(back to article index)
CGP Grey is a very intelligent man, but he does sometimes make mistakes. In his video Humans Need Not Apply, he joins the growing crowd of thinkers warning that improvements in automation will make humans unemployable because robots will be able to make everything cheaper than us. Of course, this kind of fear is not new — even David Ricardo worried about technological unemployment, and (as I shall endeavour to shew below) he of all people should have known better — but it has gained a new impetus in recent years thanks to progress in artificial intelligence. Another clever chap, Scott Alexander, also "continue[s] to think it obvious that robots will push humans out of work or at least drive down wages" (deep within a long essay), and reports that this view is normalised among AI researchers.
So if, hypothetically, I happened to know that they were all not just wrong but hilariously wrong, I should probably mention that, and explain why.
I'm going to assume without proof the basic results of price theory / marginal equilibrium theory, namely:
Now, in market equilibrium, the price of goods made by human labour and the price of the same goods made by robots must be equal. (In practice there is only a tendency toward equilibrium, a restoring force counteracting disequilibrium; but for simplicity of description I'm going to ignore that and pretend we have a perfectly frictionless market. It can also be spherical and in a vacuum, if you like; it doesn't change the results.) This is achieved by changes in the quantity of human and robot labour, because the supply curves of both are (in the short run) fixed. So as the price of 'bot labour' decreases, humans do less and less work. So far, so technological-unemployment-dystopia.
But remember that these supply curves and prices are all being measured by the currency, and that's only an intermediate good: actual value is determined by what it will purchase. Economists since at least Adam Smith have warned that prices and wages should be reckoned not as a quantity of coins and bank notes but by the goods and conveniences that those coins will buy. If most human labour has been replaced by cheap automation, the costs of production of most goods have decreased (probably a lot), so the reduced 'total wages of humanity' is not necessarily borne out in terms of purchasing power. The model above isn't wrong as such, it's just that it's formulated in terms of irrelevant variables like 'number of round disks of metal' rather than 'quantity and quality of the affordances and conveniences of life'.
So let's try looking from a different angle, one that recognises that the opportunity cost of robot labour is measured in the products of robot labour, while the opportunity cost of human labour is measured in the products of human labour. We are looking for a way of comparing these two kinds of costs, of finding a common yardstick that abstracts out all considerations of currencies and pretends that we're using barter without any double-coincidence-of-wants problem. Fortunately, such a model exists: the Ricardian theory of international trade.
David Ricardo realised that to say one nation is 'better than' another at 'everything' is to make a meaningless claim (I would call it a type error), because 'how good' England is at making cloth, say, is measured by the quantities of English land, English labour and English capital used to produce a given amount of cloth, whereas the corresponding figure for Portugal is measured in Portuguese land, labour and capital; thus the price of English cloth can not be directly compared to the price of Portuguese cloth, but only compared to (say) the price of English wine, and then the ratio between the prices of English cloth and English wine can be compared to the ratio between the prices of Portuguese cloth and Portuguese wine.
This leads to his famous Law of Comparative Advantage, the observation that since ratios of prices are reciprocal (i.e. the price ratio of English cloth to English wine is one over the price ratio of English wine to English cloth), it is not possible for every good to be cheaper imported than produced locally, nor for every good to be produced cheaply enough to export, but that instead international free trade leads nations to export goods in the production of which they have a comparative advantage, and import goods in which they have a comparative disadvantage, and reap gains thereby that could not be achieved by producing all goods domestically.
The Law of Comparative Advantage has been called "the only proposition in the whole of the social sciences that is both surprising and true"; and I am not going to devote further effort today to proving it, but hereonin take it as obvious.
Now it is I hope not too great a leap to apply this same model to an economy in which human labour must compete with bot labour, by thinking of it as trade between human workers and the owners of the bots (since the bots themselves do not, at least yet, own their labour). In exactly the same way that it is a type error to say that Portugal can produce everything more cheaply than England, it is a type error to say that robots can produce everything more cheaply than humans. Human labour will become concentrated in those tasks or niches in which it has a comparative advantage over robot labour, and in doing so it will reap gains from that specialisation, since humans are now only doing the tasks they are best at while still being able to consume goods whose production involves other tasks.
In short, then, free trade with bots will be better for humans than attempting to 'protect' existing human jobs. (And that's without getting into the public-choice hazards of giving governments the powers to do so!)
In fairness to Scott Alexander, I must admit that I earlier quoted him a little selectively. A less truncated version of the quotation is: "I continue to think it obvious that robots will push humans out of work or at least drive down wages (which, in the existence of a minimum wage, pushes humans out of work)." I accept that robots are likely to drive down nominal wages, and minimum wage laws generally tend to specify a 'number of round disks of metal' rather than 'quantity and quality of the affordances and conveniences of life'. But all we have shown here is that minimum wage laws are a folly that harms the poorest in society by preventing them from engaging in mutually beneficial contracts — and if you didn't already know that, you weren't paying attention in Econ 101 and should probably rectify that before you go around having opinions on economic questions.
Unfortunately, Scott then goes on to say that "Once a robot can do everything an IQ 180 human can do, only better and cheaper, there will be no reason to employ humans at all" and that humans will be "totally locked out of the group whose values capitalism optimizes for". So I think I am not doing him any injustice in characterising his views within the humans-need-not-apply school of thought. However, it does get a bit more subtle, because his argument also involves humans having nothing to contribute as customers, since unemployment has rendered them penniless; I have discussed this further.
Postscript: I should point out that this argument does not prove that automation cannot lower workers' wages, just as it does not prove the same thing about international trade under mobility of capital. (Ricardo knew and noted the exceptions to both cases.) Instead it merely puts some constraints on the production functions that would allow such a thing (which constraints rapidly become pathologically improbable as the reduction in wages is increased). These nuances were pointed out to me in a conversation with David Friedman, who is always worth reading.
(back to article index)