HOMEPAGE

A Dutch government algorithm financially ruined thousands of families who were wrongly accused of fraud in the childcare benefits case. The system, trained on biased data and operating without transparency or human supervision for over six years, caused a scandal that led to the government's resignation and raised concerns across Europe about the risks of blind trust in algorithms.

J. Buolamwini's research on the datasets used for facial recognition highlighted the bias in systems trained on them. By testing their effectiveness in recognizing faces, she revealed the systems' inability to recognize dark-skinned women's faces. For example, Microsoft's system had an error rate of 21 percent, while IBM's equivalent was about 35 percent. However, when the same systems were tasked with recognizing light-skinned men, the error rate dropped below 1 percent.

A widely used healthcare algorithm in the United States was designed to identify the most ill patients. However, doctors and patients discovered that it operated with significant racial bias. Instead of calculating based on illness, the algorithm calculated based on the cost of care. For a variety of reasons, it was found to offer significantly fewer services to African American patients due to racial patterns in its data. Thus, the algorithm overlooked patients in need of help, reproducing and reinforcing pre-existing social inequalities.

Despite the massive investments in artificial intelligence, the vast majority of generative AI systems fail to deliver measurable results. The problem does not lie in the algorithms themselves, but in the lack of operational maturity: weak data governance, insufficient model monitoring, absence of ethical oversight, and limited integration into real workflows. 

The full potential of AI depends not only on the technology itself, but on what is known as operational excellence. Key factors include data quality, appropriate infrastructure, and workforce expertise. In practice, it seems that the rapid adoption of AI does not necessarily lead to its beneficial application.

The work of two Harvard scientists argues that the grand narratives surrounding AI obscure the real picture. They compare the development of AI to that of electricity and nuclear energy, emphasizing the dimension of time: progress will come slowly as innovation is absorbed by society. Rather than creating utopias or dystopias, we must establish the new entry conditions for these technologies into production and society.

Google employs thousands of people to train and improve its Gemini AI system. These workers, often paid very low wages and without benefits, perform tedious tasks such as content moderation, data labeling, and checking AI responses, forming the "hidden backbone" of the industry. This practice reveals the human labor behind the digital "magic" of AI.

The Amazon Go retail chain promised a unique experience for consumers. No cashiers, no scanning, no codes — customers “Just Walk Out” of the store. An AI system using sensors, cameras and object recognition algorithms tracks customers’ movements and automatically charges them for their purchases.

But behind the company’s promises, the reality was quite different: 1,000 workers in India were monitoring the video footage from the stores, doing the work that cashiers once did.

American tech giants exported technologies such as facial recognition, predictive policing systems and cloud infrastructures, which supported, among other things, the “Golden Shield,” a nationwide surveillance system in China. Officially, it was promoted as a technological advancement enhancing public safety.

In practice, however, it was used for mass surveillance, control, and repression targeting members of ethnic minorities, dissenters, and even ordinary citizens, often violating privacy and other human rights. Despite global criticism and sanctions, the companies continued to provide or support these technologies for years, fueling an opaque and undemocratic mechanism of control.