Plate vii in print

The Robot

automation and the substitution of capital for labor; whether the machine takes labor's share of the product or makes new work for it.

The machine does not simply replace the worker, nor simply spare him; it does both at once — automating the tasks he used to perform and, where new tasks arise that he performs better, handing him those instead. Whether his share of the product rises or falls is the outcome of that race, between the work the machine takes and the work it leaves or makes; and nothing in the technology itself decides which way the race is run.

— the argument of this plate, after Daron Acemoglu and Pascual Restrepo, 2018

The volume has named several forces that set the division of the product — the shape of production, the reach of market power, the long swing of struggle between the classes, the recent slipping of labor's portion. Behind all of them stands an older and cruder suspect: the machine. The fear is ancient: that the machine, by doing what the worker did, will take his work and with it his share, until labor is left a smaller and smaller slice, or none. Against the fear stands an equally old reassurance, that every past machine, having destroyed some work, created more, so that labor always found new things to do and kept its place. The two have quarreled for two centuries without a settlement, because each is partly right, and the real question is which of them holds the stronger hand in a given age. Daron Acemoglu and Pascual Restrepo gave that quarrel its first precise frame, in which the volume's question — what sets labor's share? — meets its most disputed answer: the machine, which can shrink labor's portion or enlarge it, and whose verdict is never given in advance.

The frame begins by refusing to treat labor and capital as two interchangeable fluids poured into production in some ratio. Acemoglu and Restrepo picture production instead as a long list of distinct tasks — the many separate things that must be done to make a good and bring it to market — each performable either by a worker or by a machine. At any moment some of the tasks are done by labor and some by capital, and labor's share of the product depends on how many of them, weighted by their value, fall to it. Automation, in this picture, is precisely the handing of a task that labor used to do over to a machine that now does it more cheaply. Each such handover is a displacement: labor loses a task it once held, the work that task represented passes to capital, and, all else equal, labor's share of the product falls by the value of what it lost.1 This is the kernel of truth in the ancient fear. The machine, task by task, does displace the worker, and each displacement is a small subtraction from labor's portion of the whole.

1. Acemoglu and Restrepo model production as a list of tasks, each performable by labor or by capital. Labor's share is the value-weighted fraction of the tasks it performs. Automation hands a task from labor to capital — the displacement effect — and each handover subtracts from labor's share. This is the kernel of the old fear: the machine, task by task, takes work that was labor's.

But the same accounting holds an answer to the fear, and it is the heart of what Acemoglu and Restrepo added. The list of tasks is not fixed. As old tasks are automated, new ones come into being — new goods, new services, new things to be done that did not exist before — and in many of them it is labor, not the machine, that holds the advantage. Each new task in which labor is the cheaper performer is a reinstatement: labor is handed work it lacked, its portion rises, and the subtraction of displacement is met by an addition.2 So labor's share is not driven one way by the machine but pulled in two at once. It falls as automation displaces labor from old tasks; it rises as new tasks reinstate it in fresh ones; and where it comes to rest — whether labor keeps its accustomed portion, or slips, or climbs — is the outcome of a race between the two. The machine is neither the enemy of the ancient fear nor the friend of the easy reassurance. It is both at once, and which face it turns depends on which effect, in a given age, runs the faster.

2. But the task list grows. New tasks arise, new goods and services, and in many of them labor holds the advantage. Each is a reinstatement, adding to labor's share. The share is thus a race: displacement subtracts as automation takes old tasks, reinstatement adds as new tasks arrive, and where labor's portion lands depends on which effect runs faster.
⁂

Seen through this lens, the long history of the machine resolves into the history of that race, and for the better part of two centuries the two effects ran close enough that labor held its ground. The power loom destroyed the handweaver's task and the tractor the farmhand's, and each time the displaced poured, after hard transition, into tasks that had not existed — the factory, the office, the laboratory, the service counter — so that the work the machine took was roughly matched by the work it made, and labor's portion, though it wandered, came back. This is why the recurring prophecy of technological unemployment — pronounced by the machine-breakers, again in the automation scare of the mid-twentieth century, in nearly every decade since — was each time refuted by events: not because the machine spared labor, but because reinstatement, in the end, kept pace with displacement.3 The fear was never foolish; it was simply, each time, outrun by the making of new work. The optimism was never groundless; it simply mistook a close race, repeatedly won, for a law that labor could not lose.

3. For two centuries displacement and reinstatement ran close. The loom took the weaver's task, the tractor the farmhand's; the displaced moved, after hard transition, into tasks that had not existed. So the recurring prophecy of technological unemployment — the machine-breakers, the mid-century automation scare — was each time refuted, not because the machine spared labor but because reinstatement kept pace.

What makes the present anxious is the suspicion that the race has lately ceased to be close. The slipping the volume has already measured — labor's portion falling across nation after nation since about 1980 — looks, in this frame, less like the impersonal work of cheaper capital and more like a race labor has begun to lose: automation, in robotics and above all software, running fast, while the making of new tasks where labor holds the advantage has not kept up. Acemoglu and Restrepo read the decline this way — not as the doing of any single villain but as an imbalance, much displacement and little reinstatement, weighted against labor for a stretch longer than any in living memory. And they press a harder point: that the imbalance is not wholly an accident of what was technically possible but partly a thing we have chosen, having aimed our inventiveness at automating the work that exists — the self-checkout, the routine the algorithm runs in the clerk's place — rather than at new tasks that would put labor back to work.4 The race has tipped; and part of what tipped it is where we have pointed the machine.

4. Since about 1980 the race appears to have tipped toward displacement — much automation, in robots and software, and too few new tasks where labor leads. Acemoglu and Restrepo read the falling labor share this way: not one villain but an imbalance. And they argue the tilt is partly chosen, invention having aimed at automating existing work rather than at creating new work for labor.

This sharpens, without settling, the oldest form of the question: is the machine of our own age different in kind? The reassurance of two centuries rests on a record of reinstatement — new work has always come — but a record is not a guarantee, and the task frame shows precisely what the guarantee would require: that the making of new tasks in which humans hold the advantage keep pace with a technology now able to take over ever wider stretches of what humans do. The pessimist's image is the horse, which the engine for a time made more productive and then, its every task assumed, made all but obsolete — labor for which demand simply vanished. The optimist answers that human wants are inexhaustible and human versatility unlike a horse's, so that new tasks will go on arriving as they always have. The model cannot say which is right.5 What it can do is show that the answer turns on a measurable race and not a metaphysical fate, and that the share labor keeps is the score that race is keeping.

5. The reassurance rests on a record of reinstatement, not a guarantee. The pessimist's image is the horse, made obsolete once the engine took its every task. The optimist answers that human wants are endless and human versatility unlike a horse's. The model cannot decide between them; it shows only that the answer is a measurable race, not a fate, and labor's share is the running score.

So the machine, the crudest and most frightening of the forces on labor's share, turns out under inspection to be the least automatic in its verdict. It does not dictate labor's portion; it sets two opposing effects in motion and leaves the outcome to their balance — struck not by the technology alone but by what is invented, what is automated, and what new work is made, every one of which is a matter of direction and choice. The share of the product that goes to those who work is not, in the end, written by the machine, any more than it was written by the shape of production. Like every answer this volume has weighed, it comes back to something the arithmetic cannot fix by itself: here, the direction in which a society aims its inventiveness, and whether that direction leaves room for human work or closes it off. The capstone will press the point to its limit — that the division of the product, over the long run, is settled less by the tools a society holds than by the powers that decide how those tools are turned.