Learning Library

← Back to Library

AGI, Job Loss, and Paradoxes

Key Points

  • The speaker defines artificial general intelligence (AGI) as an AI system that can perform virtually all economically valuable work, noting that current chatbots are far from this level.
  • While many fear that ubiquitous AGI will cause total job loss and push societies toward universal basic income or token‑ownership models, the speaker argues this panic overlooks the nuanced ways AI will affect different occupations.
  • Existing studies on AGI’s economic impact are criticized for treating the technology as a single, interchangeable variable, ignoring the “ragged edge” where AI performance varies across job families.
  • To properly anticipate AGI’s consequences, the speaker highlights the need to incorporate economic concepts like Jevons Paradox (increased efficiency leading to higher overall consumption) and a second, less‑known “Morx Paradox.”
  • Recognizing these paradoxes suggests that greater AI productivity may actually expand demand for certain services rather than simply eliminate work, challenging the assumption of inevitable mass unemployment.

Full Transcript

# AGI, Job Loss, and Paradoxes **Source:** [https://www.youtube.com/watch?v=053O3UkfC3k](https://www.youtube.com/watch?v=053O3UkfC3k) **Duration:** 00:11:54 ## Summary - The speaker defines artificial general intelligence (AGI) as an AI system that can perform virtually all economically valuable work, noting that current chatbots are far from this level. - While many fear that ubiquitous AGI will cause total job loss and push societies toward universal basic income or token‑ownership models, the speaker argues this panic overlooks the nuanced ways AI will affect different occupations. - Existing studies on AGI’s economic impact are criticized for treating the technology as a single, interchangeable variable, ignoring the “ragged edge” where AI performance varies across job families. - To properly anticipate AGI’s consequences, the speaker highlights the need to incorporate economic concepts like Jevons Paradox (increased efficiency leading to higher overall consumption) and a second, less‑known “Morx Paradox.” - Recognizing these paradoxes suggests that greater AI productivity may actually expand demand for certain services rather than simply eliminate work, challenging the assumption of inevitable mass unemployment. ## Sections - [00:00:00](https://www.youtube.com/watch?v=053O3UkfC3k&t=0s) **Untitled Section** - ## Full Transcript
0:00we are going to do an entire post on job 0:03loss and fears of job loss with 0:05artificial general intelligence buckle 0:07up your seat belts this is going to be 0:09very in-depth so first I'm going to 0:13start with a definition of artificial 0:15general intelligence that open AI is 0:17using I think it's 0:18useful roughly speaking if an AI system 0:23is widely deployed and is capable of 0:25doing almost all valuable work that 0:28humans do we should call it an AR arcial 0:30general intelligence by the way that 0:33includes yard work it includes car 0:36repair it includes the physical services 0:39that humans do that are economically 0:41valuable that AI is not close to 0:43touching 0:45today by that definition there is no way 0:48the chatbot that is on your laptop is 0:52close to artificial general intelligence 0:54so we will start 0:56there what happens in a world where 0:59artificial general intelligence truly 1:01becomes ubiquitous and we could even 1:03give ourselves an easier bar what 1:06happens if it can't do physical work but 1:09knowledge work it just becomes really 1:11good at doing most economically valuable 1:13work I have seen the studies I have seen 1:17doomers on YouTube I have seen panicked 1:20people in the Tik Tock comments on my 1:22Tik Tok Channel they all essentially say 1:25the same thing if we were to get AR 1:29artificial general 1:31intelligence we would all lose our jobs 1:34there would be no economically valuable 1:36work left to be done there would be 1:38widespread unemployment and the only way 1:41to prevent massive societal disruption 1:44would be Universal basic income or 1:48perhaps tokenomics where you are all 1:50sort of tiny investors in the AI that is 1:52doing the actual work of 1:55society I 1:57disagree I have disagreed for a long 2:00time but I don't think I've stated it as 2:03plainly as I am here and the reason I'm 2:06stating it really clearly is because I 2:08am tired of seeing study after study 2:12that treats the most crucial technology 2:15we are likely to see in our lifetime as 2:18if it were a 2:19commodity as if it were something that 2:22is a single variable to plug into a 2:24study AGI the variable you plug it into 2:27studies and then you can like work your 2:29math equation and you can see what 2:31happens that is not how actual 2:34artificial intelligence rule out is 2:36going to go we know it's going to be a 2:39ragged Edge we know that AI That's 2:41relevant for particular job families is 2:44going to look very different we know all 2:46of that already why don't our studies 2:48cover that and I'm not done yet that is 2:51just one initial critique of where the 2:53studies are falling down but there are 2:55two much more foundational pieces that 2:58the studies aren't taking account of of 3:00that we desperately need to fully 3:03understand they are jevans Paradox and 3:05morx Paradox and we are going to talk 3:07about both because they are crucial for 3:10understanding this moment jevans was an 3:14economist thinker in the 19th century he 3:17was writing about coal Coal at the time 3:19was very valuable and if coal became 3:22more abundant the thinking went that 3:26demand would stay flat like you have 3:29more coal there's only so much you can 3:30do with coal you can burn it but like 3:33how much use do you really have probably 3:35like it will become worth 3:37less jevans observed that that wasn't 3:41true as the abundance of the commodity 3:45increased demand for that commodity rose 3:49that is a foundational insight into how 3:52the way humans interact with technology 3:57works I want you to think about the 3:59internet 4:00we had I kid you not an original 4:03application for the internet that was 4:05watching a coffee maker to see if it had 4:07coffee in it you can look it up have we 4:10found other things to do with the 4:11internet since then we have we are 4:16really really good at finding new 4:18utility when supply of something useful 4:21grows another example right now 4:23currently is renewable energy we are 4:26producing so much renewable energy 4:28everybody's ejections keep breaking it 4:31looks like a vertical line it is 4:33absolutely insane how much renewable 4:36energy is being produced every year 4:38nobody can get the projections correct 4:40everybody keeps predicting a tapering 4:42off and it's just not happening because 4:46we keep finding more use for renewable 4:49energy it's jeevan's 4:51Paradox but I have not yet seen any 4:56studies that actually take account of 4:58jev's par Paradox when something like 5:02intelligence becomes cheaper and more 5:04abundant and you would think that would 5:06be a relevant thing to talk about 5:09because 5:10anecdotally I am seeing it happen all 5:12over the place this is not just for me 5:14by the way this is what I've observed 5:16working with dozens of people over the 5:19last few years hundreds of people 5:20speaking to many 5:22people when we use chatbots at work in 5:25our personal lives they are not one for 5:28one replacing 5:30people they are actually doing work that 5:33would not get done otherwise we are 5:34living out jeevan's Paradox I give my 5:37chatbot things to do that would just not 5:40get done otherwise like there nobody 5:42would do it it just wouldn't 5:45happen and yet we don't take account of 5:48that when we do our studies on the 5:50projections for the future of artificial 5:52general intelligence like it makes no 5:54sense like we need to fully load in the 5:57idea of jebin's paradox to really 5:59understand what the future looks like 6:02and it's one big reason why I'm more 6:03bullish on the future of jobs than a lot 6:06of the other people talking about AI 6:08right 6:09now Paradox number two morx paradox this 6:12one specifically about computer systems 6:14in AI a few decades ago moravec observed 6:18that it is very very easy to teach a 6:22machine to do things that humans find 6:24difficult and very very hard to teach a 6:28machine to do things that find Easy A 6:31few examples chess I find it hard to 6:34play chess even though I really enjoy it 6:36I still remember as a kid when a machine 6:39named deep blue from IBM beat Gary 6:41Kasper off I was so 6:44impressed it was relatively easy for the 6:47machine and we've since built machines 6:48that are even better at chess and now 6:49they've solved go or almost solved go I 6:52don't know the point is these tasks that 6:54humans find really hard are things that 6:57machines find easy and you can find 6:58numerous other examples like 7:00that on the other side humans find it 7:04relatively easy to walk most of us most 7:06humans find it relatively easy to catch 7:08a ball these are things that machines 7:11find really really hard and I'll go into 7:13knowledge work because you might say 7:15well this is all physical stuff right 7:16like what are we doing with knowledge 7:17work humans find negotiating politics 7:21and stakeholder Management in the 7:22internal people dynamics of a business 7:25relatively easy now some of us are 7:27really good at it some of us are okay at 7:29it some some of us we kind of know are 7:30not great at it but we know it exists 7:33and our Baseline level of fluency far 7:36exceeds what you could do with a machine 7:39and it's not that hard for us we don't 7:41really think about it a lot we are 7:43basically taking the social dynamics and 7:45cues we learn as small children in 7:47family and social situations and we're 7:49applying them in the organization 7:50outside the scope of this YouTube the 7:53point is we find it easy the machine 7:55finds it hard it's more of X 7:58Paradox so so think about it right now 8:02is chat GPT better than me at 8:05remembering every single fact about 8:08product management yes it is and I've 8:12been doing product management a long 8:13time it's unquestionably better than me 8:16at remembering all of the 8:19facts but would I hire chat GPT to 8:22replace me 8:24no because all of the other things that 8:28go into more of 8:30Paradox it can't do it's not close to 8:32doing it can't have a complex 8:35conversation about timing another 8:37conversation relative to a particular 8:40event that's happening in the calendar 8:42so we maximize Team Dynamics it can't 8:45have a conversation about how to best 8:48align the constraints we face in a sales 8:50environment with value propositions and 8:52deal rooms and figure out what that 8:54means for what we build next it can't 8:57even agentically proo protype without a 9:00lot of help right now will it get better 9:03at some of that it 9:05will but my point stands more ofx 9:08Paradox means that things that we find 9:12easy including in the work environment 9:14are hard for systems to learn and if 9:16they're hard to learn they're hard to 9:18learn well they're hard to learn in an 9:21AGI sense in the sense that it would 9:23cover most economically useful work 9:27because that requires a very high degree 9:28of 9:30completeness so yeah I think morx 9:33Paradox matters for AI I think Jin's 9:36Paradox matters for AI I have yet to see 9:39any projections of the future of the 9:41labor market take these really seriously 9:44and I'm not saying I have all of the 9:46answers I'm just saying if this is a 9:48transformative technology it deserves to 9:50be studied carefully and specifically 9:53and not just as one more variable thrown 9:56out in a YouTube video for clicks and 10:01and I'm tired of that and I'm tired of 10:03studies that project off of flawed and 10:07incomplete assumptions I think we can do 10:09better I think this technology is 10:11important enough that it deserves a 10:13better 10:14look and so if you are worried about Ai 10:18and 10:18jobs if you are in uh an economist role 10:22this is hopefully some fodder for 10:24thought if you are worried about Ai and 10:26jobs and you're just a white collar 10:28professional maybe this is encouraging 10:31but I also hope it's something that you 10:33reply to that you share that you save 10:35because from what I've seen this 10:39question that I am spending 10 minutes 10:41talking about is the question everyone 10:44has under their breath and no one really 10:47wants to talk about except by saying 10:48Doom and Gloom it deserves a wider 10:50conversation we deserve to have a more 10:52nuanced conversation about what 10:55assumptions hold true as AI systems grow 10:58in capab 10:59ility unless you think that I'm the only 11:02one who thinks about this I'm going to 11:05call out something that's a small hint 11:07that open AI themselves may be thinking 11:11a little bit 11:12more thoughtfully about how long they 11:15have to keep employing people they have 11:18recently changed their policy this was 11:20in the summer around how they handle 11:24stock grants for people who choose to 11:25leave the company and long story short 11:28they are making it easier for people who 11:30leave to continue to get value based on 11:32their tenure and the equity that's 11:34vested in other words they are expecting 11:38a future for a long while to come where 11:41people will be employed at open AI for a 11:43period and then leave that is a normal 11:45employment 11:46pattern they are not expecting the end 11:48of all things and if they're not 11:50expecting it why are we