Sketching the new dysto-utopian world with massive presence of artificial intelligence

A four-class society (a long read)

The massive deployment of artificial intelligence (AI) presents economists with perhaps the unique situation in the past five or six decades. The artificial intelligence is an all-purpose technology and its effects are, from the current perspective, quite difficult or even impossible to predict. If one wonders whether there would be increased unemployment due to the introduction of the AI it makes no sense to talk to labor economists. Labor economists might be able to answer questions such as whether any specific technology would lead to unemployment or whether the increase in the minimum wage would make some people redundant but they cannot address the question of the effects of a new general purpose technology because both the extent of the application of such a technology and its effects on labor demand are simply unknown. The new general purpose technology could be, like electricity, included in 90% of the activities or it might fail and the hopes, and fears, that we entertain for the artificial intelligence may be disappointed.

This then requires not looking for concrete answers which I think at this stage are impossible to make but defining a meaningful set of questions regarding the possible effect of the AI, and then sketching the new world it might produce.

The first meaningful question is whether AI would lead to an increase in total output or would keep total output more or less the same but would reshuffle the roles of machines (including artificial intelligence itself which can be called a machine) and live labor. In other words, whether AI would be an output-augmenting technology or factor-shifting technology, meaning that it substitutes capital for live labor but does not lead to the production of many new goods and services, or increased production of the already existing goods and services. I have dictated most of this essay; I did not write it as I normally do in Word, nor did I write in longhand as I used to do twenty years ago. But in this example, neither word processing technology nor machine-dictation have made me produce more essays. There was probably some improvement in productivity, in the sense that a readable essay was produced in less time, but little else. Yet the all-purpose technology of the information revolution has made this essay much more accessible to many people. In that sense, there was an increase in overall welfare (assuming that reading this Substack does not produce anger and unhappiness in many of its readers).

The second related question is whether AI would lead to increased productivity defined as output per live unit of labor. There, as both the common sense of labor substitution by AI, and the previous example, suggest, we can relatively safely assume that AI would increase output per unit of live labor. And in some cases, dramatically so. In order not to come to this conclusion we would have to assume that total output would shrink by the application of AI to the same extent that the use of live labor is expected to shrink. This seems most unrealistic. Consequently, the ratio between total output and total live labor, we can safely predict, would go up.

The third issue is whether AI would create a significant surplus population. I prefer to use the term surplus population to a more neutral term of higher unemployment because it reflects better what may be the effects of a general purpose technology. It might create people who are basically unemployable and whose skills belong to a revolved era. These may be skills and knowledge that are entirely made redundant by the deployment of the AI; or people whose skills require a “retooling” for the needs of a new economy that would take five or more years. In either case, such people are over a medium- to long-term a surplus population. If the surplus population becomes large, for example reaching 15 or 20 percent of the labor force it would create political problems. The issue will then be how to solve this political problem. It may be dealt within the existing system of unemployment benefits with the proviso that in the case of the surplus population unemployment benefits might run for years rather than for six or nine months as they do nowadays. Or through, as many people argue, universal guaranteed income. Both would require a change in the way the welfare system is organized and funded. We would have to accept the possibility of people who would never work during their lifetime or might work for very little time in some subaltern occupations.

Money transfers however do not eliminate the political issue of a large surplus population. It would be naïve to believe that simply giving people money to survive would make them happy. Surplus population could create a permanent pool of malcontents who would be politically difficult to handle, a malleable population who would be there to be exploited and used by political operators. The very presence of a large group of people who have nothing to do and lots of free time to plot rebellions would by itself be politically destabilizing.

The next issue is the question of profits. Very few people doubt that the total amount of profits, when the AI comes to be deployed, would increase. AI is like any other capital and the people who own the AI would of course make money. They would become richer. In some cases they might become extraordinarily rich and gain a huge political influence. That huge wealth and political power of a few combined with the existence of a large surplus population would create an extremely dangerous political cocktail. It is not necessarily true that the new techno-lords would always act as a class. Some among them might have ambitions to lead the army of malcontents and displace other techno-lords from their positions. The fall of the Roman republic provides one such historical precedent; the struggle between different oligarchic clans in post-Communist Russia another.

But here my question is whether the profit rate would go up or not. The higher mass of profits due to the AI does not necessarily translate into a higher overall rate of profit because the amount of new capital that would be sunk into AI may be so high that it outpaces the increase of total amount of profit. In other words, the profit/capital (π/K) ratio may go down. We might thus face what, at first sight, looks like a paradoxical situation: the average rate of profit in an economy where AI is deployed might go down. But there are countervailing forces. Schumpeterian creative distraction brought about by the AI, might destroy the value of lots of capital that would no longer be usable. For capital, AI will thus do the same that it did for labor: create a surplus capital the way that it has created a surplus labor. Such surplus capital no longer enters into the calculation of the average profit rate; the denominator K in the calculation of the rate of profit would be reduced and the average rate of profit might go up. But notice that the creative destruction that maintains the rate of profit at a high level means large capital losses for the people who own “wrong” capital. Thus the malcontents may not be only the unemployable laborers but also former capitalists who are now pauperized.

What do we get when we put all of this together? There would be almost certainly an increase in the share of national income going to capital-owners because the quantity of capital employed would have increased and the average rate of profit would have remained more or less the same. The share of labor would go down. Since most of capital, as we know empirically, is held by people who are at the upper ranges of income distribution, rising capital share would exacerbate income and wealth inequality. This too is likely but not certain. New technologies of such magnitude might actually increase capital income not only at the top, among the current owners of capital, but also among those who are in the middle of the wealth distribution. In that case it becomes less evident that the increasing capital share would necessarily by itself drive income inequality up.

But there is a second element that would push inequality up for sure. This is the surplus population whose new state-guaranteed payouts would be much lower than their former wages.

Finally, we come to the key question which is in truth a social question: how would the new income distribution look like? This would be a tetra society, that is society composed of four classes. On the bottom there is the surplus population, the lumpenproletariat or the idle helots with no jobs and nothing to do. They would have to be kept quiet and docile through dispensation of bread and circuses. Bread and circuses in the modern case obviously means some kind of minimum unemployment benefit for life and streaming of low-cost entertainment 24/7.

Then there would be a relatively large service-sector working class employed in the areas that cannot (as yet) be replaced by the AI. They would subsist on low to moderate wages, and they will be permanently living under the threat of their replacement by the AI and thus descent into the lumpenproletariat.

The third class are people who would combine, as they do now, high capital incomes and high labor incomes. This is what I called the homoploutic class that currently represents about 3% of the US population: people who are both among the top 10 percent of capitalists and the top 10 percent of wage-earners. That particular group might increase in size. Even if they currently hold no shares in new companies their high labor incomes as well as their high capital incomes allow them to expand their shareholding in the most profitable sectors and thus not only to retain their position but to improve it and to increase the circle of people who are homoploutic.

Finally at the top of the income pyramid would be extraordinary rich people whom we already see now. This may not be just five or six or ten extreme plutocrats but there could be possibly thousands of people with extraordinary wealth.

Such a social pyramid would look, both in the sharpness with which incomes rise as we move towards at the top and also in the flatness of incomes among many on the bottom, like pyramids in pre-industrial societies which failed to create sufficiently large middle classes. The sketch applies, of course, only to the more advanced and richer societies. Other societies –not as advanced in terms of AI—might retain a more familiar social and class structure.

The interesting thing could be that more advanced and richer societies would lack the middle class and be politically more turbulent than the less advanced societies. One of the dogmas of the twentieth century was that development (higher overall income) and education create a strong middle class that is in turn a guarantor of political stability. But perhaps that that dogma was true only for a particular type of industrialized Fordist societies but not generally for the societies with high overall income.

What could be in principle called the middle class in societies with lots of AI would be service sector workers who have not been displaced by artificial intelligence. Yet they cannot be treated fully as a middle class because their incomes are relatively low compared to the incomes of the top two classes of homoploutic individuals and techno lords, and their position is precarious as they are in the constant danger of falling down, rejoining the lumpenproletariat.

There is little doubt that such a society, as we sketched it, would display strong dystopian characteristics. The utopian promise of machines doing lots of work that is currently done by humans, and allowing us to enjoy free time and leisure, would turn out in reality to produce socially extremely unequal and politically much more unstable societies than the current proponents of abundance created by the artificial intelligence want us to believe.