Philosophy
What if India’s Right to Information movement offered the real template for AI governance? The argument is bold: democracy in the age of AI means not just transparency, but the right to challenge hidden systems, question their categories, and protect knowledge that should never be reduced to data.
Cognitive Justice as AI Governance Principle
There is a domestic Indian precedent for epistemic democratisation that has not yet been sufficiently connected to the AI governance debate. The Right to Information movement, built over decades by activists like Aruna Roy and the Mazdoor Kisan Shakti Sangathan in Rajasthan, established a principle that now seems obvious but was genuinely radical when it was fought for: that citizens have the right to know what the state knows about them and does in their name. The RTI Act of 2005 was not merely a transparency measure. It was a redistribution of epistemic power, a recognition that information asymmetry is a form of domination, and that democracy requires not just the right to vote but the right to know.
The AI governance challenge is structurally analogous, and the RTI movement’s political logic maps onto it with uncomfortable precision. AI systems know things about citizens, about their creditworthiness, their health risks, their likelihood of reoffending and their political sympathies that citizens themselves do not know. They make consequential decisions based on this knowledge. And the communities most affected by these decisions are the poorest, the most marginalised and the least digitally literate; those who have the least ability to contest them.
What would AI governance that takes cognitive justice seriously actually require? At a minimum, four things. First, genuine epistemic diversity in development teams: not as a diversity-and-inclusion metric, but as an epistemological necessity. Systems built by people who share a single framework will encode that framework as the default. Second, community consent protocols modelled on the gram sabha principle: not individual opt-in checkboxes, but collective deliberative processes through which communities decide how data generated by and about them may be used. Third, plurality in evaluation frameworks: the recognition that there is no single metric for what makes an AI system good, and that communities must be able to specify their own criteria for benefit and harm. And fourth: most fundamentally, the recognition that some knowledge cannot and should not be datafied.
This last point is the essay’s most philosophically original claim, and it deserves to be stated carefully. The argument is not that traditional ecological knowledge is too fragile for the digital world, or that indigenous communities need to be protected from technology. The argument is that some knowledge is constitutively relational, embodied, place-specific, and seasonal and that its relational quality is not a limitation to be overcome but the very thing that makes it what it is. The Bishnoi community’s fifteen-generation relationship with the khejri tree cannot be entered into a dataset. Not because the dataset is too small, but because the relationship is constituted by being-there, over time, in reciprocity. It is not information about the khejri tree. It is a way of being with the khejri tree enacted across generations, embedded in ritual, song, and seasonal practice. An AI system trained on data about the khejri tree, its botanical properties, its ecological role, and its drought resistance does not have access to any of this. And the ecological management decisions made based on that AI system will be, at the precise point where it matters most, blind. This is what cognitive justice, applied to AI governance, ultimately demands: not just that more voices be included in the existing process, but that the process itself be reconceived around a more honest account of what can and cannot be known computationally and around the political recognition that what cannot be known computationally is often what is most important to the communities whose knowledge it is.