We already know that artificial intelligence amplifies inequality. For several years now, algorithms have been used, among other things, in credit assessments, social benefits allocation, and “at-risk” profiles identification. These automated decisions disproportionately affect more vulnerable and marginalized populations. Deliberately opaque, algorithmic processes of ranking, triage, and sanctioning dilute responsibility until it effectively disappears. If you’re economically, educationally, or socially marginalized, chances are you already live in Kafka’s The Castle.
Other applications of immense computing power affect much broader segments of the population. Consider monitoring of warehouse/field workers, manipulation of voters, and electronic surveillance. After the September 11, 2001, attacks, experts wondered how to deal with the staggering volume of data they were already able to collect. Today, that question no longer arises. Artificial intelligence, efficient and versatile, takes care of it.
In late 2022, generative AI (GenAI) burst onto the scene and made significant waves. It can create things for us, including writing. The machine does not understand the meaning of its responses, yet it appears to think. The simulation is remarkably convincing. In reality, these powerful microprocessors do not think at all. They answer your questions (prompts) by calculating probabilities from an unimaginably huge quantity of text samples (vector database) broken down into small segments (tokens). Constraining formulas (algorithms) govern the composition of the responses. Even if it is recycled material, the result is mesmerizing.
The question, then, is no longer simply who is marginalized by algorithms, but who has the autonomy needed to assess texts produced by GenAI. A new digital divide is emerging. It is not enough to have access to GenAI or to know how to use it. Mastering prompt engineering will not save you. You need to be able to correct machine-generated outputs.
Can you detect simplifications, errors, and biases? The usefulness of GenAI depends heavily on your prior knowledge. Without benchmarks, you risk being dazzled by it and swallowing its results whole. Speed and flattery play a key role: the answer is immediate and confirms your assumptions. A sense of pure wonder that acts as a gilded cage: GenAI was designed to charm you and keep you engaged.
AI and inequality
With its user-friendly interface, GenAI deepens inequality. People who are already skilled get more out of it than newcomers and ordinary users. For the educated elite, it is a promise of learning, improvement, and productivity. For the vast majority, it is an uncontrollable magic trick performed at their own risk and peril. This divide is not hypothetical, it is already visible in national survey data. Pew Research Center’s “Americans and AI 2026” report (published June 2026) found that about half of U.S. adults now use AI chatbots regularly, even as trust in the companies building them continues to lag well behind adoption. Forget Neo in The Matrix; think instead of the sorcerer’s apprentice in Fantasia.
What matters is the effort required to validate the results. Can you draw on prior knowledge, identify inconsistencies, and verify sources? Do you have the time to do so? These are privileges enjoyed by a small segment of society, people who have received a good education, know how to read, and cultivate doubt. As a consequence, GenAI could create a mass of functionally illiterate people accustomed to the illusion of immediate truths, formulated with reassuring kindness. A study from MIT’s Media Lab (Kosmyna et al., 2026, sometimes referred to as “Your Brain on ChatGPT”) reveals that participants who wrote essays with a GenAI showed measurably weaker brain connectivity than those who wrote unaided, and most could not accurately quote a sentence from the essay they had just finished.
Still, these instant, attention-grabbing, seemingly coherent answers are GenAI’ssignature feature and the secret to their success Who will have the ability to maintain critical distance from an extremely seductive technology? That is the question.
Everyone’s education and willingness to learn certainly matter, but the environment in which people operate can be decisive.
Privilege of having a validation network
In professional, scientific, and academic networks, validation mechanisms are multiple and mutually reinforcing: colleagues, communities of practice, peer-review committees. GenAI is one tool among many. We used it ourselves to prepare this column.
The general public, including students at all levels of education, may come to regard AI as a reliable source of information and delegate the task of writing to it. Inequality then concerns access to validation networks and, also, the time available to users. Learning is a slow, repetitive, demanding exercise. Reading and writing likewise require concentration and persistence. In an age that cultivates anxiety, reading and writing are a chore, habits from another era.
The inability to judge the quality of knowledge produced by a machine facilitates and amplifies technological domination. The resulting inequalities are persistent. They affect individuals and institutions alike. For governments, businesses, and organizations, GenAI is already a marker of success and distinction. Or of being out of touch. Are you, or are you not, at the technological forefront?
To stay current, organizations showcase innovation labs, massive investments in GenAI, and subscriptions to the most advanced tools. We can even observe forms of conspicuous consumption of GenAI: companies proudly displaying their “AI-first strategy” (in English, please), government ministries touting their use of tailor-made AI, universities presenting the integration of GenAI into teaching as a competitive advantage. It serves as both a status symbol and a hallmark of excellence, a signal that attracts customers and taunts the competition.
This relentless race for novelty continually creates new gaps. As soon as an AI application becomes widespread, more sophisticated uses appear on the market. Here too, we encounter the mirage of unattainable modernity, of a status that slips away just as we believe we have attained it. Inequalities continually reinvent themselves in ways that make their reproduction more effective.
AI & international development
One question remains. How will public and private international aid donors respond to the wave of GenAI? Since the rise of Reaganomics in the 1980s, we have seen international donors demand adaptation without regard for how local economies functioned. Projects were funded only to the extent that they mimicked, even on a modest scale, structural adjustment programs (the marketization of goods and services, promotion of the private sector, labor-market deregulation, openness to foreign investment, and the transfer of Western management models). The absurdity reached remarkable heights: there were even claims that farmers should abandon drug crops, which remained profitable, and instead grow roses and pineapples, whose returns were uncertain. The assumption, unspoken and unspeakable, was that, from the depths of their valleys, Indigenous communities would follow the global flower and fruit market and grow whatever happened to be profitable the following year.
Will we dare to do the same with AI? The magic formulas are at the ready: AI for development/equity/inclusion, responsible AI, AI readiness, AI adoption, local AI ecosystems, AI scaling, AI skills, and so on. International institutions are preparing as well. The UNDP has established the AI for Sustainable Development program and its Artificial Intelligence Landscape Assessment (AILA) tool. The World Bank is not standing on the sidelines. It plans to help poorer countries build the infrastructure needed to deploy AI. The title of its latest annual report reveals a certain fascination: The Promise of Artificial Intelligence. Implicitly, AI is presented as “the” key to moving toward a modern economy and efficient public services.
As Indermit Gill, Senior Vice President and Chief Economist of the World Bank Group, argues in the World Development Report 2026, developing economies should approach AI with“an optimistic mindset”. Because many remain heavily reliant on agriculture and small enterprises, AI may be more likely to support workers than replace them. “Wielded adroitly”, asIndermit Gill almost casually notes, AI could also expand access to essential medical, legal, educational, and agricultural services for billions of underserved people, achieving “in a decade what might otherwise take a century”. Yet one must ask: is this promise grounded in possibility, or does it drift into reverie? Does the World Bank really suggest that the poor in politically unstable countries can afford the luxury of assessing AI’s potential while maintaining an optimistic mindset?
But for now, AI is not a general “conditionality” of official development assistance (ODA). Recipient countries are being encouraged to prepare for it (by developing local capacity and infrastructure and adopting laws and regulations governing the use of AI). What will the next step be? Will the “AI divide,” based on unequal capacities to adopt, understand, and correct AI systems, become an object of research and consultation? Will the use of AI by certain groups become a requirement for ODA?
With regard specifically to GenAI, the questions could be particularly troubling. Will donors measure the actual effects on illiterate and semi-literate populations? Will they simply celebrate how GenAI reduces inequality by providing personalized access to education while concealing the loss of users’ intellectual autonomy? What about disinformation, polarization, and voter manipulation?
At its core, there is perhaps a touch of cynicism in promoting GenAI in countries that still struggle to provide affordable Internet access and reliable electricity. After all, AI is particularly demanding of both. If data centers already consume around 20% of Ireland’s electricity, what might we reasonably expect in poorer countries with unreliable electricity systems?
Let us end with a major issue: the concentration of wealth. In January 2019, in a sparsely attended room at the World Economic Forum in Davos, Rutger Bregman confronted billionaires about taxes and tax avoidance. He urged them to stop talking about philanthropy and start talking instead about taxation. “Taxes, taxes, taxes!” Seven years later, the question remains. Can the promotion of AI as a tool for combating inequality ignore the debate over the taxation of the ultra-rich who own and run the companies behind it?
By Remy Smida, Founder of Research for Purpose &
Guillermo R. Aureano, PhD, Internship Coordinator and Lecturer in the Department of Political Science at the Université de Montréal