#28 Society and Technology

Behind the “intelligent” and “magical” appearance of artificial intelligence (AI) products lies a vast amount of human labor. Thousands of “AI evaluators” who assess, correct errors, and filter harmful content from outputs of Google’s Gemini and Search’s “AI Overview” products form an invisible middle layer of workforce. These workers, employed through contract via GlobalLogic, a subsidiary of Japanese Hitachi, play a critical role in ensuring the system’s reliability and its “human-like” appearance.

The experience of writer Rachael Sawyer lays bare the intensity and difficulty of this work. Tasks that initially seemed simple have gradually become subject to increasing pressure to work faster. These evaluators, earning between 16 and 21 dollars per hour in the United States, report that the time allocated to read and score 500-word responses has been cut from 30 minutes to just 15 minutes. This pace makes it nearly impossible to verify content accuracy. Yet expectations remain unchanged despite deteriorating conditions.

Workers complain about constantly changing guidelines, deliberate omission of information, and being assigned tasks outside their expertise—such as evaluating medical or financial content. One of the most striking issues is the relaxation of standards around AI safety. One evaluator notes that certain content once deemed “never acceptable” is now labeled “acceptable.” Guidelines suggesting that harmful content such as hate speech is not considered a violation if generated by users rather than the model itself increase risks.

This intense and insecure work environment undermines workers’ trust in the products they help produce. Many evaluators now avoid using AI tools themselves and advise others to stay away from them as well.

While Google argues these evaluators provide only “external feedback,” experts say the situation is more complex. Adio Dinika from the Distributed AI Research Institute (DAIR) describes AI as a “pyramid” built on invisible, indispensable, and easily replaceable middle-class labor. Both engineers employed by major AI companies and white-collar workers within this ecosystem are essential components of the system.

Ultimately, as speed and profitability take precedence in AI development, ethical and safety concerns are pushed to the background. This leads to problems such as incorrect medical advice, toxic content, or nonsensical errors. Rather than marketing AI as a “magical” technology, attention must be paid to improving the working conditions of those behind it. Otherwise, the brilliance of this technology will continue to cast a growing shadow over the workers laboring in darkness.

Artificial intelligence has become a daily reality in educational settings. Yet this development has reignited the debate over whether AI can replace teachers. With the start of the new academic year, this question has reemerged for parents, educators, and policymakers alike.

According to a survey by the Pew Research Center last spring, about one-third of AI experts believe the teaching profession will be at risk within the next twenty years . A July report by the Learning Policy Institute states that as of 2025, one in eight teaching positions is either vacant or filled by individuals lacking full competency in their field. Thus, AI emerges as a concrete labor issue.

However, education policy experts, particularly since the pandemic, emphasize the irreplaceable value of the human connection teachers provide. Thomas Toch, director of FutureEd at Georgetown University, says students need to rebuild connections—with peers, families, and learning tools. The isolation caused by online learning during the pandemic led to serious psychological issues among students. Toch therefore characterizes an education system based solely on machines as “problematic.” Similarly, Jenkins argues that teachers offer authentic relationship-building and a sense of community that AI can never provide.

Today, AI is widely used by teachers as a practical tool to ease their workload. It assists with lesson planning, saves time in grading student assignments, and reduces administrative burdens. From this perspective, for educators who use technology appropriately, AI has become a transformative support tool for teaching. Yet clear boundaries must be established.

Philadelphia offers a striking example of this transformation. Just a few years ago, teachers and administrators debated banning ChatGPT. Today, they have launched a “AI 101” training program in collaboration with the University of Pennsylvania’s Graduate School of Education. Philadelphia’s Superintendent Tony Watlington describes this approach as “We are not hiding from AI—we are learning with it.” He argues that a responsible attitude toward AI requires both exploring its opportunities and seriously monitoring its potential downsides.

In conclusion, while AI in education is becoming an increasingly powerful tool, experts and teachers agree on one clear point: no technology can replace the teacher. The human connection in the classroom—the student’s feeling of being seen and heard, the emergence of community—is a fundamental value that AI cannot provide. AI can assist teachers, save time, and enrich lessons, but it cannot fill the place of the human element in education.

Albania has made history by appointing the world’s first AI government official. Named “Diella” (meaning “sunlight” in Albanian), this digital minister is not an “AI minister” managing AI-related agendas, but a direct “AI public procurement minister” responsible for overseeing government purchasing and tender processes.

According to Prime Minister Edi Rama, from now on, all public expenditures—from cement for road construction to office supplies—will be managed step by step through Diella, in a fully transparent and corruption-resistant manner. Rama described this initiative as “not science fiction—it is Diella’s job.”

Diella will not only manage tender processes but also help break bureaucratic bias and inertia by hiring talent from around the world. However, it remains unclear whether Diella’s public procurement version shares the same technological infrastructure as the existing Diella, which already provides services to citizens via the e-Albania state portal. Some sources suggest Diella is built on OpenAI and Microsoft technologies.

It is also unknown whether Diella, currently depicted in the portal wearing traditional Albanian clothing—a pink-and-white outfit with a headscarf—will be represented by a similar avatar in its new role. Moreover, despite being deployed to prevent corruption, Diella could still be manipulated by malicious actors or make erroneous decisions.

Despite all risks, this step is viewed as a historic experiment in integrating AI into government administration. Diella may serve both as a testing ground against deeply rooted corruption in the Balkans and as a model for other countries.