Epistemic Agency and the Politics of Social Work in the 21st Century

In his recent opus magnum, Luke Kemp argued that power comes in four different incarnations: (i) control of decision-making, (ii) control of resources that others depend on, (iii) control through threats and violence, and (iv) control of information (Kemp, 2025). Whereas the first three forms of power have remained relatively unchanged since the advent of nation states, the control of information that allows us to make sense of the world and “understand and manipulate” (Kemp, 2025, p. 1) fellow human beings is changing rapidly with the rise of Artificial Intelligence (AI). This article outlines the implications of AI-based Large Language Models (LLMs) for control over knowledge production and the concentration of epistemic power in a handful of corporations—actors that shape what we know, how we know it, and ultimately, our democratic processes.

Fusing cognition and convenience

AI—and particularly LLMs—are becoming pervasive at breathtaking speed. As Kurzweil (2024) predicted, our cognitive functions are effectively fusing with AI systems, forming a hybrid intelligence more capable of analysis than comprehension. Tools such as Copilot illustrate how AI is becoming embedded in our working lives; convenience itself drives adoption. Yet the trade-off is agency. Users, often pressed for time, are drawn into reliance on systems they neither control nor fully understand. Studies show increasing use of AI-driven search by multilingual Australians, even amid discomfort over data control and transparency (.auDA, 2025). The resulting ecosystem entrenches the dominance of Microsoft and Google, centralising epistemic authority and weakening the visibility of original sources (Thomson et al., 2025).

The shifting epistemic architecture

Scholars such as Coeckelbergh (2023) argue that AI is reshaping our epistemic architecture—defaulting knowledge production to statistical patterning, prone to echo chambers, and vulnerable to manipulation. LLMs are not true epistemic agents; responsibility lies with their designers and corporate overseers. The Grok incident (Fehres et al., 2025) demonstrated how model weighting can encode ideological bias, exposing the fragility of our informational commons. As right-wing populism weaponises culture and information, control of knowledge becomes the new political frontier. Corporate AI systems now operate as epistemic filters, mediating what the public perceives as reality.

Hybrid epistemic authority

Traditional truth claims—anchored in peer review and deliberation—are being replaced by iterative, opaque, post-hoc optimisation. In public LLMs, this process is confined to technical teams serving commercial rather than civic goals. Leiter (2024) warns that long-standing norms about whom to believe are collapsing under the weight of algorithmic mediation. LLMs thus instantiate an epistemic oligarchy: a small cadre of executives and engineers controlling the gateways of knowledge. Convenience becomes complicity; our “epistemic laziness” (Coeckelbergh, 2023) renders us vulnerable to harm, dependency, and ideological capture (Seger et al., 2025).

Epistemic security and social work

The hollowing out of democratic institutions parallels the centralisation of information supply chains. Advocacy groups have begun pressing for corporate epistemic responsibility (Lamy, 2022) and public epistemic authority (Neame, 2025). Social work, rooted in critical inquiry and human rights, must respond by cultivating critical digital literacy and AI awareness. As facial-recognition systems, automated transcription, and chatbot counselling enter mainstream practice, social workers must interrogate how these tools reshape professional knowledge and ethics. Resistance lies in education: embedding critical thinking, media literacy, and data justice within training and practice. The challenge of the twenty-first century is not merely to use AI ethically but to reclaim epistemic agency itself.

References

1. .auDA. (2025). Digital Lives of Australians 2025. auda.org.au
2. Associated Press. (2025). Google unveils ‘AI Mode’ in the next phase of its journey to change search. theguardian.com
3. Coeckelbergh, M. (2023). Democracy, epistemic agency, and AI. AI and Ethics, 3, 1341–1350. doi.org
4. Fehres, A., Conroy, L., & AI4Media. (2025, Jul 14). How do you stop an AI model turning Nazi? theconversation.com
5. Kemp, L. (2025). Goliath’s Course: The History and Future of Societal Collapse. Viking Penguin Random House.
6. Kurzweil, R. (2024). The Singularity is Nearer. Vintage / Penguin Random House.
7. Lamy, E. (2022). Epistemic responsibility in business. Journal of Business Ethics, 183, 1–14. doi.org
8. Leiter, B. (2024). Free speech on the internet: The crisis of epistemic authority. Daedalus, 153(3), 91–104. jstor.org
9. Neame, L. (2025). Facts, Arte-Facts, and Fabrications. Political Theory, 1–38. doi.org
10. Seger, E., Perry, H., & Hancock, J. (2025). Epistemic Security 2029. demos.co.uk
11. Thomson, T., Nagappa, A., & Weinbrand, S. (2025). AI overviews have transformed Google search. theconversation.com