, so my remarks are based on general knowledge. I just think that there would be some transition period when some fraction of work is still made by humans. We would need solutions for that time. And what if a company with human labor has a non-human labor companies as some of subcontractors or spends some revenue on buying products/services from such companies? I am afraid that being a human-labor company wouldn't be zero-one. And effectively buying more expensive product/service from human-labor company to get tax break would IMHO work similar to the state subsidizing the labor which would make human-labor companies cheaper.
Jan Matusiewicz
2024-04-02 00:27:13 +0000 UTCAs the unit value of human output falls to approach zero, the wages of subsidised humans necessarily falls to approach the value of the subsidy. Taxing earnings from a subsidy alone does not produce enough to pay for the subsidy, let alone a UBI as well. This leaves humans doing unnecessary work for minimal pay, in a situation reasonably described as drudgery.
Poss
2024-04-01 21:07:57 +0000 UTC100% agree. And in fact I almost feel like Judges could be an earlier use case for AI than some more subjective tasks like writing a great novel or even having a genuinely witty banter relationship with a person over time. Just because the law is supposed to be very rule based and there is a lot of data on judgements over time and the benefits of even a small increase in average 'fairness' of decisions is a huge gain for society. Like self driving it doesn't have to get much beyond human performance before you'd ask yourself, as you said, are we choosing a human bias of inferior performance at the cost of actual human lives just because we like the idea of humans doing activity X.
Tasty3141
2024-04-01 10:18:12 +0000 UTCCould try, yes! I am contacted by media organisations from time to time, even the BBC
Philip
2024-04-01 10:05:14 +0000 UTCWell UBI could also be viewed as giving money to worthless activity, depends on your framing. Subsidising human labour for a while at least eases the transition of derived dignity from labour to intrinsic worth, which will have to come at some point.
Philip
2024-04-01 10:04:50 +0000 UTCThanks Yuri!
Philip
2024-04-01 10:03:06 +0000 UTCIndeed! But you don' hear that in Bloomberg.
Philip
2024-04-01 10:03:00 +0000 UTCNot sure about the forcing part! This would sustain a tax base, one that's politically palatable, which can they be used to fund UBI for example. And the jobs I am speaking of need not be drudgery, in fact I would anticipate that if a company is funded regardless of the job their human workers are doing, they might as well let them do enjoyable, customer-facing roles - no need for back-office drudgery. And yes, I believe humans have innate value, separate from whether they have economic value in the future. My idea is to harness the market to economically reflect that value, which for me is there whether or not human labour itself is efficient.
Philip
2024-04-01 10:02:42 +0000 UTCGreat points but I am positing a scenario in which no human labour can compete effectively in entire industries/categories. So even removing income tax/national insurance would not incentivise human labour, necessitating this additional scheme. And my worry about a simple AI tool tax is that companies will be incentivised to off-shore, whereas if they could avoid fees entirely by using some human labour, they wouldn't have a need to leave.
Philip
2024-04-01 09:59:31 +0000 UTCWould have to be piloted. I think it might do better than an unfunded UBI, have you noticed OpenAI have not released the results of their pilot UBI, despite saying they would a while back?
Philip
2024-04-01 09:57:35 +0000 UTCYou occupied the top 3 spots simultaneously, incredible.
Philip
2024-04-01 09:56:52 +0000 UTCI actually read that book a year or so ago! It's great. And on my tax idea, it would be less about proving AI use, which is hard, and more about proving human employment, which is easier as we already have the systems in place for verifying that.
Philip
2024-04-01 09:56:27 +0000 UTCGreat insight, thank you.
Philip
2024-04-01 09:55:23 +0000 UTCThe quality of the code is key, but yes, once it passes various quality-thresholds, and is coherent across long contexts, then the volume will be unbelievable.
Philip
2024-04-01 09:54:26 +0000 UTCI take your point Leander but we would need some tax base to subsidize in that manner. I think the proclivity toward tax evasion would be reduced if you simply don't have to pay the tax if that slice of your income goes to human-labour consumption.
Philip
2024-04-01 09:53:47 +0000 UTCThe acceleration of change, which you describe Fetian, is the core point that needs more addressing. And a deeper question too: will there ever be a decade in the future that has less change than the current one, and if so, how?
Philip
2024-04-01 09:52:18 +0000 UTCYes, I am very much following the participation rate, it is one of the best indicators of what we are discussing! Thanks Gabor
Philip
2024-04-01 09:49:56 +0000 UTCTaking the broader point though, I am not certain that AI will end up being less biased than human sources, in the medium to long-term. We might have to 'be biased in favour of bias' to keep human judges.
Philip
2024-04-01 09:49:21 +0000 UTCThe problem with x% of jobs are at risk of automation is that no job just consists of an accumulation of tasks some of which can be automated. That doesn't mean that there is no risk but the discussion is more nuanced and a mix of optimism and pessimism, a good book on that topic: A world without work by Daniel Susskind, a Brit btw ;-) https://www.danielsusskind.com/a-world-without-work As for policy decision, complicated tax schemes like you suggest just seem like a hard way to control and would most likely miss the target. For example how to define if it uses AI? Is using a chatbot for some internal Q&A considered AI? In the long term I suspect we simply need to move away from taxing labor and start taxing capital and big business.
Adeoxymus
2024-03-30 08:45:32 +0000 UTCBeware of using thinking fast and slow. Much of the book does not hold up to scrutiny/replication. I think the hungry judges is one of those that did not replicate, and the results could be explained by the judges taking up different type of cases before or after lunch. Hence the difference. Small video on that topic: https://www.youtube.com/watch?v=Qgr1uBbNQ4w
Adeoxymus
2024-03-30 08:22:31 +0000 UTCWill GPT5 allow a user to produce a 4000lines of code application within an hour? For me, GPT3 could do 40lines, GPT4 can do 400lines, on an exponential curve that means GPT5 will be 4000lines. Thoughts? If true that's a lot of jobs being created, removed, and displaced 🤔.
GGuy
2024-03-30 02:11:03 +0000 UTCRegarding the human-deductible AI tax: It would be interesting to put this to a test, experimentally. Can we simulate this (or various other) scenarios? Instead of levying an AI tax (which naturally could be evaded), I believe it's more feasible to drastically reduce (subsidize?) common goods like food, housing, etc so that people can do more with the same amount of money. From a monetary policy, deflation is to be avoided but I am not sure if this holds true in this case.
Leander Maerkisch
2024-03-29 20:21:23 +0000 UTCAGI is coming, and if we accept that, then all these talks are either trying to dilute the radical transformation ahead into something more palatable (the post AGI society will be just like the current one, but maybe everyone's rich, etc), or flat out don't believe in AGI at all (they think AI is just a one-time gain like past technologies).
Feitian Li
2024-03-29 12:45:59 +0000 UTCBut AGI ramps up exponentially at least, and it by definition is capable of replicating all human labor. This may sound trite, but I want to point out much of human society is built around the fact that human labor constantly turns over. What's the point of teaching calculus 101 over and over again to 18yos if you only have to teach it once to AGI? What's the point of having elections if AGI is capable of administrative tasks and doesn't lose institutional memory? What's the point of training human workers if said workers often forget or refuse to learn, have poor work ethics, and strike or develop antagonistic ideas every now and then?
Feitian Li
2024-03-29 12:42:32 +0000 UTCThe take home lesson from this report is the real mechanism of labor accomodation of new technology is generational, ie children of blue-collar workers becoming white collar workers. Therefore we should expect a large one-time technological advancement to take about a generation to absorb and translate into increased economical output, as was the case for the past 150 years. A corollary of this observation is the ability of human society to accommodate change scale linearly with time, ie twice the time twice the change and so on.
Feitian Li
2024-03-29 12:37:32 +0000 UTCMy critique is not that gpt4 isn't currently smart enough to do the detailed analysis on 20000+ human activities. It's this: all the similar analysis I can get my hands on basically applied a one-to-one replacement framework, ie AGI is a drop-in replacement for human workers. While this is convenient for the author, I believe it's ultimately very off. We're currently not a one-to-one replacement to the medieval serf-and-lord economy, nor are we a one-to-one replacement to the 19th century crude industrialist economy. Eventually even those jobs that requires human interaction and human interaction only will disappear because all other jobs disappeared. How can you have a standalone seniors care sector? The real endgame is either great economical strife the like of which human society had never seen or ubi. And the timescale involved is mere decades.
Feitian Li
2024-03-29 12:33:23 +0000 UTCThank you for covering and critiquing the report. Please be on the lookout for empirically supported lump of labor fallacy fallacies. One possible smoking gun could be the Labor Force Participation Rate (except it does not help). * http://fred.stlouisfed.org/series/CIVPART <br> <iframe scrolling="no" frameborder="0"></iframe>
Gabor Melli
2024-03-29 09:57:37 +0000 UTCI think the hungry judges thing has been either debunked or heavily called into question.
Tasty3141
2024-03-29 09:50:11 +0000 UTCI think you should write an article about your idea in The Guardian or a similar place. Even if your idea will be useful only in a transitional period, or if it will need to be modified to be made workable, it will start a discussion. This is something coming very soon, and the next Labour government should have some plans for it or at least start searching for the solution.
Arek Stryjski
2024-03-29 07:36:53 +0000 UTCThe paradigm of taxation may not be the right lens to consider this through. Our entire Monetary-fiscal framework is effectively a reward function. While it works, it may only be a local minima rather than the optimal economic equilibrium. The Mundel-Fleming trilema for instance is entirely composed of drastic oversimplifications and unaccounted externalities. These data and processing limits are endemic throughout macroeconomic theory. Eureka showed us that machine learning is better than humans at generating reward functions. Is machine learning capable of similar optimisation of a suitably accurate economic simulation? It seems unlikely national currencies, localised taxation, divergent governance systems and traditional monetary policy are the optimal mechanisms to govern the world economy. If the survival of our species is on the line, we'd be wise to not just accept existing economic paradigm as the boundaries of our problem space. Its a cage of our own making, and we can remake it.
Poss
2024-03-29 04:31:35 +0000 UTCenjoyed video thank you!
YuriMassiveCat
2024-03-29 03:42:16 +0000 UTCNot sure about your tax idea, when people only use services of human labor companies because of tax deductions then their labor is actually worthless and you could have equally as well just donate money. Founding a charity to give money for unemployed people would be effectively the same.
Phillip Lakaschus
2024-03-29 03:37:03 +0000 UTCMaybe government should tax AI usage and use the money to pay for the fullfiling and socially useful jobs? For example pay for supporting, teaching and entertaining other people. The key problem is how to wisely tax AI.
Jan Matusiewicz
2024-03-29 02:39:51 +0000 UTCThat prompt was… not good, to put it mildly.
solarapparition
2024-03-29 00:55:30 +0000 UTCThat's pretty close to the 28M jobs lost (give or take) that I predicted as hypothetically possible. But of course, it's infinitely more complex than all that. Great reporting! EDIT: "Ring fencing" or what I call "statutory jobs" (jobs that are required by law) might be one the biggest hangups towards a better future IMHO. EDIT2: On the topic of "real empathy" (affective) vs cognitive empathy, as a neurospicy person I will say that cognitive empathy is still valid.
David Shapiro
2024-03-29 00:53:19 +0000 UTCYour policy idea seems likely to result in forcing the majority of humans into subsistence labor in unnecessary and preventable drudgery. The key ethical question is: do humans have an innate value? This determines if we are intrinsically worthy of food, housing and higher fulfilment. If humans don't have innate value, then subsidising humans is inefficient. The market will replace the entire species, just as it has done with every other species in the current mass extinction. If humans do have innate value, then we should optimise the market to create human fulfilment rather than optimizing for abstract profit while arbitrarily requiring constant proof of worth from humans.
Poss
2024-03-29 00:40:15 +0000 UTCThanks for your interesting and thorough analysis. As of your idea for solution, for one thing - it would be difficult to measure how much less people a company hires because remaining employees use AI. I would need to think how this proposal compares to much simpler decrease of taxation of work in affected occupations plus a tax on AI tools and robots.
Jan Matusiewicz
2024-03-29 00:23:48 +0000 UTCThanks for going the step further. Great review. What could be the next step in further checking / validating the tax policy mentioned in the end?
Adin Softic
2024-03-29 00:10:33 +0000 UTCThird
Dane Wagenhoffer
2024-03-28 23:45:27 +0000 UTC2nd
Dane Wagenhoffer
2024-03-28 23:45:18 +0000 UTCFirst
Dane Wagenhoffer
2024-03-28 23:33:52 +0000 UTC
