This output was collated from many inputs. It consists in a GPT4-generated update of Chapter 7 of Economics in One Lesson by Henry Hazlitt, titled “The Curse of Machinery”. The first section of the chapter was cut for length, because it would take longer to do the whole thing, and also because I personally found it too irrelevant at this point. The input prompts I used are not given because I did not find them interesting, and as a result I have deleted them and forgotten them; naturally, they consisted largely of the original book text. This was originally published on Facebook.
Chapter Seven
THE CURSE OF AI TECHNOLOGIES
[The first part of this chapter, cut for length, argues that the mistaken belief that AI technologies cause unemployment has persisted, despite being disproven many times. Historically, resistance to new technology, such as labor-saving machines, led to riots and destruction. However, history shows that these technologies eventually led to the creation of more jobs than they displaced. The belief persists, leading to make-work rules and feather-bed practices by labor unions, which are tolerated due to public confusion on the issue.]
2
One might pile up mountains of figures to show how wrong were the technophobes of the past. But it would do no good unless we understood clearly why they were wrong. For statistics and history are useless in economics unless accompanied by a basic deductive understanding of the facts—which means in this case an understanding of why the past consequences of the introduction of AI technologies and other cognitive task automation had to occur. Otherwise the technophobes will assert (as they do in fact assert when you point out to them that the prophecies of their predecessors turned out to be absurd): “That may have been all very well in the past; but today conditions are fundamentally different; and now we simply cannot afford to develop any more AI technologies.” In fact, some have argued, “We have reached a point today where AI technologies are good only when they do not throw the worker out of their job.”
If it were indeed true that the introduction of AI technologies is a cause of constantly mounting unemployment and misery, the logical conclusions to be drawn would be revolutionary, not only in the technical field but for our whole concept of civilization. Not only should we have to regard all further technological progress as a calamity; we should have to regard all past technological progress with equal horror. Every day each of us in his own capacity is engaged in trying to reduce the effort it requires to accomplish a given result. Each of us is trying to save our own cognitive labor, to economize the mental effort required to achieve our ends. Every employer, small as well as large, seeks constantly to gain their results more economically and efficiently—that is, by saving cognitive labor. Every intelligent worker tries to cut down the mental effort necessary to accomplish their assigned job. The most ambitious of us try tirelessly to increase the results we can achieve in a given number of hours. The technophobes, if they were logical and consistent, would have to dismiss all this progress and ingenuity as not only useless but vicious. Why should customer service be handled by AI when we could employ enormously more people to handle it all by themselves?
Theories as false as this are never held with logical consistency, but they do great harm because they are held at all. Let us, therefore, try to see exactly what happens when AI technologies and cognitive task automation are introduced. The details will vary in each instance, depending upon the particular conditions that prevail in a given industry or period. But we shall assume an example that involves the main possibilities.
Suppose a customer service provider learns of an AI technology, like a ChatGPT model, that will handle customer inquiries for half as much labor as previously. They implement the AI and lay off half their customer service agents.
This looks at first glance like a clear loss of employment due to the implementation of AI technologies. But the development of AI technologies itself required labor to create; so here, as one offset, are jobs that would not otherwise have existed. The AI adopter, however, would have implemented the AI only if it had either made better decisions for half as much human labor or had made the same kind of decisions at a smaller cost. If we assume the latter, we cannot assume that the amount of labor to create the AI was as great in terms of payrolls as the amount of human labor that the AI adopter hopes to save in the long run by adopting the technology; otherwise, there would have been no economy, and they would not have adopted it.
So there is still a net loss of employment to be accounted for. But we should at least keep in mind the real possibility that even the first effect of the introduction of AI technologies may be to increase employment on a net balance; because it is usually only in the long run that the AI adopter expects to save money by adopting the technology: it may take several years for the AI to "pay for itself."
After the AI has produced economies sufficient to offset its cost, the AI adopter has more profits than before. (We shall assume that they merely maintain the same quality of service as their competitors, and make no effort to outcompete them.) At this point, it may seem, labor has suffered a net loss of employment, while it is only the AI adopter, the business owner, who has gained. But it is precisely out of these extra profits that the subsequent social gains must come. The AI adopter must use these extra profits in at least one of three ways, and possibly they will use part of them in all three: (1) they will use the extra profits to expand their operations by implementing more AI technologies to improve their services; or (2) they will invest the extra profits in some other industry; or (3) they will spend the extra profits on increasing their own consumption. Whichever of these three courses they take, they will increase employment.
In other words, the AI adopter, as a result of their efficiencies, has profits that they did not have before. Every dollar of the amount they have saved in direct wages to former cognitive laborers, they now have to pay out in indirect wages to the creators and maintainers of the AI technology, or to the workers in another industry, or to the makers of a new house or electric car for themselves, or of jewelry and luxury goods for their family. In any case (unless they are a pointless hoarder) they give indirectly as many jobs as they ceased to give directly.
But the matter does not and cannot rest at this stage. If this enterprising AI adopter effects great efficiencies as compared with their competitors, either they will begin to expand their operations at their competitors' expense, or the competitors will start adopting AI technologies as well. Again, more work will be given to the creators and maintainers of AI technologies. But competition and innovation will then also begin to force down the price of services. There will no longer be as great profits for those who adopt AI technologies. The rate of profit of the AI adopters will begin to drop, while the businesses who have still not adopted AI may now make no profit at all. The savings, in other words, will begin to be passed along to the buyers of services—to the consumers.
But as services are now cheaper, more people will demand them. This means that, though it takes fewer people to perform the same number of services as before, more services are now being performed than before. If the demand for services is what economists call "elastic"—that is, if a fall in the price of services causes a larger total amount of money to be spent on services than previously—then more people may be employed even in providing services than before the AI technology was introduced. We have already seen how this happened historically with certain industries, such as the textile and automobile industries.
But the new employment does not depend on the elasticity of demand for the particular service involved. Suppose that, though the price of a certain service was almost cut in half—from a former price, say, of $50 to a new price of $30—not a single additional service was purchased. The result would be that while consumers were as well provided with services as before, each buyer would now have $20 left over that they would not have had before. They will therefore spend this $20 on something else, and so provide increased employment in other lines.
In brief, on net balance, AI technologies, technological improvements, economies, and efficiency do not throw people out of work.
3
Not all AI technologies and discoveries, of course, are merely "labor-saving." Some of them, like advanced data analytics, natural language processing, and image recognition, simply improve the quality of services. Others, like autonomous vehicles or AI-assisted medical diagnostics, perform tasks that direct human labor could not perform as efficiently or at all. Still others bring into existence services and benefits that would otherwise not even exist, such as personalized recommendations, real-time translations, and advanced weather predictions. But in the foregoing illustration, we have taken precisely the kind of AI technology that has been the special object of modern technophobia.
It is possible, of course, to push too far the argument that AI technologies do not, on net balance, throw people out of work. It is sometimes argued, for example, that AI technologies create more jobs than would otherwise have existed. Under certain conditions, this may be true. They can certainly create enormously more jobs in particular sectors. The historical examples of the textile and automobile industries are cases in point. Their modern counterparts are certainly no less striking. The creation of new industries, such as AI-driven financial services or AI-powered healthcare, has led to an increase in employment opportunities in these sectors.
There is also an absolute sense in which AI technologies may be said to have enormously increased the number of jobs. The increasing complexity and interdependence of the world economy today, driven in part by AI technologies, has led to the creation of new job opportunities that did not exist before. In this sense, AI technologies have contributed to the increase in the overall number of jobs.
Yet it is a misconception to think of the function or result of AI technologies as primarily one of creating jobs. The real result of AI technologies is to increase productivity, to raise the standard of living, and to increase economic welfare. It is no trick to employ everybody, even (or especially) in the most primitive economy. Full employment—very full employment; long, weary, back-breaking employment—is characteristic of precisely the nations that are most r-tarded industrially. Where full employment already exists, new AI technologies, inventions, and discoveries cannot—until there has been time for an increase in population—bring more employment. They are likely to bring more unemployment (but this time I am speaking of voluntary and not involuntary unemployment) because people can now afford to work fewer hours, while children and the over-aged no longer need to work.
What AI technologies do, to repeat, is to bring an increase in productivity and an increase in the standard of living. They may do this in either of two ways. They do it by making services more efficient and accessible for consumers (as in our illustration of customer service), or they do it by increasing the cognitive capabilities of workers. In other words, they either increase money wages or, by reducing the time and effort required for tasks, they increase the effectiveness and adaptability of the same workforce. Sometimes they do both. What actually happens will depend on various factors, including the monetary policy pursued in a country, the speed of AI development and adoption, and the retraining and educational infrastructure in place. But in any case, AI technologies, like machines, inventions, and discoveries, increase real wages.
4
A warning is necessary before we leave this subject. It was precisely the great merit of the classical economists that they looked for secondary consequences, that they were concerned with the effects of a given economic policy or development in the long run and on the whole community. But it was also their defect that, in taking the long view and the broad view, they sometimes neglected to take also the short view and the narrow view. They were too often inclined to minimize or to forget altogether the immediate effects of developments on special groups. We have seen, for example, that the English stocking knitters suffered real tragedies as a result of the introduction of the new stocking frames, one of the earliest inventions of the Industrial Revolution.
But such facts and their modern counterparts have led some writers to the opposite extreme of looking only at the immediate effects on certain groups. Joe Smith, a customer service representative, is thrown out of a job by the introduction of an AI system like ChatGPT. “Keep your eye on Joe Smith,” these writers insist. “Never lose track of Joe Smith.” But what they then proceed to do is to keep their eyes only on Joe Smith, and to forget Tom Jones, who has just got a new job in developing the AI technology, and Ted Brown, who has just got a job maintaining and improving the AI system, and Daisy Miller, who can now access customer service more quickly and efficiently than ever before. And because they think only of Joe Smith, they end up advocating reactionary and nonsensical policies.
Yes, we should keep at least one eye on Joe Smith. He has been thrown out of a job by the new AI technology. Perhaps he can soon get another job, even a better one. But perhaps, also, he has devoted many years of his life to acquiring and improving a special skill for which the market no longer has any use. He has lost this investment in himself, in his old skill, just as his former employer, perhaps, has lost his investment in old processes suddenly rendered obsolete. He was a skilled worker, and paid as a skilled worker. Now he has become overnight an unskilled worker again, and can hope, for the present, only for the wages of an unskilled worker, because the one skill he had is no longer needed. We cannot and must not forget Joe Smith. His is one of the personal tragedies that, as we shall see, are incident to nearly all industrial and economic progress.
To ask precisely what course we should follow with Joe Smith—whether we should let him make his own adjustment, give him separation pay or unemployment compensation, put him on relief, or train him at government expense for a new job—would carry us beyond the point that we are here trying to illustrate. The central lesson is that we should try to see all the main consequences of any economic policy or development—the immediate effects on special groups, and the long-run effects on all groups.
If we have devoted considerable space to this issue, it is because our conclusions regarding the effects of AI technologies on employment, production, and welfare are crucial. If we are wrong about these, there are few things in economics about which we are likely to be right.
No comments:
Post a Comment