"US AI is forcing its view of the world on us."
Artificial intelligence promises progress. But who actually decides which problems it should solve? Indian researcher Urvashi Aneja warns of an AI boom in which the so-called Global South serves as a testing ground, while power and value creation remain in the Global North. AufRuhr met her at re:publica in Berlin and spoke with her about digital colonialism, resource consumption, language, and alternatives to Large Language Models.Ms. Aneja, how is AI changing India?
India and other countries in the Global South see enormous potential in AI to address challenges such as access to healthcare and education with relatively little time, money, and effort. India aims to position itself as a leader in the Global South and is currently building a public digital infrastructure. This infrastructure will allow companies to offer relevant services. Expectations are high: If AI works in a country with many languages and cultures, then it could also bring about social progress elsewhere.
However, they themselves are concerned about these developments. Why?
Technological optimism and enthusiasm are understandable in a country like India, which faces enormous problems and has limited resources. My skepticism stems from the fact that budgets and political priorities are already being shifted in favor of AI without any solid evidence of the actual benefits of these AI solutions. This development is being driven primarily by technology companies from the Global North. State actors in the Global South, on the other hand, often lack the capacity to thoroughly evaluate the products being offered. It seems as if the AI companies have sold us a hammer, and now we're desperately searching for the right nail.
Critics describe the countries of the Global South as a testing ground for AI. Do you share this assessment?
Historically, we see such patterns time and again. One example is Aadhaar, the biometric identification system in India. It is now required for all basic public services, from buying a mobile phone to even taking a taxi. If Germany or France wanted to introduce such a system, there would be massive public resistance. But in countries with weaker regulations or in countries experiencing humanitarian crises, companies can often test new technologies without giving the population a choice.
It seems as if the AI companies have sold us a hammer, and now we're desperately searching for the right nail.
- Urvashi Aneja, AI researcher, political consultant and author
Is colonial thinking continuing here in digital form?
It is clearly a continuation of colonial patterns. That's why many rightly speak of a new "digital colonialism": the Global South serves as a source of resources , while the actual value creation takes place almost exclusively in industrialized nations. In response, India is now trying to develop its own AI models to position itself further up the value chain and maintain its national sovereignty. The problem, however, is that this national sovereignty often fails to consider the interests of its citizens. Many Indian AI companies ignore issues such as labor law, data protection, the fight against discrimination, and environmental damage.
Technology is not neutral. How are US AI models shaping the tech landscape in India?
These large language models (LLMs) are predominantly based on English-language training data. And language is identity: US culture, worldview, and prejudices are thus inscribed into these AI applications. Local developers can adapt the models for specific purposes and for one of India's many languages. But their fundamental values and assumptions cannot be corrected. US AI systems thus impose their worldview on us.
We have stopped asking whether AI is even the right political project and whether it truly serves humanity.
- Urvashi Aneja, AI researcher, political consultant and author
What alternatives do you see for India – and what can Europe learn from them?
We need AI models that offer clear societal benefits. Before deploying them, we must define the problem they are intended to solve. This also means moving away from large-scale machine learning (LLM). LLMs consume enormous amounts of resources and are not precise enough for many applications. Smaller, specialized systems, on the other hand, are more transparent, resource-efficient, and easier to verify. Furthermore, they can be trained more effectively on specific cultural or professional contexts.
What was your core message at re:publica ?
The debate surrounding AI has narrowed considerably. While a few years ago we were discussing the democratization of the AI ecosystem, today it's almost exclusively about the technology's proliferation. We've stopped asking whether AI is even the right political project and whether it truly serves humanity. That's why I keep urging you: Don't stop asking critical questions. Those who don't ask today for whom technology is being developed may have no choice tomorrow.
The interview was originally published online at Aufruhr, the Mercator Foundation Magazine.