Artificial intelligence (AI) is transforming the world of work on several fronts simultaneously. It alters the conditions of those who work to make it possible, often in invisible and precarious jobs; it changes hiring, organizational, and career advancement processes, where automated decisions can reproduce historical inequalities under the guise of neutrality; it heralds a new labor paradigm in which part of the economic value may grow without a proportional increase in human labor; and it transforms the way we educate professionals who will have to coexist with it, altering both teaching methods and the competencies required of graduates. This text proposes exploring these areas to reflect on how to guide the development and use of AI toward institutional, educational, and social collaboration that places the dignity of people and their working conditions at the center.

The Human Infrastructure Behind Artificial Intelligence

The apparent magic of AI algorithms rests on a deep human infrastructure that remains out of the picture. Although it is often imagined as an autonomous, purely computational system, it actually relies on millions of people who produce, correct, and validate the data that makes its operation possible. Those who label images, classify texts, or moderate content construct that “ground truth,” without which state-of-the-art models could not continue to grow.

This invisible—or rather, rendered invisible—dimension reveals how the global AI value chain is organized: fragmented, delocalized, and, in many cases, deliberately opaque. According to World Bank estimates,^[1] between 154 and 435 million people participate in some form of online contract work, as a primary, secondary, or marginal activity. Part of this ecosystem involves data production, sorting, and annotation tasks required for AI systems. These tasks are often outsourced to countries in the Global South, such as Kenya, the Philippines, India, or Brazil. There, workers perform them under precarious conditions, with a heavy psychological burden and little contractual stability. Within the broader context of online gig work, this situation is even more acute for women, who particularly value the ability to balance remote work with unpaid caregiving responsibilities.

Various studies have documented these working conditions. In Kenya, some workers tasked with classifying toxic content to train AI systems earned between $1.32 and $2 per hour and were constantly exposed to descriptions of violence, sexual abuse, and other traumatic content.^[2] In the Philippines, thousands of such workers, known as “taskers,” work through data annotation platforms, many of them from internet cafes, their homes, or precarious workspaces. One worker interviewed reported working approximately 12-hour days, seven days a week, while others complained about the lack of contracts, social protection, and effective mechanisms to claim unpaid wages or wage reductions.^[3] Precariousness, therefore, goes far beyond compensation: it includes unpaid waiting time and training, instability, a lack of recognition for their work, and exposure to psychologically harmful content.^[4]

In this context, the recent encyclical Magnifica Humanitaswarns against the consolidation of a new form of digital colonialism, which is no longer exercised solely over territories or bodies, but also over the vital data of vulnerable populations that feed algorithmic power systems.^{[5 ]}From this perspective, Pope Leo XIV explicitly speaks of “new forms of slavery” to describe labor practices that, under the guise of technological innovation, reproduce hidden relationships of exploitation.

The argument is clear: no technology that presents itself as emancipatory can be ethically sustained if it depends on precarious, deliberately invisible labor. In light of this, social justice in the digital age demands much more than “universal” access to technology. It entails guaranteeing decent working conditions, transparency in data use, and mechanisms for public oversight that prevent economic profit from becoming the sole criterion for decision-making.

The Most Advanced Technology Reinforcing the Oldest Stereotypes

The problem with AI, however, is not limited to the working conditions of those who train it. It also arises in contexts where algorithms make decisions about other people. One of the most sensitive areas is hiring. Increasingly, companies use AI systems to screen resumes, rank applicants, analyze recorded interviews, or predict job candidates' future performance. Although these tools are often presented as objective and efficient, various studies have shown that they can reproduce and amplify biases present in the historical training data.^[6]

The problem arises because algorithms learn from past decisions. If an organization has historically hired more men than women for certain positions, or if certain social groups have had less access to educational and employment opportunities, the system may interpret those patterns as indicators of success and replicate them in its future recommendations. Instead of correcting preexisting inequalities, automation risks reinforcing them under the guise of technical neutrality. A particularly troubling phenomenon is also at play: a series of experiments involving a simulated medical diagnosis task showed that people can internalize and subsequently reproduce the biases in AI recommendations.^[7] Thus, social prejudices feed into the algorithms, and the algorithms’ responses, in turn, influence human decisions, creating a feedback loop that can reinforce and normalize those very same patterns of exclusion.

Algorithmic discrimination is particularly concerning because it often operates invisibly, using criteria hidden within complex, opaque models. As a result, people excluded from a selection process are rarely able to learn the reasons for that exclusion, effectively challenge it, or even be aware that it has occurred. A recent study provides evidence of this, showing that, in certain scenarios, some generative AI models recommend significantly lower salary offers for women than for men with identical professional profiles.^[8]

From this perspective, AI can replicate social vulnerabilities. People who already face structural barriers due to gender, ethnicity, age, disability, or socioeconomic status may see their situation worsen when automated systems translate historical inequalities into predictions about the future—and these predictions, in turn, become self-fulfilling prophecies.

There is a high risk that technology will institutionalize and escalate existing forms of exclusion, thereby granting them apparent legitimacy. That is why we need mechanisms for auditing, transparency, and accountability that enable the detection of discriminatory biases and ensure that decisions respect human dignity and the principle of equal opportunity.

Productivity, Replacement, and Labor Transition

There is a third impact we observe in the world of work, stemming from the paradigm shift across many productive sectors. The International Labor Organization (ILO) warns of the particular vulnerability of certain administrative jobs and of female-dominated occupations.^[9] The International Monetary Fund (IMF) estimates that AI could affect nearly 40% of global employment and notes that low-income countries have less infrastructure and capacity to benefit from it.^[10] The United Nations Conference on Trade and Development (UNCTAD) warns that this unequal preparedness could widen the gaps between countries.^[11]

We do not know to what extent AI will replace human cognition in each sector.^[12] But it does seem clear that it can increase economic value not only through more human labor, longer hours, or more jobs, but also through an artificial capacity for prediction and cognitive processing that is increasingly cheaper, scalable, and faster. This points to an economy in which part of productivity growth will depend less on expanding the labor force and more on deploying systems capable of performing cognitive work at low cost.^[13]

However, we can anticipate certain tensions associated with a productivity increase of this magnitude: Will it translate into an improvement in workers’ quality of life, or simply into increased profits through reduced labor costs? In a system geared toward reducing costs and expanding margins, AI makes it possible to increase production and the value generated without increasing the workforce by the same proportion. This suggests that certain cognitive jobs, especially those that are more routine or have lower added value, may be reduced, transformed, or eliminated within a relatively short period.

The main risk may be the pace of the transition. Employers expect that 39% of current skills will change or become obsolete between 2025 and 2030, while 63% already identify a skills gap as a major barrier to business transformation.^[14] If the adoption of AI proceeds faster than the retraining of workers, the reorganization of productive sectors, regulatory adaptation, and the strengthening of social safety nets, the adjustment period could lead to job displacement, income loss, and greater inequality.^[15]

This potential increase in wealth concentration should prompt governments to review their distributional and social protection mechanisms. During the transition, it will be necessary to mobilize part of the value generated by AI to fund reskilling, support those who lose income, and facilitate the adaptation of productive sectors.

Educating About AI and Educating with AI: Digital Humanistic Leadership

AI is expanding its capabilities across an increasing number of professional fields. Therefore, it is essential to prepare future professionals for the effective, efficient, and ethical use of this technology. This requires vocational training focused on the practical, valuable use of AI, with the teaching profession playing a central role. In this context, teacher training presents one of the greatest challenges—both in teaching how to use these tools and in integrating them ethically into the practice of teaching.

But it is also true that AI can amplify errors, biases, and poor decisions, with very dangerous consequences when we use it to automate processes and delegate decisions about organizations and individuals. What is new, compared to the industrial revolution of automation, is that AI does not act solely on physical systems or industrial processes, but rather on cognitive, organizational, managerial, and decision-making tasks that directly impact people. In this new revolution, the issue of education takes on a different urgency: it must foster the ability to discern what should be accelerated, which processes require constant supervision, and which decisions should not be automated without ethical evaluation. ^[16]/[17]

How can the education system respond? The University of Deusto, for example, has launched several initiatives at both the institutional and operational levels. It has committed to an ethical stance on the use of AI, envisioning its integration into teaching, research, technology transfer, and management, guided by criteria of honesty, transparency, responsibility, sustainability, equity, and service to the individual.^[18] On a practical level, it has established an AI Commission, tasked with providing institutional guidance and support for this process, and a Digital Innovation Unit that provides resources, guidelines, and best practices for integrating technology into university activities.^[19] In the area of teaching, the university is rethinking student competencies and training, promoting educational innovation projects that incorporate AI into the teaching-learning process, and reviewing assessment and research processes, the use of different types of AI, and the manner in which the use of these tools should be disclosed. This effort also finds a natural home in the “Human Development and Values” courses, which are common to all undergraduate programs at the University of Deusto, where questions about technology are linked to justice, professional responsibility, and care for people and the planet.

Educating about AI and educating with AI are two distinct tasks that must advance together in pursuit of a humanistic digital leadership that is not limited to merely allowing or prohibiting technologies, but rather to providing discerning guidance on when they are valuable and when they are harmful, and when to use them and when not to.

Conclusions

The impact on the most vulnerable societies can be profound, and the challenge overwhelming. But at this point, the encyclical Magnifica Humanitas offers a call to hope: it warns against the temptation to believe that the problems are so great that our decisions make no difference, and, in the face of such resignation, it reminds us that the civilization of love is also built on small, steadfast acts of loyalty. From this conviction, a question arises: beyond defending and educating, how can AI be put at the service of those already working with vulnerable populations?

Some AI systems can analyze information, detect patterns, and help make sense of complex realities, such as situations of vulnerability or risk of exclusion. The social sector, which serves, supports, and raises awareness of these realities, must transform fragmented and scattered information into actionable knowledge—often with limited resources. As a general-purpose technology, AI could enhance diagnostics and strengthen public advocacy by providing a more solid foundation. But general-purpose AI that works with poor-quality, biased data—or data too far removed from the real-life experiences of the most vulnerable people—will perpetuate exclusion rather than correct it. At least three conditions would be necessary: critically reviewed data, human oversight of sensitive decisions, and a clear focus on care, support, and advocacy. The question is: Is general-purpose AI useful for this purpose? Are there specialized AI systems that are more suitable? Or is it essential to build a specific AI system to serve this purpose?

It is clear that no single actor can tackle this paradigm shift alone. The social sector is familiar with realities that often do not translate into useful data for decision-making or advocacy. The academic and technological communities, for their part, can contribute technical knowledge and tools to organize, interpret, and bring coherence to part of that complexity. A valuable opportunity lies precisely there: in the collaboration between those who understand technology and those who have firsthand knowledge of the social realities of injustice that technology should help resolve.

AI is already making its way into the social sector: the question is whether it will succeed in doing so to advance a better understanding of vulnerability and more just social action.

References:

^[1] Datta, N., Chen, R., Singh, S., Stinshoff, C., Iacob, N., Nigatu, N. S., Nxumalo, M., & Klimaviciute, L. (2023).Working without borders: The promise and peril of online gig work. World Bank. https://doi.org/10.1596/40066
^[2] Perrigo, B. (January 18, 2023). Exclusive: OpenAI used Kenyan workers earning less than $2 per hour to make ChatGPT less toxic.Time. https://time.com/6247678/openai-chatgpt-kenya-workers/
^[3] Simon, T. (April 24, 2025). The Filipino workers at the sharp end of artificial intelligence.Equal Times. https://www.equaltimes.org/the-filipino-workers-at-the-sharp
^[4] Salim Wagner, C., Miceli, M., Kinyua, J., Dinika, A., & Hanna, A. (2026).Data work in the ILO Platform Labor Convention [Policy report]. DAIR Institute. https://data-workers.org/policy-iloconvention/
^[5] León XIV. (May 15, 2026).Magnifica Humanitas: Encyclical Letter on the Care of the Human Person in the Age of Artificial Intelligence. Holy See. https://www.vatican.va/content/leo-xiv/es/encyclicals/documents/20260515-magnifica-humanitas.html.
^[6] Collett, C., Neff, G., & Gouvea, L. (2022).The Effects of AI on Women’s Working Lives. UNESCO, OECD, & Inter-American Development Bank. https://doi.org/10.18235/0004055
^[7] Vicente, L., & Matute, H. (2023). Humans inherit artificial intelligence biases.Scientific Reports, 13, Article 15737. https://doi.org/10.1038/s41598-023-42384-8
^[8] Sorokovikova, A., Chizhov, P., Eremenko, I., & Yamshchikov, I. P. (2025).Surface fairness, deep bias: A comparative study of bias in language models [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2506.10491
^[9] Gmyrek, P., Berg, J., Kamiński, K., Konopczyński, F., Ładna, A., Nafradi, B., Rosłaniec, K., & Troszyński, M. (2025).Generative AI and jobs: A refined global index of occupational exposure(ILO Working Paper No. 140). International Labour Organization. https://doi.org/10.54394/HETP0387
^[10] Cazzaniga, M., Jaumotte, F., Li, L., Melina, G., Panton, A. J., Pizzinelli, C., Rockall, E. J., & Tavares, M. M. (2024).Gen-AI: Artificial intelligence and the future of work (Staff Discussion Note No. 2024/001). International Monetary Fund. https://doi.org/10.5089/9798400262548.006
^[11] United Nations Conference on Trade and Development. (2025). Technology and Innovation Report 2025: Inclusive Artificial Intelligence for Development. United Nations. https://unctad.org/publication/technology-and-innovation-report-2025
^[12] Acemoglu, D. (2025). The Simple Macroeconomics of AI.Economic Policy, 40(121), 13–58. https://doi.org/10.1093/epolic/eiae042
^[13] Agrawal, A., Gans, J., & Goldfarb, A. (2018).Prediction Machines: The Simple Economics of Artificial Intelligence. Harvard Business Review Press.
^[14] World Economic Forum. (2025).The Future of Jobs Report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
^[15] Cazzaniga, M., Jaumotte, F., Li, L., Melina, G., Panton, A. J., Pizzinelli, C., Rockall, E. J., & Tavares, M. M. (2024). Gen-AI: Artificial Intelligence and the Future of Work(Staff Discussion Note No. 2024/001). International Monetary Fund. https://doi.org/10.5089/9798400262548.006
^[16] Bond, M., Khosravi, H., De Laat, M., Bergdahl, N., Negrea, V., Oxley, E., Pham, P., Chong, S. W., & Siemens, G. (2024).A meta systematic review of artificial intelligence in higher education: a call for increased ethics, collaboration, and rigour. International Journal of Educational Technology in Higher Education, 21, Article 4. https://doi.org/10.1186/s41239-023-00436-z.
^[17] Mirsch, M., Moreno, S. G., Schultz, B., & Leicht-Scholten, C. (2026). Responsible engineering in the age of AI: The value of responsible AI education from engineering students’ perspectives. SEFI Journal of Engineering Education Advancement, 3(1), 6–51. https://doi.org/10.62492/sefijeea.v3i1.48
^[18] University of Deusto. (March 2024). University of Deusto’s Position on the Use of Artificial Intelligence(Version 1.1). https://www.deusto.es/document/deusto/es/posicionamiento-deusto-uso-ia.pdf
^[19] University of Deusto. (n.d.). Artificial Intelligence at Deusto. Retrieved July 15, 2026, from https://www.deusto.es/es/inicio/somos-deusto/estrategia-academica/ia-deusto

Author

Cover image credits: Cheiro annotates data in his wooden home in a slum in Cagayan de Oro. Photo: Théophile Simon. Published on 24 April 2025 in Equal Times.