This article originally appeared in French in the journal Lutte de Classe no. 232 May-June 2023.
Last November, the company OpenAI launched ChatGPT, a programme capable of writing a text on any subject by imitating a human being. This program has passed the fi nal exams of several major schools, and books written by ChatGPT are already on sale. At the end of March, a group of scientists and engineers of new technologies, including Steve Wozniak, co-founder of Apple, and Elon Musk, the boss of Tesla and SpaceX, were alarmed by the threats that the developments of what is called artificial intelligence would represent for humanity and demanded a moratorium. What is behind these cries of alarm from tech capitalists? What prospects does this new technology open for humanity?
Computers have long mastered the game of chess - the Deep Blue computer beat the world champion Kasparov in 1997. They have mastered facial recognition and can translate texts into almost any language. And now they are able to automatically produce original texts, without it being possible to distinguish between what they have written and what a human being could have done. In fact recent advances in computer science and the infinite possibilities for their application in all fields make one’s head spin: computer-assisted medical diagnoses and surgical operations, increasingly autonomous robots, natural disaster prediction, etc. And yet, like every technical upheaval since the first industrial revolution, these innovations are raising anxiety levels, since they could also mean the loss of millions of jobs, the generalising of surveillance, the manufacture of mass disinformation and the production of weapons capable of killing without human intervention. Catastrophe theorists imagine computers eventually becoming entirely autonomous, manipulating and replacing humans and taking over society, reviving the 1950s science fiction of Isaac Asimov!
But computers, no matter how advanced, are still machines. The impact on society of the latest advances in computer science will depend primarily on who uses these discoveries, and for what purpose: the advances in chemistry at the beginning of the 20th century made it possible to produce both fertilizers that could feed the whole of humanity and deadly gases for chemical warfare. The same is true for radioactivity, which can be used to treat cancers and produce energy, but also to make bombs.
Ultra-sophisticated machines, but not intelligent
The expression “artificial intelligence” implies that the functioning of these computer programs is somehow equivalent to the functioning of the human brain. It is true that computers may be able to produce “wonders” but they cannot reproduce human intelligence. Nevertheless there are those who imagine that this is possible - or may be possible in the future. A recent open letter, signed by alarmed researchers and digital capitalists was published in various media, including the French newspaper “Le Monde” (29 March 2023) warned that “Artificial intelligence systems are now capable of rivalling human beings”! They base this on the fact that tasks that used to be thought impossible without human intelligence can now be automated. Yann Lecun, director of Facebook’s artificial intelligence lab, in a lecture at the Collège de France in 2016, defined artificial intelligence as “A set of techniques that allow machines to perform tasks and solve problems normally reserved for humans and some animals”.
To call it “intelligence” just because a machine performs a task for which there was previously a need for human intervention, is to confuse intelligence with automation. In this respect, the Jacquard loom, which made it possible to automate the weaving of patterns on silk fabrics at the beginning of the 19th century, could also have been called “intelligent”, because before its invention, this work was carried out by highly-qualified humans.
The ambition to create “artificial intelligence” is as old as computing. The term itself dates back to 1956. At that time, researchers realised that a computer could be programmed not only as a calculator, but also to solve geometry problems, to plan a series of actions to be performed by a robot, or to imitate a conversation - already! These successes made computer scientist Herbert Simon say in 1965, “Machines will be able, within twenty years, to do any job that man can do”. A prediction that proved to be somewhat exaggerated!
The basic principle of these programs was to try all possible answers to a question until the right one was found. But this only works for simple problems: to generate a text, it is easy to store all the words from a dictionary and all the rules of grammar inside a computer. And while the number of texts that it is possible to construct with these elements may be infinite, unfortunately the vast majority of them make no sense! To resolve this problem, we had to wait for the miniaturization of electronics, the explosion of computing power and computer memory, and the development of the Internet, which makes it possible to centralise information which is scattered all over the world. Thanks to these developments, it is no longer necessary to program all the steps that the computer must follow in order to find the solution to a question: we can use mathematical probabilities.
In order to “work”, ChatGPT relies on a mathematical model of what a meaningful text is. Using databases containing millions of words, it learns how to calculate the probability that the beginning of a sentence is followed by a particular word. For example, “At night, the sky is...” is more likely to be followed by “black” than by “red” because the words “sky” and “night” are more likely to be associated with the colour black. By selecting words one after the other according to these probabilities, it can generate a whole text. The more sentences there are in the database, the finer the model, and the more realistic the texts generated.
This learning by repetition is indeed one of the mechanisms of our human brains, but in the case of ChatGPT, this is training, not understanding. Each time it reads a new sentence, ChatGPT increases its probability of finding the corresponding words, but it has not understood the sentence. Even if the texts produced are original, in the sense that they are not simple copy and paste of texts already written, their content is a function of the database on which the program is trained. The Chatbot is a machine to imitate what has already been written. Unlike mechanical machines that automate gestures, machine-learning algorithms automate repetitive learning processes which take place in our brains, but this does not make them intelligent.
Human intelligence, the result of biological and social evolution
The functioning of human intelligence is obviously and incomparably richer than these training mechanisms. Unlike machine-learning algorithms, it does not simply reproduce what has been done in the past. Take for instance the mastery of fi re, agriculture, writing, and more recently the discovery of electricity or antibiotics. Most of the revolutionary discoveries in human history were the product of trial and error, where the need to adapt to survive as well as undue curiosity played at least as important a role as systematic research. Computers are unable to do this, because curiosity, the instinct of survival, as well as the crowd of feelings and emotions which intervene permanently in our thoughts and imaginings, cannot be expressed in mathematical equations.
Unlike computers, our intelligence is not the product of a real time manufacturing process: it is the result of biological and social evolution that spans millions of years. This is what gives it a capacity to explore the unknown, without any fixed objective. Our nervous systems and our brains are adaptable and malleable, the connections between neurons are made and unmade throughout life. When a gesture is repeated many times, the area of the brain dedicated to this movement is stimulated and strengthened, which allows us to gain in precision, speed, etc. This cerebral plasticity, by which our brains are linked to the whole body, has been favoured by natural selection, because it allows our organism to learn and thus to adapt to extremely different environments and situations.
This is all the more important because an essential characteristic of humanity is not to adapt passively to the pressure of the environment, but to transform the environment according to our needs. Work has played a determining role in the evolution of thought, because it requires that we project ourselves into the future, planning our actions by anticipating the consequences. For instance, for a prehistoric hunter to make a spear that would allow him to kill a reindeer, he first had to find suitable flints, cut them, tie them to a handle, and then test the spear to see if it would serve its purpose.
This thinking and acting relies on social interactions; it is not the product of a single brain. In order to organise collective work, humanity has created languages, allowing the sharing of concepts. This has, in turn, allowed the development of abstract thought: Egyptian peasants were able to anticipate the flooding of the Nile and sailors to find their way around on the high seas by looking at the stars, thus beginning to construct the science of astronomy. This was long before the laws of gravity or the origins of the solar system had been discovered. A computer cannot replicate a living, organic body, with its all functions and its social life, which is what would be needed for it to be able to think like us.
As mankind learnt to control the forces of nature, more and more sophisticated tools were needed. By irrigating and ploughing the land, fields sprang up where there was only desert. By mastering the power of steam, and then inventing the internal combustion engine, it was possible to build machines which could move by themselves. The development of electronics meant that it was possible to program machines to run autonomously for years. And the latest algorithms now allow them improve their performance automatically, over time.
But no matter how simple or complex the instruments which humans use happen to be, whether a flint fashioned out of stone, a plough, or a satellite, they are all adapted by humans for a specific purpose. Unlike the even most powerful computer, humans have set their own aims and objectives, even when they have not fully understood how they will achieve them, or what the consequences of their actions may be. This is where true intelligence begins.
Apprehension, pessimism and the rivalry between capitalists
The fact that we are able to reproduce certain processes of human thought is just further proof of the power of human intelligence. Our brain is self-aware, conscious; it seeks to understand its own functioning and to reproduce it. However, today, instead of these technical feats reinforcing our confidence in the possibilities this represents, they feed fear. The problem is that despite the unprecedented mastery over nature achieved by humanity, we still have no conscious control over our own social organisation.
The conclusion of the signatories of the letter quoted in the introduction is to ask governments to impose a six-month moratorium on artificial intelligence research programs, because, they say, we are on the verge of “developing non-human minds that would render us obsolete and replace us”, thus risking “the loss of control over the future of our civilization”. They insist that decisions regarding the further development of AI cannot be left to “unelected leaders”.
Coming from an upstart capitalist like Elon Musk, the main signatory of the letter, these democratic considerations are obviously only a pretext to hide his own economic interests. On March 26, Goldman Sachs published a study entitled “The Potentially Significant Effects of Artificial Intelligence on Economic Growth”, in which it estimates that 300 million jobs worldwide could be automated by machine-learning algorithms. These figures should be taken with a pinch of salt, but what is certain is that the largest digital companies (the GAFAMs -Google, Apple, Facebook, Amazon, Microsoft) are already in a race to grab the biggest share of this multi-billion dollar market. With ChatGPT, Microsoft got a head start. Google and Facebook quickly followed with their own software, Bard and LlaMA. With Musk missing the boat, he’s probably hoping that a six-month hiatus will allow him to catch up.
Yes, before anyone knows what this software is really capable of, capitalists and executives around the world are already fighting over who will benefit! Who will get the copyright for a book or a computer generated image? Who will have to pay the fi ne if content violates the law? Everyone knows that whoever succeeds in establishing a monopoly will be in a strong position to impose his or her own conditions. The Chinese government has banned ChatGPT, in favour of a competing version, Ernie Bot, developed by a Chinese company. Information may circulate at the speed of light, but in the digital domain, as in all other domains, it is capitalist competition with its borders and protectionism that dictates its laws. In this economic war, philosophical considerations about the future of civilization are only there to serve as a smokescreen for the decisions of the trusts and the states at their service.
A “framework” always at the service of the capitalists
The struggle is all the more fierce because these so-called intelligent algorithms are considered a strategic sector, both by the capitalists and by their states. Whether it is a question of anticipating market developments in order to adapt their commercial strategy, or of having the most modern “intelligent” weapons - such as the robot soldiers deployed by Samsung on the Korean border since 2013, capable of automatically spotting and shooting down a target at a distance of more than 3km - they anticipate that tomorrow, they will be dependent on the trusts that will have succeeded in imposing their stranglehold on the sector. The digital giants are in a better position to know this, because they did exactly the same thing twenty years ago: by dominating the Internet, Google has become one of the world’s largest companies within a few years, with a market capitalisation far exceeding that of an oil trust like ExxonMobil or an investment bank like Goldman Sachs.
A key point in this confrontation between rival capitalists is the question of the data used to drive these programs, i.e., all the texts, images and videos stored on computer servers. This data represents a gigantic and rapidly growing market: in 2020, users of electronic devices generated 64 zettabytes, the equivalent of 64 billion hard drives. At that time, Wilbur Ross, Donald Trump’s former Commerce Secretary, estimated the value of data exchanged between Europe and the United States at $7.1 trillion! Competition is fierce over who gets to use that data, and on which terms. In 2015, the Safe Harbor treaty, which regulated data transfers between the European Union and the United States, was annulled by the European Court of Justice, which considered “that it did not sufficiently protect the privacy of European citizens”. Its replacement, the Privacy Shield, was in turn cancelled in 2020, for the same reason.
European companies which want to take advantage of the predictions of GAFAM algorithms to adapt their strategy to market changes, are forced not only to give GAFAM a share of their profits, but also to send them information about their customers, their products and their manufacturing processes. Apparently they fear that this information will end up in the hands of competitors. In the same way, European states have no desire to entrust the GAFAM monopolies with confidential and strategic data, which could be accessed by the US government. Behind the general declarations of European officials on the “defence of privacy”, there is the defence of the private interests of European capitalists against the monopoly of US companies over the digital market.
Those who demand a framework for the use of these new technologies are counting on state governments - and the most powerful among them, to impose this control. Eliezer Yudkowsky (director of MIRI, the Institute for Artificial Intelligence Research) said in an interview with Time Magazine (29 March 2023), that he considered a six-month moratorium on AI development insufficient, as it would require a complete ban, on a global scale. To impose such a ban, he said, military intervention should be an option: “If intelligence indicates that a country outside the conventions is building a GPU cluster [computer equipment used to train machine-learning programs], be less afraid of an armed conflict between nations than of a moratorium violation; be prepared to destroy an illegal data centre by bombing”. With such a policy, it is not artificial intelligence that is the mortal danger!
This suggests what an international agreement on artificial intelligence might look like. Similar treaties already exist against the proliferation of nuclear weapons. In practice, they serve to ensure that the greatest powers have a monopoly over these weapons, achieved through brutal methods, if need be. It was in the name of non-proliferation of atomic weapons that Israel, supported by the United States - the only state to have used the atomic bomb, against two Japanese cities - recently bombed a nuclear facility in Iran. In 2003, it was in the name of fighting (non-existent) bacteriological and chemical weapons that the United States invaded and razed Iraq. Within the framework of capitalism, the “control of artificial intelligence” can only be the law imposed by the world’s imperialist powers, in the interests of their national capitalists.
Consciously mastering our social organisation: a struggle for the future
The constant threat that the greatest scientific discoveries will be turned against the interests of the vast majority of humanity, is a symptom of the fact that society can no longer progress within the narrow framework of the law of the market and competition for profit. Learning-algorithms could be a tremendous step forward, saving millions of hours of human labour, if used to plan how the economy works. Machine-learning algorithms offer the possibility to computerise repetitive tasks in inventory management, administration, or accounting. Based on information measured in the past, they can estimate future needs by adapting to production cycles, seasons, etc. Factories equipped with sensors already exist; the data measured throughout the production process is centralised on computer servers; algorithms could be used to operate entire factories. But all this could only be fully implemented by breaking the straitjacket of private property, and centralising information at every stage of the chain. It presupposes that humanity consciously takes its social organisation in hand.
Rotting capitalism, however, not only imposes a straitjacket on the economic and material development of humanity, it also puts blinders on its intellectual development. It is becoming increasingly popular, even in academic circles, to claim that it is possible to replace systematic scientific understanding with statistical learning. Frédérique Vidal, a researcher and French president Macron’s former minister of higher education and research, said in 2017 in a speech at INRIA (France’s National Institute for Research in Digital Science and Technology), that “Science is going through [...] an epistemological revolution with the implementation over the past decade alone of a ‘fourth paradigm’ of scientific discovery, based on the analysis and intensive exploitation of data, without the a priori need for a model describing reality”. But the analysis of data can only produce a description of the world, whereas a model identifies causes and effects that allow human beings to act on it. Thus, from the data collected by the astronomer Tycho Brahe, Kepler was able to propose his laws that describe the orbit of the planets. But it is the model invented by Newton that allowed us to imagine a gravitational force, valid not only for planets, but for all bodies in motion, which is used today to fly aeroplanes and to put satellites into orbit. To place the two (scientific discovery and data accumulation) on the same level, is both reductionist and conservative. It is a perspective which excludes finding new ways of acting upon the world around us.
This lack of perspective is characteristic of a society in crisis. The decadent bourgeoisie spits on its own heritage. In the 17th and 18th centuries, when it was still a revolutionary class and fighting against the power of the nobility, this bourgeoisie gave birth to the Newtons, the Diderots, the Voltaires - the Age of Enlightenment. These enlightened thinkers sought to look far into the future, to dig into scientifi c and social problems and to tackle them at their roots. They had the audacity to confront the society of their time in order to open up new paths and, in their own way, to take a leap into the unknown: in 1600, the Christian Church burnt Giordano Bruno at the stake, because he asserted that truth had to be sought in the study of the real world, and not in the holy texts. In the introduction to the Encyclopedia, published in 1751, the Enlightenment philosopher d’Alembert put forward the idea that: “The universe, for those who could embrace it from a single point of view, would be, if it may be said, a single fact and a great truth”. Despite the gaps in their understanding of the workings of nature, due to the technical limitations of the observational instruments of the time, they dared to assert that human intelligence was capable of understanding the world without a God. And the most consistent deduced that they should also be able to organise society without a king.
This intellectual audacity is only possible when a social class has confidence in the future it has to propose to society. This has not been the case for the bourgeoisie for more than a century. The history of the intellectual evolution of the bourgeoisie illustrates the fact that human thought did not undergo a slow and progressive development, from the obscurantism of ignorance, to the light of reason. On the contrary, it reflects social upheavals and the class struggle. It is the fruit of the many fights led by women and men. This is how Marx expressed this, in 1845: “Philosophers have only interpreted the world in different ways, what matters is to transform it”. These struggles are not the result of great individuals, great women and men who once in a century take society a step forward. The greatest intellectuals do not draw their ideas from nothing, but from their capacity to find in the reality of social relations, in the interests of opposing classes that tear society apart, the answers to the questions that confront their time. Marx’s genius was to see that the only social class capable of resolving the deadly contradictions in which capitalist society is mired is the working class, because it has no private property to defend.
Only by placing ourselves on this communist perspective can we confidently envisage a future in which humanity will consciously take its fate into its own hands and rid itself of the shackles of competition between capitalists, collectivise their factories, banks and computer servers and put them at the service of everyone’s needs. A future where these algorithms, which are intelligent in name only, combined with the formidable existing productive forces, will allow the human brain to free itself from the mind-numbing routine of today’s productive work and concentrate on truly intelligent activities. Once the exploited masses are freed from the obligation to devote the best of themselves to daily survival, they will be able to embrace culture, take advantage of leisure, of science, of the arts, which today are the privilege of a small minority. By generalising this intellectual development, by freeing social relations from the prison of material and moral misery, humanity will finally be able to reveal its full potential: how many Archimedes, Mozarts, and Marie Curies could then emerge? In the words of Trotsky: “Socialism will mean a leap from the realm of necessity into the realm of freedom in that other sense too, that the present-day contradictory and disharmonious man will pave the way for a new and happier race”.
May 9, 2023