Many of the problems with AI the author is using as premises to justify his thesis are already being addressed and working towards a solution for daily.
Nonetheless, even if humans are required, it would be 1/50th or 1/100th less, a sufficiently high enough reduction that the outcome would remain the same.
I disagree with this premise, there is still a lot of work to do. We don't even have space elevators, nanobots, immortality drugs and all other manner of technologies yet, and just like every other technological innovation in history, I assume that we will reallocate people to do said work. Work expands to fill the time and space (and workforce) available.
I've mentioned this before in other AI threads, but humanity has an uncanny ability to use all resources available. As you say, the amount of work will only grow with the increased productivity from AI.
AI threatens to make all human-sourced work redundant. Because AI is really an automated culture machine - in the widest sense. It will eventually be able to innovate, rather than replicate, culturally ground-breaking work in art, science, engineering, business, politics, and media.
What jobs are humans going to do then? All that's left is manual labour - and eventually that will be automated too.
This isn't exaggeration. It's already happening in a very limited way with today's super-crude AI 1.0. AI 5.0 won't be a step change, it will be something entirely new.
But with LLMs we are as close to that as climbing Everest takes us to the Moon. No small feat, but not particularly relevant. After all, all they do is return statistically likely text responses from text inputs, based on a very large corpus of text previously written by humans. Any claim above that is technical hubris and/or funding strategy.
You would have to define work for this to be meaningful. Generally, demand for "work" is unlimited if the price is zero, but demand falls to zero as price increases. Humans are willing to work for at least what buys them "bare necessities", food and maybe shelter (in a madmax-style world.)
In these discussions, it is suspected that AI will "underbid" larger groups of workers than in previous waves of automation.
The unemployment rate at the height of the Great Depression was "only" about 25%. If 3 people doing the work of 4 is already a huge problem, 1 person doing the work of 50 or 100 is massive.
The obvious and oft-repeated response to this would be to point out how automation has been going on for many decades (arguably centuries) and so far people have just evolved into different jobs, which is true, but it seems like a risky bet to me to assume that trend will continue if/when a non-task-specific automation technology has the potential to very quickly cut wide and deep across an awful lot of the job sectors that currently have reasonably desirable jobs.
The speed of disruption, I think, is likely to matter quite a lot.
So your hypothesis is that the Great Depression was caused by a sudden 25% increase in productivity?
If so, what data do you have to back this up?
I think generally it’s hard to think that modern economics has only really exist for the past couple hundred years. The last major shift that created it? The cotton gin starting the Industrial Revolution. AI is akin to the cotton gin.
No, that isn't the hypothesis nor is it relevant to the point I am making.
The cause of the 25% unemployment rate in the Great Depression is actually completely insignificant to why I mentioned it. The point being made with that is to highlight that 1 in 4 people no longer being in a position to make a steady living was enough to cause an economic situation so bad that nearly 100 years later in the USA it's still treated as the common example of how bad things can get when things are Really Really Bad.
The oncoming rush of automation across virtually every aspect of the "knowledge worker" economy easily has the ability to create a situation in which the percentage of people made redundant is an order of magnitude more than that benchmark 1 in 4, and since it is likely to cut across so much of the economy very fast all at once there is likely to be no time for the economy to reasonably deal with.
The go-to advice for people who are automated out of jobs is "retrain and find another one", how do you do that when the disruption is "everything everywhere all at once" and your competition for the new jobs is likewise coming from everywhere all at once and the job contraction impacts nearly every company?
How scary this all sounds is probably colored to some degree by where you live and all of my concern is admittedly biased towards the USA where our current social safety net systems are woefully inadequate for dealing with this sort of disruption (even in the short term) and still constantly under attack.
The cause isn’t irrelevant, in fact, you've reversed cause and effect.
> The oncoming rush of automation across virtually every aspect of the "knowledge worker" economy easily has the ability to create a situation in which the percentage of people made redundant is an order of magnitude more than that benchmark 1 in 4,
So 10 in 4? Seems... unlikely.
See the word "percentage" which you quoted and would suggest I meant 1 in 40?
Or are you doing The Internet Thing of not actually reading/understanding and just trying to look for things to clumsily attempt to dunk on without being able to elucidate whatever it is your own point is supposed to be?
Mixing fractions and percentages doesn't change that an order of magnitude more than 1 in 4 (25%) is 10 in 4 (250%).
If you meant 1 in 40, you should have either said “less” instead of “more” or, “the denominator in the fraction of people made redundant is an order of magnitude more than in that benchmark 1 in 4” in place of “the percentage of people made redundant is an order of magnitude more than that benchmark 1 in 4”, and in either case that’s 2.5% , which while not insignificant if intended as a delta on top of normal unemployment, is less than the delta in a most recessions (the Great Recession peaked at about +5% from the pre-recession trough, which wasn't itself a long-term steady baseline), is also not significant enough to warrant the kind of hyperventilating you are doing. Its also very much not the obvious intent of the words you wrote.
Steps up from that are plausible, but less certain. Do we all want personal pleasure yachts? Flights on a Dragon capsule or a Starship?
-
But there are too many paths the future might yet take for me to be confident of anything much.
If the AI are always narrow, but can be made superhuman for any given skill in 6 months, that has one set of issues; if they are fully human-like general but we never make them more than IQ 85 (plus access to whatever narrow AI exist by then that actual humans also use), that's a different set of issues; and that's just off the top of my head.
It's basically trying to ensure they get this person away from whatever place or job they happen to be currently working in. If AI was able to empower many people with this skill, they wouldn't be paying anywhere near 900K by simple offer/demand.
So, I'm a very unusual person in that I can't be bothered to spend more than about €12k/year. The idea of owning most stuff just doesn't motivate me. I earn significantly more than €12k, because I think the immigration office wants me to, but I'm not motivated by it myself.
If everyone was like me, most advanced economies would just collapse as everyone went down to 5-20 hours per week.
The observation that the economies have not collapsed, demonstrates that most people do actually want to acquire stuff.
The observation of a job offer for $900k strongly implies that there are people who are motivated to acquire $900k (less taxes) worth of stuff each year.
From that it follows that at least some people will be motivated to work just as much as they currently do, in order to earn about 16x as much money, in order to spend it on 16x as much stuff.
(16x is here an observation, not an upper limit).
1. Not everything is stuff. You can book a flight for 5000 Euros to the other end of the world.
2. Not all prices are the best. You might overpay significantly for said flight in order to not have to think about finding a cheaper one. Same for everything else, such as grocery shopping.
3. Housing. It‘s easy to imagine someone buying a house for 5 million. Depending on the location this may not even buy an _extremely_ fancy house. If you don‘t like debt and want to pay it down in 10 years then 900k (and what really remains of that? 400k?) might be useful. Even if you don‘t care about „stuff“.
Maybe all you care about is not having to care about it.
A flight ticket itself may be ephemeral, but the plane and the fuel are not. Overpaying for convenience becomes oversupply for convenience, which at the scale of "everyone gets it" requires massive overproduction.
Houses will indeed always have a "location location location" premium; but the dynamics of this also changes if everyone has far more economic power — $900k is currently enough to commute by private plane, opening up many small places that are currently too out-of-the-way to get such attention, though I don't have any idea if that can actually scale given airport capacities and densities.
10x solar power and we get meet all current electricity demand in a climate friendly way; 10x lithium etc. mining and we can roll out electric vehicles much faster; 10x bioreactors makes… well, very little difference to ethical eco-meat, but mainly because it's still new and the kinks have yet to be worked out, but the principal still applies.
(Unfortunately, my point would still stand even if higher production only sped up disaster; the tragedy of the commons is because everyone is motivated to consume more than is sustainable given the size of the group).
The Earth is a massive planet relative to human scales, there are more abundant resources in it than we can use in a thousand lifetimes.
Roughly speaking, we consume more by consuming "better" things rather than "larger" things. A $1000 iPhone is much "more consumption" than a 1990s cell phone plus portable CD player plus a calculator, but takes less resources than those things. Even more so when consumption of digital entertainment replaces live entertainment.