“The More AI Progresses, the Greater the Need for Human Labor”: Extensive Interview in the German Journal Soziopolis (24 Jan. 2024)

The German journal Soziopolis presents an interview-essay conducted by researcher Nikolas Kill with me. This English translation reproduces the original German content.

“The further the development of artificial intelligence progresses, the greater the need for human labor”

Mr Casilli, in the debates about the development of artificial intelligence, it is often heard that the increased use of smart machines would remov many jobs in the future. In your research, you turn around the question and investigate how much human work is behind artificial intelligence. Basically, what is artificial intelligence?

I fear that there is no easy answer to this question. First of all, it must be said that artificial intelligence is now an overused term. AI can mean anything and nothing. Things that were described as algorithms three or four years ago are now called AI. Things that were known as software ten or twenty years ago also operate under this label today. The term artificial intelligence means hardly anything specific in the commercial context of the 2020 years. Perhaps one better not approaches the matter directly, but by a detour, by holding on to what the term does not denote. To be sure, today’s programs and systems have nothing more to do with the ambitious scientific programs of the 1950s, which pursued the goal of developing a thinking machine. Even if the interested side repeatedly claims otherwise: The idea of actually being able to develop a machine that thinks like a human independently has now become a different research has now become different.

What are the main differences compared to the past?

Today’s AI systems and the underlying understanding of intelligence are the result of a number of content-related adjustments and shifts. The first shift was limiting to a certain aspect of human intelligence, namely learning. Instead of a thinking machine, they wanted to develop a learning machine. Following this change, the concept of automatic learning, which in reality is based solely on an in-depth game type of statistics, prevailed. At its core, it is about collecting and evaluating enough data to read out certain regularities. Even if it is always claimed that the machine has decided these regularities, it has of course not done anything like that. It only highlights statistical properties of the data sets.

Two years ago, you wrote that there was no artificial intelligence, just the click work of other people.[1] They speak of micro-work (micro-work). What does the term describe?

Microwork is a special type of digital laboratory that is required for training and programming artificial intelligence. In recent years, it has become synonymous with data work. Microwork is, to some extent, a hybrid between ordinary online work, which has been common in many areas since the pandemic, and “uberated” work. The latter describes the fact that many micro-workers are not employed as employees of a company or sub-company, but work as an independent micro-entrepreneur, each applying for individual work orders and working on their own account.

In general, these are tasks for which both the time and the spectrum of human skills used are extremely manageable. At the same time, this also requires talents and abilities, which cannot be easily required, but must first be developed. A good example of such a micro-task is the classification of dozens, sometimes hundreds of images based on specific criteria. For example, a sorting of images by hand is carried out in order to filter out adult images from a database. Another example is the sorting of images, in which a green or a red traffic light can be seen. Such tasks are typically assigned in the context of the production of image recognition software for autonomous vehicles.

However, the use of microwork is not limited to images, often it is also about texts. Since texts serve as the basis for the automatic generation of new texts by chatbots and other text-based applications, this is even very often the case. A typical task in this context, for example, is in the completion of sentence beginnings. The phrase “Today is a more beautiful” is then supplemented, for example, by the word “day”. A microworker can receive hundreds of such sentences to be completed for editing. If you ask for the meaning of the whole, well, it is explained very quickly and easily. The corresponding work is used to train generative AI systems such as ChatGPT. These systems are intended to enable the mass input of data to complete sentences based on tips. Users: provide a “Prompt”, a note, which is then completed by the system. Of course, the efficiency of a system like ChatGPT is very impressive because it is able to not only complete sentence beginnings, but also to provide answers to complex queries. But the system works according to the same principle. Based on various requests, the system produces meaningful connections based on a vast amount of entered data and statistical probabilities. Similar programming tasks are also required for sounds or videos. Videos are categorized, annotated and commented, sounds are transcribed. These are all examples of micro-tasks.

With your research team, which is called DiPLab (Digital Platform Laboratory), you have been conducting field research in numerous countries for several years. They observe the reality of platform work and conduct interviews with affected actors. For some time now, you and your colleague have been busy arranging and systematating the enormous number of cases that you have covered during your field research. How do you proceed and which categories do you assign the investigated cases?

The systematization we have reached so far provides for three large categories of microtasks. First, there are microtasks that serve to train AI systems. These are mainly due to the development phase before the product hits the market. Here, too, the case of ChatGPT is relevant, as the reference to the corresponding practice is already included in the acronym GPT (short for Generative Pre-Trained Transformer). A data set was compiled and annotated for training purposes from huge data collected on the Internet after the 2010s. The required microtasks fall into this first category.

Once artificial intelligence is on the market, its ongoing operation must be ensured and ensured that it works according to the wishes of the manufacturer’s company and, above all, of course the user. This is the context of the second category of microtasks: tasks for review. For example, individuals ensure that the chatbot functions properly, or carry out activities that fall into the area of moderation or maintenance. In our field research, we have also encountered cases where people check the transcripts of voice assistants used by Apple, Amazon or Microsoft for possible errors and inconsistencies.

What exactly do they do with them?

You can imagine it as follows: Let’s assume you are the customer. You ask the voice assistant: “What is the weather in Lyon today?” The voice assistant transcribeds your request and then processes it. The programme must always check whether it has understood the request correctly. If there is background noise – for example, from a running washing machine – or if someone, like me, speaks with a special accent or dialect, voice assistance systems often reach their limits. In these cases, the click workers must: check whether the machine has reproduced and processed the information correctly.

In addition to training and review, there is a third category of tasks that seems a bit more problematic or at least more irritating to people than the other two. It is about the imitation of artificial intelligence. There are actually actors among AI providers who claim that they have developed a super powerful algorithm, a model that supposedly performs true miracles while in reality allowing certain work to be carried out by hand. This approach is sometimes cheaper for companies than the development of the promised software. For example, my team and I have noticed this in connection with AI solutions that supposedly manage appointments or transcribe images for their customers. Corresponding examples, which we have come across in our field research as well as in specialist literature, come in part from large corporations.

That sounds bizarre. Could you describe the practice in more detail using an example?

Safe. Google, for example, launched a voice-controlled artificial intelligence called Google Duplex in 2018, which was supposed to be able to keep calls for its customers, make appointments or make reservations. The results were incredible, and for good reason. As you soon found out, around 40 percent of the calls were made not only by AI, but by humans, or at least supported by people.[2] Because the machine was not up to the task at a certain level of complexity, micro-workers had to step in: a call center and help out behind the scenes.

How popular is this kind of AI mimicry through microwork?

Oh, that’s not so rare. The companies just don’t like to talk about it. That is why people are also irritated when this kind of microwork becomes public. They feel cheated because they are not aware that it is a routine practice. Sometimes the functionality of a dysfunctional system has to be simulated. This is especially the case if there is a bug or other technical problem. In such cases, the voice assistant may misinterpret requests such as “choose the photo of a beautiful sunset and send it to my sweetheart”. He is then, for example, unable to process the information “my treasure” correctly. If you want to prevent potentially fatal misunderstandings from coming from happening and the customer being disappointed, you have to run the machine, as you say, in graduated mode – while a person looks at her fingers.

In the test phase, such a procedure occurs very often to further develop artificial intelligence and to resolve errors. In addition, numerous other tests are also carried out, such as tests of user-friendliness, to find out how users use the product, or acceptance tests to determine whether users can accept the result or not.

If an AI system has achieved a certain degree of functionality in this way, it may be more favourable for the company in question to do without its further technical optimization and support artificial intelligence with human intelligence. Instead of developing a machine or model even further, which costs a lot of money – you have to hire experts, find data centers, pay for computing power and data – it is sometimes more useful, if not necessary, to have a human being this machine simulated for technical or economic reasons.

Their sceptical findings on the illusory nature of artificial intelligence stands in striking contrast to the predominantly positive statements with which the progress in the development of increasingly powerful AI systems such as ChatGPT or Dall-E was commented on. Do you see your assessment more confirmed or relativised with regard to the generative artificial intelligences of the last generation?

In general, my approach and conclusions have not changed. However, I would like to state that almost every new commercial product that comes onto the market with the stamp AI is hyped in the media. ChatGPT, Dall-E or Midjourney are no exceptions in this respect. These are products that work relatively well, but at the same time are very far from the wonders that they present their manufacturers. They are machines that often generate false or pointless results, as in the case of ChatGPT, or create mediocre illustrations, as in the case of Midjourney. However, under the spell of commercial discourse, we tend to be forgiven for mistakes or bugs and forgive dysfunction. It’s a bit like pampered children who are adored so much that you overlook their bad qualities. If you take a closer look, you notice that the highly acclaimed chatbot ChatGPT sometimes simply hallucinates.

What do you mean?

ChatGPT is a machine that improvises texts; it is not a search engine and also no encyclopedic AI. It does not serve the purpose of providing you with true answers to questions, but to improvise based on the information you give her. On the basis of “Today is a beautiful day” she can improvise several pages long. But as soon as I try to get substantial answers from ChatGPT, for example, in terms of actual events, such as the birthday of a particular personality or the course of historical events, their improvisational art sometimes tends to invent supposed facts. A test for the inability of ChatGPT, which quite a few people have been for fun, for example, is to ask the chatbot about one’s own biography. If you ask the chatbot four times about your biography, you will have four different life stories. These stories also contain accurate information, but also many others that are fictitious. That’s what I mean by hallucinating.

Does the regular excitement about the development of new AI systems influence your research?

No, not at all. Every time a new, supposedly brilliant AI comes onto the market, I just wait for what happens. Or, to say it with an old Chinese saying, I’m sitting on the river bank and waiting for the body of my enemy to pass by. It is usually only a matter of time until in practice the weaknesses of each new AI system are revealed. In the case of ChatGPT, it was not even necessary to wait very long for this moment. The chatbot was launched at the end of November 2022 and in January 2023 it became public that the machine is still dependent on the use of numerous people who constantly continue to train and moderate it.[3] In this case, these people are based in Kenya and were recruited via the SAMA platform, which is very well known in the industry. But these people are just one of many contingents of trainers, examiners and, who knows, perhaps also imitators, keep the ChatGPT running. After that, internal documents from the developer company OpenAI show that people in South Africa, the Philippines, India, Turkey and the United States are also being used for this type of work. OpenAI has micro-workers around the world: recruited via various platforms.

From where and to what conditions do these people work?

Microworkers: there are everywhere in the world, but the largest number of them are in low-wage countries. This is not surprising, after all, most companies are trying to minimize their labour costs. The companies in the tech industry are no exception. In addition, the countries concerned are often states where labour law is not as strictly regulated as in most European countries. It is therefore no coincidence that the companies that produce AI systems are based in countries from the global north, while the click work needed to keep the corresponding machines going is mostly done by people in the global South.

In this context, post-colonial or neocolonial relations of dependence are also becoming quite clear. In Southeast Asia, for example, there are numerous companies that mainly work on large companies of the world powers China and the USA and which are dependent on these orders. However, many people also work directly or indirectly for European and US companies in other regions of the global South, especially in Africa and Latin America.

Do there are similarities in the biographies of these workers despite the different geographical contexts: or recurring sociological features?

With our research team, we were primarily in Africa and Latin America, but we are still doing research in Europe. We are currently working on a systematic study in several European countries, including in Germany. We are seeing large differences in the profiles of microworkers: women. For microworkers: in France or in Germany, there is a high probability that they have an above-average level of education and a university degree and make this job a secondary job improve their income. Above all, single women with children often bring an important additional income to an important additional income, which can be done flexibly. However, even in cases where microwork is paid relatively well, we say between 70 cents and a few euros per task, which can amount to a few hundred euros at the end of the month, it usually does not form the main source of income.

The situation is changing radically with regard to Africa or South America. Unlike in Europe, the existence of those affected depends on microwork. In this context, two countries from our research can serve as prime examples, namely Venezuela and Madagascar. Venezuela is a global hub of micro-work for both Spanish-speaking and English-speaking countries, and for France it is Madagascar. In both countries we have conducted several hundred interviews and distributed thousands of questionnaires. For both countries, the cost of living is high compared to wages. In Venezuela, the average daily income is between five and eight dollars. Although they do tasks that are only remunerated at a few cents, people can double their microwork salary there – if they have a different income at all.

In Madagascar, the situation we observed was slightly better, partly because we were concentrating on companies that were established on the market and their microworkers: women up to ninety, sometimes even hundred euros per month. As a monthly wage, this is not worth mentioning by European standards, but in Madagascar it is only a little less than the average wage. But even there it is not enough for a life in the capital Antananarivo, where the cost of living is significantly more expensive, as in many capitals. We must not forget that.

Microworkers: So in countries such as Venezuela or Madagascar, do not earn badly compared to the national average. But that does not mean that microwork is well paid for. Above all, it means that the living conditions in many countries are so precarious that even comparatively small sums make a difference. Many AI companies are taking advantage of this situation.

What about the qualification of micro-workers: appointed in these countries? And what about gender distribution?

In the course of our research, we have found that the micro-workers: the majority of the majority in the countries concerned are people who tend to be overqualified. It is often people with a university degree or even with a master’s degree. In Madagascar, we are among the micro-workers: even a person who had obtained his doctorate. The local labour market is severely restricted, especially for highly qualified people, there are only a few jobs and these are rarely at attractive conditions. Therefore, overqualified individuals are often unemployed or underemployed and therefore willing to do this kind of work.

Moreover, in these countries, the distribution of micro-workers: after gender, is another. In countries where access to the labour market is difficult, people who are privileged by the country are beneficial. So you meet more men there. In some cases, 80 percent of men are 20 percent women among microworkers, while in France there are 56 percent women and 44 percent men.

In the countries of the South, the majority are people who are better qualified compared to the population average, but earn little compared to the skills they possess. There is a big problem behind this. It is a crisis symptom of liberal ideology and its promise of advancement through education. This promise is losing more and more credibility. In fact, many countries are now observing a global decorrelation between education and salary.

How is it that microwork is being carried out in both the global south and in Europe and the United States? Would it not be more favourable for companies to completely outsource this kind of work to low-wage countries for business reasons?

There are several criteria that play a role in deciding on outsourcing. Some of them are cultural. In the course of the development of artificial intelligence, many work steps have to be completed. In this context, for example, voice recordings must also be interpreted to verify the transcription, translation or other forms of processing of texts. For this purpose, it is necessary that the people who do this work have a sense of language and their local characteristics. Take only the German-speaking region, for example. It already makes a big difference, whether people speak High German or Swiss German, whether they come from Berlin, Vienna or Zurich. In the development of AI products, it often needs local experts: in the women. That is why companies are willing to pay more as soon as they seek people living in high-income countries, such as Switzerland. They want to be sure that they recruit workers whose skills meet the cultural needs of the companies.

In other cases, it is not cultural factors that are decisive for the decision to outsource, but legal factors. Aspects of data security also play an important role, because sensitive information is often processed in the development of an AI, which concerns the security of a company or a country. For example, just think of AI systems that are used in the context of health care or national defence. Companies are interested in ensuring that such sensitive data does not reach the public, and therefore they have an interest in employing workers who are bound by strict requirements or working in separate, strictly shielded areas.

These requirements make it clear why AI companies do not orient themselves exclusively towards financial aspects when selecting their workers. The protection of data is of at least as important importance. Regardless of whether the persons concerned were recruited in California, Europe or Madagascar, we were able to see in all countries where we have carried out our participating observations, we have noticed that certain safety-related forms of micro-related work take place only in particularly protected places that have an intercom, a camera, and sometimes even armored doors. Companies also accept higher costs.

What can you tell us about people’s working conditions? How stressful are the different forms of microwork for psyche and physique?

The risk of mental illness is not equally high in all kinds of microwork. The most affected group is that of the content moderator: women. These are people whose work is to train algorithms in such a way that they recognize, filter and sort out or block certain online content. It is usually about representations of pornography and violence. To do this, the content moderator must: look at thousands of images with corresponding content every day and classify them as rules, or legally compliant or not according to their specifications. These contents are often very explicit, they can show scenes of extreme violence and therefore seem traumatic. This work is indeed burdensome, but it is not representative of micro-work in its entirety.

One thing that has proved to be very variable in our field research, depending on the location, is the degree of social contact. We have met people who work in complete isolation from home, as well as people sitting in an office who are sitting in an office and cultivate the kind of conviviality that is common between colleagues. And we have had to deal with people who work from home, but who succeed in mobilising social resources.

One example of this is micro-workers: in Venezuela, some of whom include their entire family, not only children, but also the grandparents. These families worked according to a predetermined schedule. For example, first the father did the father do some tasks at certain times, so the daughter and finally the grandmother took over. In these cases, a network of social relationships that emerged around this work was actually observed; and which enabled companies to benefit not only from the work of a person who is already very poorly paid, but by three people.

On one point, however, all micro-workers agree with whom we have interviewed, in the global north as well as in the global south, namely in dissatisfaction with their working conditions and especially with their pay. They have an increasingly strong awareness that their work is a qualitatively challenging job, which also generates significant added value for companies. In most cases, they are aware of the fact that their work is contributing to the production of products that, if you like, belong to the spearhead of the tech industry. After all, AI is on everyone’s lips, and they are involved in the development of the AI systems everyone talks about and everyone wants to buy. In terms of their contribution to the production process, their remuneration is very small.

And in what respect do the experiences of microworkers differ?

An important point where not only, as they say, the spirits, but in this case also separate the sexes, concerns the assessment of one’s own career prospects. Women who carry out microwork are looking a lot more pessimistic about the future than men. This is also evident from other studies that we have conducted. Men can also more often imagine to continue to practice micro-work in the future. Or they regard this work as part of a transition phase that serves them to learn a profession that they will later pursue. Even if such optimism is rare overall, it is more common in men than in women.

Many women, on the other hand, see micro-work primarily as a makeshift, a temporary activity to bridge a difficult situation that requires an additional income. In addition, many women have a strong awareness of the fact that they are exposed to discrimination. These are mostly more indirect: micro-work platforms do not deliberately offer women work on worse conditions. Many of them, however, award contracts according to a point system that rewards particularly efficient microworkers. Women with children who only occasionally accept micro-work orders or only pursue this work for a few hours a week usually do not manage to increase their score on the platform in the same way as it would be necessary to receive better orders in the future, i.e. simpler tasks that are better rewarded and especially providers who then actually pay. On these platforms, there are unfortunately many shady companies that remain guilty to the workers: their wages and exploit their legal defenselessness. At present, there is no indication that this situation will change anything in the foreseeable future.

You just mentioned a point system. What do you have in mind?

There are different systems, but the whole thing works like this: A company is looking for workers to carry out certain tasks, for example to sort images according to specific criteria. The company publishes appropriate advertisements on a mediation platform. Let’s assume that 10,000 people respond to these advertisements, who then work through the tasks. Depending on their speed and the quality of their work, the people receive not only their wages, but also points. Sometimes these points are also distributed directly from the platform in question. In these cases, other aspects are often taken into account. For example, micro-workers can obtain additional points if they are considered particularly reliable, i.e. staying on the platform for a longer period of time, or if they have done more demanding and better paid tasks satisfactorily.

Sometimes points are also awarded directly by the customer. This is usually the case with more complex tasks. If the company has enjoyed the work of a particular person, there may be a good grade for them, as is usual with Uber or Deliveroo, for example. For the microworkers: the points system is an ambivalent affair. On the one hand, they have the opportunity to improve their earnings through good work and appropriate assessments, on the other hand, they are dependent on the benevolence of their customers: what makes criticism more difficult.

On which platforms do you encounter such valuation systems? And how are they related to the applicable wage models?

Evaluation systems are particularly widespread on the platforms on which the tasks are actually carried out. But there are also points systems on recruitment platforms. There, for example, you can accumulate points if you have taken on a particularly large number of small tasks.

As already mentioned, there are also a number of subcontractors who specifically provide workers for certain micro-tasks and offer their services on their own platforms. There are also various remuneration models: In some cases, different tasks are paid according to corresponding rates, in other cases the workers receive fixed hourly, daily or monthly wages. The contractual structures are sometimes dubious and often to the detriment of the workers. For example, the contractually guaranteed wages are often incongruent and extremely low. A not insignificant part of the wage consists of performance-related bonus payments for work well done, a high score or the number of tasks completed. To a certain extent, these bonuses have to be earned each time, you can’t be sure of them.

How and where does mediation between companies and workers take place?

In our first field research project in 2019, we investigated Amazon’s Mechanical Turk platform. During our work, we realized that it was by no means the only model for microwork placement. Amazon’s Mechanical Turk was quite simple at the time and functioned a bit like a bulletin board: companies that had certain jobs to offer could post them on the platform, and interested parties could then apply for these jobs independently. This model was based on a simple intermediary, namely the platform, which mediated between customers and providers, i.e. entrepreneurs and workers.

In the meantime, the situation has become more complex and we can observe a situation that can be described as deep labor – a network of different platforms, companies, subcontractors and freelancers.

Unlike just a few years ago, it is now often extremely complicated to understand and untangle the sometimes convoluted legal relationships. A person who worked for Microsoft on the development of a voice assistant and whom we interviewed during the aforementioned research project gave us an insight into the procedure. According to this person, Microsoft first contacted a Chinese platform to recruit microworkers. The Chinese platform then contacted a Japanese recruitment platform. The latter then commissioned its subcontractor in Spain, which in turn had concluded contracts with microworkers in France. As you can see, it’s a trip around the world. In this case, at least five platforms were involved in the placement process. Some platforms are simply there to ensure the placement, while others mainly take care of the recruitment. Still other platforms take care of the payment process. And then there are the platforms where the tasks are actually carried out.

Can certain developments nevertheless be recognized in this opaque web?

Today, we are seeing a multiplication of platforms, a multiplication of levels, which makes it increasingly difficult to understand the entire process and to assign both responsibilities and accountabilities. As researchers, we must first explore this field and talk to those involved, which usually means interviewing the workers directly. We then try to trace the chain of contractual relationships and employment relationships in order to understand today’s micro-labor market, which has exploded in recent times. Research in this area is much more difficult today than it was ten or fifteen years ago. Back then, roughly speaking, there were only two or three other platforms to look at besides Amazon’s Mechanical Turk. Once you had collected your data, you already had a solid basis for research. Today, there are far more platforms and people working or looking for work via them. This is also a consequence of the pandemic and the subsequent economic crisis.

However, there are also structural reasons for the increase, which are directly linked to the hype surrounding AI. After all, microwork is an activity with no end in sight because, contrary to the promises of the tech industry, AI systems cannot train and test themselves. For this reason, microwork will not become superfluous in the future, but will be increasingly in demand. The more the use of artificial intelligence progresses, the greater the need for human labor.

What factors play a role in the formalization of employment relationships?

Formalized employment relationships are usually tied to certain requirements. As already mentioned, this may involve certain language skills or other relevant skills on the part of the microworker or special requirements on the part of the company, such as increased confidentiality levels or stricter security requirements. Particularly high demands on the quality of the work required can also be a motivation. Companies rely on contractually regulated employment relationships in particular if they are interested in planning security and want to ensure that the required workers are not gone from one day to the next. As this is difficult on platforms such as Amazon’s Mechanical Turk, there are now several platforms that try to employ a stable contingent of workers.

I can give you two examples of such platforms that are among the largest in the world today, if not the two largest. The first is called Appen and is Australian-owned, the other, TELUS, is an international holding company based in Canada. Both have several million users and several million workers. In order to be accepted into the pool of workers, applicants have to go through various stages, just like in a traditional company. This also applies vice versa. Even companies that want to recruit workers for certain tasks cannot simply advertise these tasks, as is the case with Amazon’s Mechanical Turk, but must go through a selection process.

Microworkers are selected through a certification system, which also requires them to take exams. I have taken a few of these exams myself. Some are very simple. However, sometimes you come across platforms that ask applicants particularly challenging questions, but still pay them very little money once they are hired. With the help of certification systems, platform operators organize the scarcity of labor and job opportunities. In this way, they are able to retain employees without having to enter into a traditional wage employment relationship.

What is the situation with regard to the self-organization of click workers? What developments have there been in recent years?

There have been developments that, in my opinion, represent a gradual process. It is important to bear in mind that digital labor includes two other forms in addition to the micro labor discussed so far: on the one hand, the “uberized” work of delivery workers, i.e. the people who work in transport and logistics; on the other hand, the unpaid micro labor of users who – often without their knowledge – also contribute to improving the functionality of AI systems.

The struggles of delivery workers, delivery drivers and logistics workers have been particularly visible and successful in recent years, not least thanks to the accompanying work of legislators around the world. For click workers, on the other hand, it is much more difficult to organize to defend their rights for two reasons. Firstly, because of the large distances involved, which makes effective self-organization to defend their own interests and rights difficult, and secondly, because of the public invisibility of microworkers, most of whom work from home. Unlike delivery drivers, we do not meet them on the street and have no idea of their working conditions. And they themselves are often unable to publicize them, as their hands are tied by confidentiality clauses.

Nevertheless, things are currently changing here too. We are seeing a kind of continuation of the struggles that emerged in the mid-2010s. Class actions are being filed to have the employee status of these workers recognized. A very prominent case in France concerns the company Click and Walk, which employed over 200,000 microworkers, and according to its own figures, even over 700,000. A French court has ruled that all of these people must be registered as employees. This was an important decision, but it has since been overturned by the Criminal Chamber of the Paris Court of Cassation.

What is the situation like in other countries?

In some other countries, things are a little better. Take Brazil, for example. In September 2023, a labor court in the state of São Paulo ruled that Uber must employ its drivers subject to social security contributions, i.e. take them on as regular employees.[4] Even though the ruling is not yet legally binding and Uber has appealed against the ruling, this was certainly a historic decision.

These successes are encouraging, but do not yet represent a breakthrough. In general, it can be said that microworkers can only defend their rights and improve their status through legal action. Courts are currently the only place where they are actually seen and heard. Demonstrations as a means of mobilizing public support seem unpromising, as most people would not understand what the protesters’ profession is and what their problems are. If you see 400 delivery drivers with their boxes and bicycles, you understand very quickly who these people are and what they are about. When you see 700 or 7000 microworkers walking by, it’s much more difficult. So far, microworkers have lacked social recognition in every respect. It’s time for that to change.