Jan 14, 2020
The following article is a translation of an article from Lutte de Classe (Class Struggle), issue # 205, the magazine of Lutte Ouvrière (Workers Struggle), the revolutionary communist workers organization active in France.
Self-driving cars, automatic translation, image- and facial-recognition software, autonomous navigation systems: whether it inspires enthusiasm or anxiety, artificial intelligence (AI) seems to be playing a larger and larger role in society. After the appearance of robots capable of replacing workers on assembly lines in the 1980s, so-called intelligent systems are today performing the labor of accountants, financial advisors, and even lawyers. Twenty years after the 1996 victory of the IBM computer Deep Blue over the chess champion Garry Kasparov, another machine, AlphaGo, this one belonging to Google, became the world champion of the board game Go, which is reputed to be very difficult to master.
However, rather than representing a considerable advance for humanity which allows for the extension of the human mind, AI appears to many as a threat which could destroy millions of jobs in the world in the immediate future. Some people are going so far as to predict the end of work. Others are worried about AI’s increasing grip on our lives, and, behind it, the companies which have mastered it, like Google, Facebook, and Amazon, which are accused of surveilling the entire planet to collect, compile, and store the personal data of its inhabitants. But discussing a technology, no matter how high-performing, without discussing the society in which it appeared, the social conditions under which it has been implemented, and whom it benefits, does not make any sense. Behind the technological progress which allowed for the spectacular advances in AI, there is the exploitation of workers at the one end and the accumulation of profits at the other.
The term “artificial intelligence” was invented in 1955 by John McCarthy, a professor of mathematics who worked on Turing machines, the predecessors of computers. Alan Turing (1912–1954), a specialist in cryptography and algorithms, created the following rule in 1950 to decide whether a machine was intelligent: that it could pass for a human during a blind conversation with a real human.
The notion of artificial intelligence is an exaggeration, to say the least. It has nothing to do with human intelligence. It is a so-called “weak” intelligence, essentially able to sort and process an increasingly gigantic mass of data, in record time, with algorithms thought up by mathematicians. This is what the physicist Hubert Krivine calls “the irrational efficiency of data.” AI is capable of predicting results better and much quicker than human intelligence, but it is not capable of understanding, and even less of innovating. It establishes correlations between phenomena without understanding the links of causation which connect them. It can lead to huge errors. And when the situation is unprecedented, the machine fails. For the same reasons, AI reproduces the biases and prejudices of the data used to train it. For example, Tay, an AI developed by Microsoft to exchange messages on social media, took less than a day before it started posting racist and Holocaust-denying tweets, feeding off of all of the messages that it found on the Web.
Although the human mind also processes and analyzes data, its ideas and reasoning, just like the feelings and intuitions that it experiences and expresses, are not the result of a simple accumulation of information. An individual’s personal and social experiences contribute a great deal to the decisions they take. A support system for medical decisions can be quicker and more effective than a doctor for analyzing symptoms and medical imaging, but healing a patient cannot be reduced to analyzing their pathological data; this requires asking them questions and listening to them, in order to know their past history and their situation.
Since the 1950s and the early stages of AI, progress in computing took place over successive stages. The power of calculators has continued to increase. The invention of the transistor allowed for the miniaturization of electronic components. In 1971, Intel created the first commercially available microprocessor, which executed thousands of elementary operations in computer programs. But these advances in AI were followed by stagnation, since the promises made by its promoters (as early as 1958, certain people announced that “a digital computer will be the world’s chess champion within ten years”) were slow to be realized. There were many technical obstacles, notably the weak calculating power of the machines. As a result, public and private financing declined.
Between 1980 and 1987, AI experienced a new boom, with the establishment of “expert systems,” or software programs capable of responding to questions, making a reasoning based on known facts and rules grouped together in data sets in a precise domain. In the same period, researchers explored new ways to make machines capable of sorting, classifying, choosing between two options, and even to improve their performances themselves. This was connectionism, which imitates the biological brain by recreating networks of artificial neurons, trained by algorithms to recognize images, contours, and faces. This method, called “deep learning,” has been known for 30 years. But the power and calculating speed of computers have long remained too weak to produce convincing results.
This phase of AI’s development in the 1980s coincided with a growth of new technologies which attracted masses of capital which were already looking for places to invest. Hundreds of millions of dollars flowed into this sector, until this bubble popped for the first time in 1987. Credit was cut off. AI disappeared from newspaper headlines.
It was not until the middle of the 2000s, with the increase in the capabilities of processors and then the development of the Internet, that it returned to center stage. To give an idea of this kind of calculating power, the University of Reims Champagne-Ardenne recently acquired ROMEO, a computer capable of carrying out one quadrillion operations per second. The massive accumulation of data, or “big data,” makes deep learning both possible and efficient for AI systems. This data collection has skyrocketed: every day, 2.5 quadrillion bytes of data are collected. Ninety percent of all available data in the world was collected over the past two years.
The collection of the personal data of hundreds of millions, if not billions, of users, in order to ultimately transform it into advertisements and profits, has become a profession. This falls to the Internet operators Google, Facebook, Amazon, and the other GAFAM companies (to use the acronym made up of their names).
In 2001, Larry Page, the co-founder of Google, declared: “If we did have a category, it would be personal information… . The places you’ve seen. Communications… . Sensors are really cheap… . Storage is cheap. Cameras are cheap. People will generate enormous amounts of data… . Everything you’ve ever heard or seen or experienced will become searchable. Your whole life will be searchable.” In 2003, three Google computer scientists filed a patent entitled “Generating User Information for Use in Targeted Advertising.” Their invention involved “determining user profile information and using such determined user profile information for ad serving,” or in other words, collecting data about the behavior of users in order to use it in targeted ads.
After it became a publicly-traded company in 2004, Google became one of the top five companies in terms of stock market capitalization in 2018. Its market capitalization is equivalent to Argentina’s GDP. It is 4 or 5 times larger than that of a traditional company like Total, even though Google produces only services of limited value, and nothing material. Such a market value, which is largely virtual, is the result of speculation. Those who buy Google or Facebook stocks are anticipating that the price of these shares will rise, so that they can sell them and make a profit. These unbelievable market valuations also reflect the fact that the masses of disposable capital can find no outlet in other productive sectors. It is one of the signs of the incurable sickness of the capitalist economy. In addition, the stock market value of these companies can collapse just as quickly as it rose: Facebook’s capitalization dropped by $120 billion in just one day, toward the end of 2018, after a giant breach of personal data was revealed.
That said, the financial power of the GAFAM companies allows them to form monopolies by buying up, sometimes at high prices, hundreds of other companies which specialize in the collection of personal data. Facebook bought the application WhatsApp in 2014 for the trifling sum of 19 billion dollars. Google bought Waze, a start-up that developed a GPS navigation app that competed with Google Maps, for 1.2 billion. Alphabet, the parent company of Google, has bought up 230 companies since it was created. Besides Google, which owns Android and YouTube, Alphabet has developed a dozen subsidiaries which carry out research and development in healthcare, artificial intelligence, robotics, materials, transportation, cybersecurity, and even agriculture. Unsurprisingly, despite its motto of “Don’t be evil,” leaks have revealed that Google worked with the U.S. Army to develop killer drones. In 2018, workers at Microsoft and Amazon denounced their companies’ sale of facial recognition software to the U.S. Border Patrol.
The market for connected devices continues to grow. There were 22 billion connected devices in the world in 2019, compared to 15 billion in 2016, and 40 billion expected in 2025. Once again, the main goal of companies in this sector is to collect data. Sleep Number, for example, which produced so-called intelligent beds, collects biometric data on the movements of sleepers, their positions, breathing, heart rate, and even the noises in their rooms. All of this information allows the company to create databases to train AI medical decision support systems.
Tapping an ever-growing mass of personal data allows digital technology companies to monetize it at high prices, either through targeted advertising or by developing various kinds of AI systems. Data collection and buying up promising start-ups are not the only source of enrichment for these firms. Like all capitalist companies, it is by exploiting workers all around the world, either directly or through subcontractors, that they enrich their shareholders.
Behind the cool image of Californian geeks like Mark Zuckerberg, Larry Page, and the founders of the GAFAM companies, there is hidden exploitation. Although computer engineers are not the most exploited workers in the world, the blue collar workforce of Silicon Valley, and the cafeteria employees of Facebook and Alphabet, security guards, custodians, and drivers, have to hold down three jobs to survive. The digital technology industry is not virtual. It requires material support, computers, telephones, networks, storage facilities for ever-increasing amounts of data, etc. It needs thousands of tons of semiconductors and other electronic components, copper and other metals, rare-earth elements, coltan and tin, extracted from the mines of Congo, which are transformed or assembled in China, Bangladesh, and elsewhere, under horrific conditions. No progress in AI is possible without the very real and underpaid labor of workers around the world.
A report on the French television news show “Cash Investigation” on September 2019 shed light on another form of exploitation. This is that of the underpaid workers who train AI systems. Before software can recognize one particular face among all the others, or a meaningful pattern on an MRI scan, it is indispensable to subject it to “supervised” learning. This initial education falls on humans whose job it is to choose the figure they are supposed to identify out of thousands of photos, or to click in order to validate. Paid 1 to 12 cents per click, the most experienced people, working on their own computers for 8 hours a day, 5 days a week, earn between $400 and $500 per month. In this work of teaching AI, someone can recognize images used to train drones how to kill without the person training them even knowing it. In the same vein, Facebook uses the labor of 15,000 “moderators,” paid less than $900 a month, to look for hours at images or videos posted by users to remove those which Facebook deems violent, pornographic, or degrading. Apart from the criteria for censorship being imposed by Facebook, the moderators must constantly face often-serious psychological shock due to the unbearable images they have to see, all without the least bit of medical attention. The Internet allows the companies to distribute these unrewarding jobs all over the world. The tech giants subcontract them to multiple companies and wash their hands of the conditions under which this new category of workers, the slaves of the click farm, are exploited.
Advances in digital and computer technology, coupled with those of globalization, led to Amazon’s success. But its algorithms and software serve primarily to organize the exploitation of workers. Amazon warehouse workers are under the constant surveillance of their scanners, the small portable device that tells them which goods to collect. This scanner not only tells them what work to do, but it also follows them around the warehouse, times each operation, and scolds them if they stay in the bathroom for too long. These exhausting and stressful work conditions accompany rock-bottom wages. At Amazon, AI has updated Charlie Chaplin’s Modern Times, a parody of the work on auto factory assembly lines in the early 20th century, but it has not changed the exploitation.
During his presidential campaign in 2017, Benoît Hamon called for the creation of a Universal Basic Income by invoking “the inevitable scarcity of work,” caused by the development of AI and robots. The CGT union federation’s monthly magazine, Ensemble, recently published an interview with the philosopher Bernard Stiegler, the director of the Institute for Research on Innovation and the author of a book called Work is Dead, Long Live Labor!, in which he writes: “Under the effect of total and generalized automation, … workers will become a sort of residue from a past epoch. There will be, of course, jobs, because in certain sectors, there will continue to be a need for a proletarianized human workforce, but this will become exceptional.” Stiegler’s conclusion is that income must be disconnected from jobs, in order to distribute “resource allocations” or “contributive income.” This sort of universal income would be paid by government bodies, meaning that it will come out of social service budgets, that is, the socialized part of the wealth produced by workers. But there can be no question of making the capitalists pay!
Stiegler cites various studies, such as that published in 2013 by two researchers at the University of Oxford, Frey and Osborne, claiming that 47% of U.S. jobs are at high risk, meaning that they are “potentially automatable over some unspecified number of years, perhaps a decade or two.” Although it cannot be denied that automation eliminates jobs, these spectacular figures are challenged by other studies. A report from the Organization for Economic Cooperation and Development (OECD) published in 2016 indicates for its part that “while Frey and Osborne find that 47% of U.S. jobs are automatable, our corresponding figure is only 9%.”
The jobs eliminated by the introduction of robots have long been those held by industrial and retail workers. Even in China, the workshop of the world, with low-paid workers, the government in 2015 launched a plan called Robots 2025 in order to boost automation in its factories. Questioned in January 2018 by a journalist from the French television show “Envoyé Special,” the director of an enormous factory in southern China belonging to Hisense, a subcontractor for Hitachi, Sharp, and Whirlpool, explained that he invested in a robot as soon as its cost fell below that of two years of a workers’ wage. In three years, this factory cut 3,000 out of its 8,000 jobs. But one should not lose sight of the forest for the trees: all over the world, even in a factory with many robots, the hard and unrewarding tasks requiring little training continue to be performed by underpaid workers. This is in essence what the director of a digital technology research institute told the newspaper Les Échos: “The lower the cost of labor, the less interest there is in replacing the worker.” Nevertheless, the use of intelligent software eliminates the jobs of accountants, financial analysts, and bank and insurance employees. These are skilled and relatively well-paid jobs.
However, although digitization and automation do eliminate jobs, how many have been cut due to plant closures following corporate restructuring, to outsourcing, and to gains in productivity, without the introduction of either robots or intelligent software? The main cause of job cuts is not automation but the aggravation of exploitation in an economy in crisis and stagnation. The tens of thousands of jobs now being eliminated by the European banks are being cut just as much, if not more, because of economic slowdown and their bosses’ uncertainty about the future as because of the effects of AI.
Using machines to produce more quickly and at a larger scale, to lower production time, has characterized capitalism since its origins. Robots and systems equipped with AI are just perfected machines. In every period, the introduction of new machines has taken place on the backs of workers. Some were thrown out of work, while those hired to run the new installations were even more exploited than before. As the system extended and developed, new jobs were created. What characterizes the current period, much more than the performance of so-called intelligent machines, is the stagnation of the economy and the capitalists’ inability to develop its productive forces, and even to update those already in use, due to a lack of potential markets.
As long as the means of production belong to private capitalists, the gains in productivity allowed by machines—intelligent or not—will not benefit workers, and the inventions that could reduce the general hardship of labor will only heighten exploitation for certain workers. Capitalism has always consisted of the marriage of extraordinary scientific and technical prowess with the worst exploitation of human beings.
For all that, no more than the introduction of machines in the nineteenth century did not stop workers from organizing to defend themselves collectively, the introduction of robots and AI in the twenty-first century will not put an end to the class struggle. Although Luc Ferry, the former French education minister under Sarkozy, said during a symposium about automation in 2018: “A robot doesn’t drink, doesn’t smoke, doesn’t join the CGT, and doesn’t go on strike. The bosses dream of it,” he may be disappointed. The workers who hold the underpaid jobs created by digital technology, like those who slave away in workshops beside robots, or those who build them, continue to create surplus value, and therefore profits for the bosses. They are indispensable for the functioning of the economy, and this gives them a central role to change society.
The global economy has been ripe for socialism for over a century. All of the elements for planning exist, but the big companies use them for their own profit, without getting rid of the competition between them or the waste that this creates. Digital technology and artificial intelligence are not only the means to improve human life, to eliminate difficult, dangerous, or tedious work, and to improve human abilities. They are also tools for planning which Marx and Lenin could not even have imagined. They offer powerful instruments for humanity to take stock of the resources, energy, and needs of everyone, at both a local and a global level. They could allow us to organize the production, transportation, and distribution of the goods necessary to everyone in a rational and planned way, all while preserving the planet and above all reducing the labor of each human being. Combined with the capacities for production which already exist, they would allow us to reduce productive labor to a minimum, all while permitting each human being to make their contribution to the functioning of society. Marx’s slogan, “from each according to his ability, to each according to his needs,” could finally become a reality.
But none of this is possible unless the working class—all categories together—takes the control of the means of production out of the hands of the big bourgeoisie.