Research in the Wild: Epistemic Authority in AI
Still writing, slowly but surely!
This is the fifth Research in the Wild post, part of a series capturing my learnings and evolving thinking as I work towards a contribution to the succinctly titled Topical Collection on AI Resistance, Refusal, Reclamation and Reimagining: Ethical Imperatives and Emerging Practices. You can read the previous instalment here.
The subaltern classes are the vast majority of people – those of us subjected to a system of thought and power that is produced by dominant classes but reproduced by us; at work, at school, in church, at the supermarket… The web of sociocultural, political and historical circumstances that rest upon us like a smog is what Gramsci called hegemony.
Bringing Gramsci’s notions of subalternity and hegemony into the world of AI makes me somewhat uneasy. The rise of fascism in Italy after World War I is the backdrop of Gramsci’s writings. That seems far greater a social issue than the rise of a handful of tech bro billionaires. The unrest within me is only subdued by AI chatbot companies having a body-count of 24 according to Wikipedia, as of 7th April, 2026; AI being used in the US’s killing of over 100 Iranian schoolchildren, according to people with knowledge of the matter; and Israel’s genocide of Palestinians relying on “AI warfare,” which is enabled by the Big Tech companies – such as Google, Meta, OpenAI and Anthropic – opening the doors to military contracts in recent years.
I struggle to tackle the gravity of the situation with what my training in the analytic tradition of philosophy gave me. To me, Gramsci’s thinking helps make sense of the rebellion brewing among subaltern classes.
My concern remains in the effectiveness of those otherwise siloed rebellions – divided and thus easier to conquer.
Recall the thesis I am working on articulating and defending:
Challengers of predominant AI narratives and practices are united in (i) their subjugation to forms of epistemic injustice, and (ii) their organisation through epistemic activism.
In the previous post, I gave us a name – AI subalterns. This is hopefully one step towards forming some sense of unity. Two major steps remain:
Defining those who commit epistemic injustice; and
Articulating the shared experience of epistemic activism among subaltern AI movements.
We already introduced epistemic injustice in the second post of this series. Roughly, it is the undermining of a person or group’s knowledge on some prejudicial basis. In this framework, individuals in positions of power –say, police officers, judges, and doctors– commit epistemic injustice by not depositing trust in their interlocutors’ –say, prisoners, victims of crime, and patients– testimonies, and by not developing the concepts necessary for interlocutors to describe their experiences of unfairness. The subjugated interlocutors are the AI subalterns in our framework. The individuals in positions of power are epistemic authorities.
Just consider how many sound bytes about AI that hit the headlines are from the mouths of Big Tech executives. “CEO said a thing” has now become a journalistic genre of its own. It is news when one says AI will become sentient within five years, another claims we won’t need real estate agents within two, and another explains how their AI chatbot is already too powerful to be released into the public realm. In this flux of misinformation of AI, we also find ourselves have to understand new, emergent and ambiguous concepts, going from NLP to LLMs to generative AI, from AI governance to AI safety, from frontier AI to agentic AI, and so on and so forth.
These are narratives that cause confusion; distract from what’s really going on; allow Big Tech to continue launching flashy chatbots to more and more random softwares and websites whilst pretending they had nothing to do with this death or that water shortage.
This, I will argue in my paper, is a form of hermeneutical injustice; a flooding of the infosphere with so much jargon –sometimes technical, sometimes entirely inadequate– and so many ideas –often fictional– that the vast majority of people simply have no resort but to take what the tech bro billionaires say for granted.
Next post…
I have about 1,800 words written so far with the intro and definition of AI subalterns. Writing these posts and seeing comments and engaged people has helped a lot – thank you! The next post on here will follow some writing – I will either get stuck with epistemic authorities and write on them some more, or move onto a case study. Anybody up for some Luddism?