Artificial General Intelligence (AGI), Artificial Super Intelligence (ASI) and fully general humanoid robotics are just around the corner, or so many people believe. So it’s time to try to understand how this will affect our economy. Will we be forced into lives of idle leisure and/or meaninglessness? Will the few remaining human workers toil below the API? Will we get fully automated luxury gay space communism?
This is a follow up after five years to my original post on post scarcity and post-capitalism.
Keynes predicted in 1930 that by 2030, automation would reduce the need for work to just 15 hours per week. We’re almost there, so what did he get right and wrong? First, most people work fewer hours in less physically taxing jobs than their grandparents did. But we’ve standardized on 40 hour weeks, and much more for people in jobs with a high degree of competition. Within those 40 hours, I am reliably informed, some people struggle to perform even a single hour of meaningful work, but these are not the rule.
On the other hand, Keynes was absolutely correct that cumulative capital increase would lead to radically higher GDP and standard of living. If anything, he underestimated how resourceful we could be in finding new things to make and spend money on.
This doesn’t surprise me much. Even in Keynes’ time, mechanization of much of labor had increased US GDP by a factor of four or five since the end of the civil war, and probably a factor of 20 since pre-industrial times.
With modern technology, we could keep our entire civilization at a c. 1700 level of abundance (i.e. mostly not starving to death, but no real health care or other tech, life expectancy of 35 due to tragic levels of child mortality) with almost trivial allocation of labor – but for the necessity of some level of economic dynamism to maintain the tech and capital stack that sustains today’s level of mechanization.
We stand on the cusp of another industrial revolution, poised to increase GDP again by many times over. Maybe this time we’ll finally run out of things to spend money on and be forced to retire? I doubt it. Let’s get specific.
Large language models and other proto-AGI technology functions as a tool. Like other tools, it increases productivity for workers in the economy, but to varying degrees. It seems highly likely to me that while nearly everyone will benefit substantially from widespread use of AGI, functioning as a cognitive prosthesis, a tiny minority of power users will increase their personal productivity not by a factor of two or three, but by a factor of hundreds or thousands. Even if they are unable to capture all the financial benefit of this productivity increase, it creates potential for out-of-equilibrium economic shifts.
First, let’s divide goods and services into four quadrants, by whether demand is saturable or unsaturable, and whether supply is rivalrous or non-rivalrous.
Rivalrous, saturable goods include housing, food (pre agricultural revolution), education (credentialed), and regulated childcare.
Rivalrous, unsaturable goods include untaxed land, healthcare, education.
Non-rivalrous, saturable goods include LCD displays as an example of the sort of luxury consumer electronics which have become universal, high quality, and cheap. It also includes cars and food (post agricultural revolution).
Non-rivalrous, unsaturable goods include software, aviation, inference compute, and robot armies.
This model explains why, despite our growing wealth and technological power, some goods’ costs are increasing relative to inflation.
That is, the recently nonrivalrous goods have become so because technology has greatly increased their supply elasticity, essentially overcoming scarcity in favor of abundance. On the other hand, growing buying power accruing to the highest productivity workers can bidding everyone else out on the unsaturable rivalrous goods, creating the effect of increased scarcity despite a broadly shared increase in overall wealth.
We can expend near infinite effort on complicated bureaucratic shell games to ration scarcity, which is itself often artificially induced and then exacerbated through this sort of regulation. We can tax gains into oblivion, obtaining equitable outcomes through more universal poverty and cutting down the possibility of out-of-distribution upsides for our entire civilization, until we’re outcompeted by a more forward thinking alternative from some other culture. Or we can let technology continue to lift the burden of scarcity, allowing supply to increase in tempo with our increasing civilizational wealth, ensuring that everyone sees benefits to generally increased prosperity.
Healthcare, education, childcare, housing are poster children for scarcity-through-over-regulation, credentialism, etc. But this doesn’t really provide a road map for how our civilization can soften, for example, the looming demographic catastrophe, other than “I wish we could deregulate zoning and permitting”. Even though I wish we could.
But it does offer some ideas for how AGI could change this landscape. In particular, if a relatively small class of AI superusers grows enormously in economic power, we should expect their consumption of rivalrous goods to drive cost increases. The nightmare scenario here is that San Francisco’s deliberately scarce housing supply gets further bid up by a generation of AI billionaires to the point where a crumbling pre-war tenement costs not $5m but $500m, and any prospect for anyone not in the cyborg class to live in that city goes from fantasy to myth. Demand for essential consumer goods, like a place to live, will drive inflation up if supply cannot increase in step.
So, how do we fix this? How to we transition the rivalrous goods into the non-rivalrous quadrants?
Historically, food was a saturable, rivalrous good. That is, you could only eat so much of it, but once you had, no-one else could. We solved food scarcity and famine with Haber-Bosch and artificial fertilizers. Food is now abundant enough that your purchase of whatever you want at Costco does not meaningfully reduce my odds of purchasing whatever I want at the same store. This is an instructive example for how technology can fundamentally alter some previously guaranteed constraint on our lives. In this case, within living memory food shortages were simply a fact of life and every winter, some fraction of people didn’t make it. Today, naturally occurring famine has been eradicated.
Healthcare is rivalrous and unsaturable, because doctors are in finite supply but one sick person can consume unlimited quantities of their time. In practice, this always leads to rationing, usually with some enormous extractive bureaucratic superstructure to try to obscure this fundamental truth. If some fraction of your society are doomed to suffer slow declines in old age, it is trivial to spend orders of magnitude more wealth on unpleasant medical interventions with poor outcomes than the patient could ever have contributed in taxes.
How can technology address this? One approach could be AI doctors. If you can spool up Grok MD for $0.10/hour, the scarcity of doctors goes away and healthcare becomes unsaturable but non-rivalrous. Every doctor I know can think of half a dozen malingering hypochondriac “frequent flyers” who consume much of their time to the detriment of patients with real conditions, and who could be equally well treated (in terms of their unsaturable need for attention) by a friendly AI.
But this hardly addresses the core problem, which is that most of healthcare is consumed by old, sick people in the process of dying. If we invent immortality technology, then healthcare goes from rivalrous and unsaturable to non-rivalrous (drugs are easy to make) and saturable (once you’ve obtained perpetual youth there’s no way you can do it again).
Land is rivalrous and unsaturable. While housing is a relatively discrete need, an army of AI-rich people attempting to buy up estates in California will quickly run out of water front. What to do about this? We could tax the entire industry into oblivion, but this will hardly stop new Chinese billionaires from showing up with the same plan in mind. Conventional wisdom would call for managed scarcity through Georgism. Tax the land to incentivize value-aligned use. This would revert it to a saturable good.
But a tech-based approach might call for the creation of more land through mass-desalination and irrigation, particularly of the US south west. Making land in space accessible also provides a pressure relief valve. This means that land is still unsaturable (as ideally it should be in a world with open frontiers) but is no longer rivalrous.
Finally, housing is the easiest of all. Upzone and build. The core problem with alleviating housing scarcity is engineering a soft landing for the landlords who have hoarded the artificially scarce supply, building their wealth in the process. It is unfortunate that we’ve chosen to build middle class wealth on home ownership rather than, say, robust equity in highly productive and fast-growing primary and secondary industry. But the presence of AI-led GDP hypergrowth should soften the blow.
Indeed, periods of rapid economic growth provide a cornucopia of opportunities for new companies to provide new products to satiate new desires. If AI is increasing civilizational productivity by some substantial margin, there is margin to be made by helping consumers to increase their consumption.
The goal of policy should be to help our civilization ingest ongoing technological improvements without creating an artificial class of “have nots” who miss out, especially when there is so much wealth creation to go around and so little standing in the way of this rising tide lifting all boats.