How can AI and generative AI be useful without the risk of destroying economies and societies?

According to current estimates investment in AI in 2025 will exceed one trillion dollars and might possibly top two trillion dollars in 2026. Market capitalization of the largest 65 AI firms exceeds 20 trillion dollars. These are big numbers. AI is responsible for one of the largest build-up of capital in history. Investments in AI were motivated by the expectation that AI would generate huge profit. The expectation of profit, in turn, has produced a great increase in market capitalization. NVIDIA is the first firm to reach a market capitalization in excess of 5 trillion dollars.

Much of this massive investment in technology is motivated by three major objectives:

1. Increase military power, as AI is considered an essential component of modern warfare, a component that, contrary to nuclear bombs, can be deployed without the risk of annihilating humanity

2. Industrial dominance, as AI (and automation in general) is believed to give companies a competitive edge virtually in every sector.

3. Reduce the amount of labor employed both in industrial production and administrative task, with the objective of reducing cost and increase profit

Globally we see at the same time a replacement of humans with machines and an increase of the potential to produce goods and services.

The critical questions are the following: how is profit produced in economies were less and less workers are needed? What is the social impact of replacing workers with machines?

According to the classical theory of capitalistic economies, workers produce more than they consume. The excess production forms investment and the consumption of capitalists. Leaving aside the Government, this is the famous equation of Michal Kalecki: P=C+I, that is, in aggregate profit is equal to the sum of investment and the consumption of capitalists. Marx observed that if salaried workers receive the amount M of salaries, they can only spend M’<M. Where does monetary profit in aggregate come from?

In practice, in growing economies, salaried workers borrow money for buying houses, cars and even to finance consumption. This borrowed money, created ex nihilo by the banking system, is the monetary profit of capitalists. In addition, the true sources of financial gains of capitalists come from financial markets where expectations of future profits are traded.

In summary, in the classical view, in aggregate, workers consume only a fraction of the goods and services they produce, the remaining fraction of production is either invested or consumed by capitalists. The financial part has progressively become a parallel system that produce money through credit and that trade expectations.

With increasing use of automation and AI, workers are progressively replaced by machines. Profit does not depend on the consumption of workers. Machines and fewer and fewer workers produce the goods and services that are partially invested and partially consumed by capitalists.

In practice situation is more complicated because of government spending and international exchanges. The flow of goods and money should include government expenditures for infrastructure, military spending and so on. However, the above is the big picture of profit and concumption.

What is the social impact of this situation? There is a continuous increase of social inequalities. A small number of people make money through substantial salaries and plans to share profit. However, the majority of people have less and less money to spend, at least in relative terms because of lack of well paid jobs. We see the growth of a parallel economy made of marginal, precarious jobs that are not immediately touched by automation.

What can be done? This is a big issue that requires much more than a short post to be discussed. At least, it is important to create a growing awareness that the problem exist and it is fundamental.

A number of people and researchers are trying to understand how to mitigate the replacement of humans with machines essentially claiming that AI will not replace humans but will increase their thinking power. According to this view, we are creating new ecologies based on symbiotic humans-machines interactions.

Well, this is nothing new. Over the centuries, the progress of technology has increased the ability of humans to reason and to solve problems. For example, the invention of the printing press has greatly increased the ability of humans to storage knowledge and to make knowledge available. The relationship of individual and collective knowledge has changed with the availability of affordable books.

Progressively industrialization has increased the ability of creating more refined instruments and perform better experiments that could be easily replicated. The strength of empiricism is ultimately based on the ability to perform experiments on a large scale. The discipline of empirical testing of hypotheses has changed the attitude of people towards ideologies.

With the invention of computers humans have acquired the ability to perform very quickly huge numbers of calculations. Mathematics has become a practical computational tool. An engineer can now study a structure and simulate different design options. The ability to simulate has changed the perspective of design and engineering.

Now comes AI and Generative AI. Let’s make a few basic points. First AI as known today is based on manipulating symbols that have no external reference. There are areas of say “intelligent automation” such as automatic driving that are based on sensors, but sensors are points in a conceptual scheme which is a given, that is movement in a tri dimensional world.

Generative AI is based on embedding words in vectors and manipulate them with numerical operations. Now, embedding is based on learning from a huge number of linguistic contexts. It is the cross check of millions or even billions of co-occurrences that creates a reasonable map of human meanings onto a purely formal vector space.

It is possible that this process becomes somewhat creative. It is possible that inner contradictions in the empirical co-occurrences generate new meaning. This is not true discovery; AI will not have the possibility of checking if these new meanings are useful. Some humans might find it useful, but it is difficult to believe that these emerging properties will make any difference.

The second point is that Generative AI is self-referential insofar as it generates many new linguistic corpora that become the new training set. This process needs to be studied. The most likely outcome is flattening of ideas, but it might also produce biased reasoning in many different areas, especially in those areas subject to ideologies.

In summary, it is difficult to believe that AI will lead to new ideas. The vision according to which humans will have daily sessions with their preferred generative system to discuss ideas and even search emotional comfort is more frightening than reassuring. Daily sessions of “confession” with an automatic mentor open the way to manipulations on a large scale.

There are areas where AI can be genuinely useful. It is important to recognize true capabilities of this new technology without indiscriminate deification of machines. AI is ultimately the mathematical modelling of languages. In this sense it can be very useful.

And there are AI technologies not specifically linked to languages that can be very useful. Neural networks can be very useful in studying complex dynamic processes, for example in the context of motor re-education after major surgeries.

However, no genuinely useful use of AI will be able to solve the big issue of employment where humans compete, and lose, with machines. The quest for profit will outnumber the useful use of AI.

To conclude, we have a new technology. We humans should acquire the ability to separate the discussion of genuinely useful applications from the discussion of true social and economic issues. If we mix the two under the umbrella of indiscriminate search for profit we might end up with really serious problems.