Predictions from a throw-away Herbert Simon article in 1960
The Corporation: will it be managed by machines?
In 1960, Herbert Simon was tasked with predicting what work life would look like in 1985.1 The whole paper is chock full of fun predictions, way more than I can list here (I deleted 75% of this post to make it palatable). Go read the whole paper! But let me list some, and keep in mind this was 1960, people!
Like Adam Smith in 1776, you could say that Simon properly saw the beginning of an exponential curve where others saw a continuing flat line.
On AI
Let me make my point perfectly clear. Heuristic programs do not merely substitute machine brute force for human cunning. Increasingly, they imitate—and in some cases improve upon—human cunning.
Here he is describing his work on the General Problem Solver, an early precursor of AGI.
It is my conviction that no major new ideas will have to be discovered to enable us to extend these early results to the whole of human thinking, problem solving, decision-making activity. We have every reason to believe that within a very short time—I am even willing to say ten years or less—we will be able technically to produce computers that can grapple with and solve at least the range of problems that humans are able to grapple with and solve—those that are ill-structured as well as those that are well-structured.
The transformer based on attention mechanisms might be a new idea, but the difference between 2017 and 1970 is basically a rounding error, so should we give him that one too?
To what extent, in 1985, will managers and other humans be occupied in thinking about and solving ill-structured problems, as distinct from doing other things? On this point the image in my crystal ball is very dim. I will nevertheless hazard some guesses. My first guess is that man will retain a greater comparative advantage in handling ill-structured problems than in handling well-structured problems. My second guess is that he will retain a greater advantage in tasks involving sensory-manipulative coordination— “physical flexibility”—than in ill-structured problem-solving tasks—“mental flexibility.” If this is true, a larger part of the working population will be mahouts and wheelbarrow pushers and a smaller part will be scientists and executives—particularly of the staff variety. The amount of shift in this direction will be somewhat diminished by the fact that as income and general productivity rise, the demand for work involving ill-structured problem solving will probably increase more than the demand for work involving flexible manipulation of the physical environment.
Similarly, we can dismiss the notion that computer programer will become a powerful elite in the automated corporation. It is far more likely that the programing occupation will become extinct (through the further development of self-programing techniques) than that it will become all-powerful. More and more, computers will program themselves; and direction will be given to computers through the mediation of compiling systems that will be completely neutral so far as content of the decision rules is concerned. Moreover, the task of communicating with computers will become less and less technical as computers come—by means of compiling techniques—closer and closer to handling the irregularities of natural language.
Big if true! Oh wait!
And there’s this one:
Computers behave like morons only because we are just beginning to learn how to communicate with them in something better than moronic language… They have not reached the point where they permit the programmer to communicate with the computer in idiomatic English, but only in a kind of simple pidgin English.
On occupations
The change in the occupational profile depends on a well-known economic principle, the doctrine of comparative advantage…Whether man or machines will be employed in a particular process depends not simply on their relative productivity in physical terms but on their cost as well. And cost depends on price. Hence—so goes the traditional argument of economics—as technology changes and machines become more productive, the prices of labor and capital will so adjust themselves as to clear the market of both. As much of each will be employed as offers itself at the market price, and the market price will be proportional to the marginal productivity of that factor. By the operation of the market place, manpower will flow to those processes in which its productivity is comparatively high relative to the productivity of machines; it will leave those processes in which its productivity is comparatively low.
Remember that Bowen and Baumol wrote their famous piece in 1965, but basically the “Baumol Effect” is just a core premise of technological change that good economists always knew.2 In fact, Simon follows up that paragraph with “I apologize for dwelling at length on a point that is clearly enough stated in the Wealth of Nations.” Is it Cowen’s 17th law, that things always have earlier origins than you think?
We can conjecture that by 1985 the departments of a company concerned with major clerical functions—accounting, processing of customers’ orders, inventory and production control, purchasing, and the like—will have reached an even higher level of automation than most factories.
New hypothesis: the tax code changes just to preserve rents for accountants.
Man’s comparative advantage in energy production has been greatly reduced in most situations—to the point where he is no longer a significant source of power in our economy. He has been supplanted also in performing many relatively simple and repetitive eye-brain-hand sequences. He has retained his greatest comparative advantage in: (1) the use of his brain as a flexible general-purpose problem-solving device, (2) the flexible use of his sensory organs and hands, and (3) the use of his legs, on rough terrain as well as smooth, to make this general-purpose sensing-thinking-manipulating system available wherever it is needed.
Machines will play an increasing role, of course, in maintenance functions, but machine powers will not likely develop as rapidly relatively to those of men in this area as in in-line activities. Moreover, the total amount of maintenance work—to be shared by men and machines—will increase. For the middle run, at least, I would expect this group to make up an increasing fraction of the total work force.
Finally, in the entire occupied population, a larger fraction of members than at present will be engaged in occupations where “personal service” involving face-to-face human interaction is an important part of the job. I am confident in stating this conclusion; far less confident in conjecturing what these occupations will be, for the reasons already set forth.
Read that one again! Then apply it to 2035. Then read my prediction.
A large part of the total middle-management job consists of decisions of the same general character as those that have already yielded to automation. The decisions are repetitive and require little of the kinds of flexibility that constitute man’s principal comparative advantage over machines. We can predict with some confidence, I think, that persons making such decisions will constitute a much smaller fraction of the total occupied group within a few years than they do now.
This comes after he lists extensively the many ways that computers, programs, and mathematics are building the field of “operations research” that (in 1960) will soon do all of the sales, inventory, procurement, order filling, invoicing, scheduling, and, well, operations research. Of course, we know what came to pass.
I think we can predict that in future years the manager’s time perspective will be lengthened. As automated subsystems take over the minute-by-minute and day-by-day operation of the factory and office, the humans in the system will become increasingly occupied with preventive maintenance, with system breakdowns and malfunctions, and—perhaps most important of all—with the design and modification of systems.
General Contours
Within the very near future—much less than twenty-five years—we shall have the technical capability of substituting machines for any and all human functions in organizations…To predict that we will have these technical capabilities says nothing of how we shall use them. Before we can forecast that, we must discuss the important invariants in the social system.
Quite perceptive. It is true that for any individual task we could think of, we could program a machine to do it. But obviously it is not economical when we have cheaper wage workers. We could also build railroads out of titanium, but it is not economical when we have cheaper iron. Will we let it become economical is the question.
The distribution of intelligence and ability in the society will be much as it is now, although a substantially larger percentage of adults (perhaps half or more) will have completed college educations.
My over-generous reading is that because college is signaling + networking + socializing into the dominant culture (which is all a kind of human capital, just not literal skill camp), and this mix develops taste in some sense which is the scarce factor.
Fin
Just as a reminder on his style, Herbert Simon starts off the article as follows:
My work on this paper has been somewhat impeded, in recent days, by a fascinating spectacle just outside my office window. Men and machines have been constructing the foundations of a small building. After some preliminary skirmishing by men equipped with surveying instruments and sledges for driving pegs, most of the work has been done by various species of mechanical elephant and their mahouts. Two kinds of elephants dug out the earth (one with its forelegs, the other with its trunk) and loaded it in trucks (pack elephants, I suppose). Then, after an interlude during which another group of men carefully fitted some boards into place as forms, a new kind of elephant appeared, its belly full of concrete which it disgorged into the forms. It was assisted by two men with wheelbarrows plain old-fashioned man-handled wheelbarrows and two or three other men who fussily tamped the poured concrete with metal rods. Twice during this whole period a shovel appeared on one occasion it was used by a man to remove dirt that had been dropped on a sidewalk; on another occasion it was used to clean a trough down which the concrete slid.
Here, before me, was a sample of automated, or semiautomated production. What did it show about the nature of present and future relations of man with machine in the production of goods and services?
Herb Simon is highly rated yet still underrated.
See my other posts on AI:
Prediction: The Growth of the Vibe Sector
What is left for teachers in the age of AI?
The first ASI will be decentralized
It was printed in a symposium collection that included F.A. Hayek, Adolf Berle, and Robert Merton (the famous sociologist father of the famous economist). See also Richard Langlois’s instructive comment on Simon’s article.
You basically can’t talk about AI with reaching semantic satiation on the word “Baumol.”