Summary: Things we could do about technological unemployment, if there was technological unemployment. (Novel bit is a top-down macroeconomic calculation.)
Confidence: 90% that the solutions I call ‘vicious’ would be. 60% that the ones I call ‘nonvicious’ would be. Worth emphasising up here that there is little evidence for tech’ unemployment right now.
Automation is maybe the main way that technology improves most people’s lives: aside from status exceptions like Apple products, big reductions in manufacturing cost usually mean big reduction in the end cost of goods. Obviously, replacing labour costs with lower-marginal-cost machines benefits rich machine-owners most, but automation also allows giant price cuts in all kinds of things; over the last two centuries these cuts have transformed society, increasing equality enormously by making things affordable for the first time.
Besides the obvious example—that we now produce a volume of food far beyond the needs of the entire world population (2940kcal per person per day, though with terrible distributive failures) - consider that a single ordinary shirt takes 508 hours of labour to produce on a spinning wheel and hand-loom—so you would expect to pay something above $3600 at current minimum wage (still $900 at 1400CE wage levels).
Getting costs as near to zero as possible is the way we will solve the easy problem of human existence, scarcity of basic goods and leisure. (The hard problem of human life is boredom and dignity and meaning and all that.)
Types of automation
I couldn’t find a rigorous list of the types of automation, so here’s my attempt:
First-wave: Mechanical control. machines which perform one task automatically because of the exact shape of their components. e.g. a) Cams, linked wheels machined to create fixed sequences of linear motions. an instance in the year 200BCE. b) Governors. c) Cataracts.
Second-wave: Numerical control. Programmable machines, where changing the program (punch-cards) changes the product. Grossly physical, analog algorithms. The first fully-auto productive machine was designed and built by Vaucanson in 1745 (not Falcon or Bouchon or Jacquard).
Third-wave: Computer numerical control. One machine, whose behaviour is dictated by algorithms encoded as digital data. The birth of software. Memory allows for ~zero human oversight once task instructions are created and loaded. Arbitrarily fine manual work, performed better than humans and much faster. 1957 or 1958.
Fourth-wave: Machine-learning model control: No need to program, after suitable domain algorithms have been written; models are automatically calibrated given enough data. Trained models can go beyond ordinary control flow, discerning non-necessary and non-sufficient conditions involved in sophisticated tasks like driving. Early successes came in speech recognition (e.g. Tangora and DRAGON in the early 80s, Rabiner et al in 1985), beginning the long march to replace stenographers and secretaries (and ultimately experimental linguists). The defining instance of the fourth wave may be driverless cars: 5% of developed-world jobs at risk. Sometimes called the ‘second machine age’.
Fifth-wave: Artificial general intelligence: No need for programming (programming either new tasks, or any future design of improved machines).
So automation has been happening for (many) hundreds of years, but in the past it probably didn’t produce long-term, “technological” unemployment. This is probably because people were easily able to think up new professions given new tech and culture, and since the increased productivity translated into lower costs for the automated good, which stimulated other parts of the economy, and people could retrain for the subsequent new jobs. The present wave might be different: it might actually reduce the number of available jobs permanently, because the machines now entering the workforce can be applied to many of the jobs that people could retrain to do.
If so, our economy—resource allocation based on employment (which we use as a poor proxy variable for productivity) - is a local maximum and we cannot expect to arrive at a good outcome without activism, since:
The new machine-learning automation could fully replace around half of jobs.
Since these jobs involve even our highest cognitive faculties, it is possible that we won’t think up new productive jobs for the replaced workers, like we did in the past.
So automation could produce an unprecedentedly high unemployment rate, ~60%.
Most existing unemployment welfare systems are inadequate and degrading.
So without intervention, a crash in human welfare is easily possible.
But, unless we automate a lot more, we the species will never have enough wealth to offer a decent basic income, and everyone will continue to waste half their lives at work. Like C20th peasants.
Gross world product divided by population is $10,600 per person! - but this naive distribution would be impossible, even assuming all the political will in the world. Depreciation costs bring this down to ~$9,300 ; maintaining our present levels of R&D investment brings it down further to $9,100.
We also have to consider the “deadweight loss” of taxation (how much you have to spend to collect the tax + how much unproductive tax avoidance behaviour you cause + how much it discourages economic activity + etc). The research on this is shockingly vague (this gives estimates between 2.5% and 30%!). Lower bound takes us to $8,800; the 30% upper bound takes us all the way down to $5,900. The mass carve-up we’re talking about goes well beyond any existing tax rate, and avoidance does scale in proportion to rates; so we probably have to assume the deadweight would be worse than any yet experienced. Call it 30%.
Most people will want to maintain government services at around their current level (besides the giant basic income expenditure); remember that this could knock another 30% off our available income flow (or a mere 28% off if we lose the military). Half of that is welfare and pensions, which are being replaced here (in our heads). So we’re down to $4,300.
So the current economy, carved up sustainably, would yield some fraction of $4,000 per person per year. Even given that most households would get about 4 of these incomes, this is simply not enough for freedom, given the needs or tastes of an average human.
(The above does not consider a host of other sad realities: e.g. rich people like their money; e.g. this much equality would destroy entire industries (luxury goods!), and so further sap the available pie; e.g. there would be a recession effect from all the fully-alienated workers downing their tools—yes, there could also be a stimulus effect from increasing poor people’s spending, but it’s extremely difficult to say which sign prevails.
Worse: only the direct cost of taxation is factored in above, without the amount that the rich manage to spirit away. We would need something like a world government for it to work even this well (badly), to stamp out tax havens and transfer pricing and all that jolly financial dancing.)
My point is not that this is the exact figure we’ll have to work with - instead it points up our paltry present capacity. Growth would be necessary, even in an ideal world without nationalism, greed, inefficiency (...)
Socially conscious people are these days ambivalent about economic growth, often for environmental reasons. But consider the enlightened definition of “productivity”: it is not “amount of output”, but the amount of output per unit of input. This is pure gain, and is actually environmentally positive, since it could reduce resource use and waste. But we do need output growth too: e.g. until every paraplegic on earth who wants one of these has one.
The above just uses world income; what about using world wealth? Even if we liquidate the whole of the world’s wealth (our stock of money, as opposed to the GDP, a flow), it would only provide a universal basic income for three and a half years.
Probably vicious solutions to the worst case:
Halt progress on automation, preserving current employment. (Via government ban, successful hostility by organised labour, mass monkey wrenching). I count this as vicious, even though it is much better than the worst-case, since it leaves almost all of us very unfree, forever.
or I guess you could try huge unemployment plus an authoritarian crackdown on desperate masses, see how that goes.
Nationalised robot factories, or full cyber-communism. Food and clothes and houses guaranteed to all, at least. Leave aside the historical failure of command economies; imagine here that a new big-dataKantorovich manages to make it fairly efficient.
Even granting this giant assumption, this is a dangerous move. Total state control of the means of production is too easily twisted. Now, this could be just my emotional overreaction to reading about e.g. Maoist China, with its food terrorismand peasant-robbing. But it rings malign: total control of production by any entity is a terrible unnecessary risk.
“Back to the land” primitivism. Humans return to subsistence farming as a means of survival.
Just raising the minimum wage without doing anything else. Misses the point entirely. (‘Should the minimum wage be called the “Robot Employment Act?”’ – Cowen and Tabarrok.)
Mass human augmentation, to keep up with the machines. (At minimum, just traditional externally-hosted software: Humans using chess assistant programs were beating solo supercomputers until relatively recently, c. 2006.) Possibly vicious, but not because there’s anything wrong with transhumanism: because doing it for purely economic reasons is 1) a neverending process, since the machines will improve as fast, and 2) it would probably be destructive of some distinctive human virtues (e.g. serenity, play, reflection, aesthetic interest). Only good if people really can’t feel dignified without having a leading productive role in things.
Potentially nonvicious solutions:
Prop up the liberal mixed economy:
with a programme of mass employee stock ownership.
or by carving each full-time job into several part-time ones, plus heavy wage subsidies.
or with a universal basic income funded through higher taxation.
2. More or less vague suggestions for a very different social structure, like embarrassingly decentralised groupings, with their own minifactories...
I am not very sure of any of the above; the actual stats on productivity growth are worrying for the opposite reason: it has been too slow to support wages for a long time. Anyway other powerful forces (e.g. global outsourcing, the decay of unions) besides robots have led to the 40-year decline in labour’s share of global income. But those will produce similar dystopian problems if the trend continues, and there’s enough of a risk of the above scenario for us to put a lot of thought and effort into protecting people, either way.
Factories that run ‘lights out’ are fully automated and require no human presence on-site… these factories can be run with the lights off.
– Wiki
Not only is it lights-out—we turn off the air conditioning and heat too.
– an executive at Fuji Automatic Numerical Control
Automatic for the people
Summary: Things we could do about technological unemployment, if there was technological unemployment. (Novel bit is a top-down macroeconomic calculation.)
Confidence: 90% that the solutions I call ‘vicious’ would be. 60% that the ones I call ‘nonvicious’ would be. Worth emphasising up here that there is little evidence for tech’ unemployment right now.
Crossposted from gleech.org.
Autonomous trucks are now in use and are already safer and more fuel-efficient than human driven ones. (Truck drivers are ~2% of the entire American workforce.)
Crap journalism (that is, 80% of (UK) journalism) is now fully automatable. Automatic art is quite good and improving fast. Consider also the cocktail bartender. And so on: maybe half of all jobs are at risk of being automated, assuming the rate of AI progress just stays constant (“over an unspecified period, perhaps a decade or two”).
Automation is maybe the main way that technology improves most people’s lives: aside from status exceptions like Apple products, big reductions in manufacturing cost usually mean big reduction in the end cost of goods. Obviously, replacing labour costs with lower-marginal-cost machines benefits rich machine-owners most, but automation also allows giant price cuts in all kinds of things; over the last two centuries these cuts have transformed society, increasing equality enormously by making things affordable for the first time.
Besides the obvious example—that we now produce a volume of food far beyond the needs of the entire world population (2940kcal per person per day, though with terrible distributive failures) - consider that a single ordinary shirt takes 508 hours of labour to produce on a spinning wheel and hand-loom—so you would expect to pay something above $3600 at current minimum wage (still $900 at 1400CE wage levels).
Getting costs as near to zero as possible is the way we will solve the easy problem of human existence, scarcity of basic goods and leisure. (The hard problem of human life is boredom and dignity and meaning and all that.)
Types of automation
I couldn’t find a rigorous list of the types of automation, so here’s my attempt:
First-wave: Mechanical control. machines which perform one task automatically because of the exact shape of their components. e.g. a) Cams, linked wheels machined to create fixed sequences of linear motions. an instance in the year 200BCE. b) Governors. c) Cataracts.
Second-wave: Numerical control. Programmable machines, where changing the program (punch-cards) changes the product. Grossly physical, analog algorithms. The first fully-auto productive machine was designed and built by Vaucanson in 1745 (not Falcon or Bouchon or Jacquard).
Third-wave: Computer numerical control. One machine, whose behaviour is dictated by algorithms encoded as digital data. The birth of software. Memory allows for ~zero human oversight once task instructions are created and loaded. Arbitrarily fine manual work, performed better than humans and much faster. 1957 or 1958.
Fourth-wave: Machine-learning model control: No need to program, after suitable domain algorithms have been written; models are automatically calibrated given enough data. Trained models can go beyond ordinary control flow, discerning non-necessary and non-sufficient conditions involved in sophisticated tasks like driving. Early successes came in speech recognition (e.g. Tangora and DRAGON in the early 80s, Rabiner et al in 1985), beginning the long march to replace stenographers and secretaries (and ultimately experimental linguists). The defining instance of the fourth wave may be driverless cars: 5% of developed-world jobs at risk. Sometimes called the ‘second machine age’.
Fifth-wave: Artificial general intelligence: No need for programming (programming either new tasks, or any future design of improved machines).
So automation has been happening for (many) hundreds of years, but in the past it probably didn’t produce long-term, “technological” unemployment. This is probably because people were easily able to think up new professions given new tech and culture, and since the increased productivity translated into lower costs for the automated good, which stimulated other parts of the economy, and people could retrain for the subsequent new jobs. The present wave might be different: it might actually reduce the number of available jobs permanently, because the machines now entering the workforce can be applied to many of the jobs that people could retrain to do.
If so, our economy—resource allocation based on employment (which we use as a poor proxy variable for productivity) - is a local maximum and we cannot expect to arrive at a good outcome without activism, since:
The new machine-learning automation could fully replace around half of jobs.
Since these jobs involve even our highest cognitive faculties, it is possible that we won’t think up new productive jobs for the replaced workers, like we did in the past.
So automation could produce an unprecedentedly high unemployment rate, ~60%.
Most existing unemployment welfare systems are inadequate and degrading.
So without intervention, a crash in human welfare is easily possible.
But, unless we automate a lot more, we the species will never have enough wealth to offer a decent basic income, and everyone will continue to waste half their lives at work. Like C20th peasants.
How much is there for everyone?
You can follow my calculation here.
Gross world product divided by population is $10,600 per person! - but this naive distribution would be impossible, even assuming all the political will in the world. Depreciation costs bring this down to ~$9,300 ; maintaining our present levels of R&D investment brings it down further to $9,100.
We also have to consider the “deadweight loss” of taxation (how much you have to spend to collect the tax + how much unproductive tax avoidance behaviour you cause + how much it discourages economic activity + etc). The research on this is shockingly vague (this gives estimates between 2.5% and 30%!). Lower bound takes us to $8,800; the 30% upper bound takes us all the way down to $5,900. The mass carve-up we’re talking about goes well beyond any existing tax rate, and avoidance does scale in proportion to rates; so we probably have to assume the deadweight would be worse than any yet experienced. Call it 30%.
Most people will want to maintain government services at around their current level (besides the giant basic income expenditure); remember that this could knock another 30% off our available income flow (or a mere 28% off if we lose the military). Half of that is welfare and pensions, which are being replaced here (in our heads). So we’re down to $4,300.
So the current economy, carved up sustainably, would yield some fraction of $4,000 per person per year. Even given that most households would get about 4 of these incomes, this is simply not enough for freedom, given the needs or tastes of an average human.
(The above does not consider a host of other sad realities: e.g. rich people like their money; e.g. this much equality would destroy entire industries (luxury goods!), and so further sap the available pie; e.g. there would be a recession effect from all the fully-alienated workers downing their tools—yes, there could also be a stimulus effect from increasing poor people’s spending, but it’s extremely difficult to say which sign prevails.
Worse: only the direct cost of taxation is factored in above, without the amount that the rich manage to spirit away. We would need something like a world government for it to work even this well (badly), to stamp out tax havens and transfer pricing and all that jolly financial dancing.)
My point is not that this is the exact figure we’ll have to work with - instead it points up our paltry present capacity. Growth would be necessary, even in an ideal world without nationalism, greed, inefficiency (...)
Socially conscious people are these days ambivalent about economic growth, often for environmental reasons. But consider the enlightened definition of “productivity”: it is not “amount of output”, but the amount of output per unit of input. This is pure gain, and is actually environmentally positive, since it could reduce resource use and waste. But we do need output growth too: e.g. until every paraplegic on earth who wants one of these has one.
The above just uses world income; what about using world wealth? Even if we liquidate the whole of the world’s wealth (our stock of money, as opposed to the GDP, a flow), it would only provide a universal basic income for three and a half years.
Probably vicious solutions to the worst case:
Halt progress on automation, preserving current employment. (Via government ban, successful hostility by organised labour, mass monkey wrenching). I count this as vicious, even though it is much better than the worst-case, since it leaves almost all of us very unfree, forever.
or I guess you could try huge unemployment plus an authoritarian crackdown on desperate masses, see how that goes.
Nationalised robot factories, or full cyber-communism. Food and clothes and houses guaranteed to all, at least. Leave aside the historical failure of command economies; imagine here that a new big-data Kantorovich manages to make it fairly efficient.
Even granting this giant assumption, this is a dangerous move. Total state control of the means of production is too easily twisted. Now, this could be just my emotional overreaction to reading about e.g. Maoist China, with its food terrorism and peasant-robbing. But it rings malign: total control of production by any entity is a terrible unnecessary risk.
“Back to the land” primitivism. Humans return to subsistence farming as a means of survival.
Just raising the minimum wage without doing anything else. Misses the point entirely. (‘Should the minimum wage be called the “Robot Employment Act?”’ – Cowen and Tabarrok.)
Mass human augmentation, to keep up with the machines. (At minimum, just traditional externally-hosted software: Humans using chess assistant programs were beating solo supercomputers until relatively recently, c. 2006.) Possibly vicious, but not because there’s anything wrong with transhumanism: because doing it for purely economic reasons is 1) a neverending process, since the machines will improve as fast, and 2) it would probably be destructive of some distinctive human virtues (e.g. serenity, play, reflection, aesthetic interest). Only good if people really can’t feel dignified without having a leading productive role in things.
Potentially nonvicious solutions:
Prop up the liberal mixed economy:
with a programme of mass employee stock ownership.
or by carving each full-time job into several part-time ones, plus heavy wage subsidies.
or with a universal basic income funded through higher taxation.
Get the government to buy every 18 year old a serious stock portfolio (??)
2. More or less vague suggestions for a very different social structure, like embarrassingly decentralised groupings, with their own minifactories...
I am not very sure of any of the above; the actual stats on productivity growth are worrying for the opposite reason: it has been too slow to support wages for a long time. Anyway other powerful forces (e.g. global outsourcing, the decay of unions) besides robots have led to the 40-year decline in labour’s share of global income. But those will produce similar dystopian problems if the trend continues, and there’s enough of a risk of the above scenario for us to put a lot of thought and effort into protecting people, either way.
– Wiki
– an executive at Fuji Automatic Numerical Control