Beyond the tool view of AI: Intelligent technologies and the emergence of new epistemic regimes
Source: https://www.semanticscholar.org/paper/cb0b272fa808402ee53a9a043890d26944f593f6 ↗
Sergeeva, Leonardi, and Faraj identify a clean analytical failure in how organisations talk about AI: adoption, automation, and augmentation frameworks all presuppose that new technology is absorbed into existing authority structures, whereas what AI actually does is introduce a rival basis for knowing that can displace professional expertise at the root.
The 'epistemic regimes' frame — organisations as arenas where competing logics of credible knowledge contend — is a genuine conceptual contribution rather than a restatement of principal-agent or information-processing models.
This is exactly the vocabulary product directors need when they encounter resistance to algorithmic outputs from clinicians, lawyers, or analysts: the conflict is not about adoption friction, it is about whose grounds for justified belief get to count.
The paper connects directly to the library's strand on bounded rationality and organisational decision-making (Simon, March) while extending it into terrain those authors could not anticipate.
Low citation count is the only material risk, but Leonardi and Faraj are established organisation-theory voices whose prior work on technology and knowing has influenced the field, and the framing is original enough to clear the bar.
Central argument
Sergeeva, Leonardi, and Faraj argue that prevailing frameworks for understanding AI in organisations — adoption, automation, augmentation — share a common flaw: they treat AI as a tool absorbed into existing authority structures rather than as a rival epistemic logic. Their core thesis is that AI introduces what they call a new 'epistemic regime' — a competing basis for what counts as credible, justified knowledge — and that this regime can displace professional expertise not by outperforming it on agreed criteria but by substituting different criteria altogether. Organisational conflict around AI is therefore not primarily about change management or skill gaps but about which grounds for knowing get to carry authority.
Critique
The epistemic regimes framework is analytically sharp but risks overstating the coherence and novelty of AI as a knowledge logic: many algorithmic systems encode the same professional priors they ostensibly displace, meaning the regime 'shift' may be a repackaging of existing power rather than a genuinely new epistemology. A sharper challenge is whether the framework offers actionable traction — diagnosing a conflict as epistemic rather than political or economic does not obviously tell leaders how to resolve it, and the paper may underspecify the mechanisms by which one regime actually supplants another in practice.
Why it matters for product
When a product director embeds an AI-driven recommendation layer — in clinical triage, legal review, or financial analysis — and experts resist its outputs, the instinct is to treat this as an adoption problem solvable with training or UX improvements; this paper reframes it as a legitimacy contest over whose reasoning process the organisation treats as authoritative, which changes both the stakeholder strategy and the design brief. Concretely, it implies that metrics framing (does the model outperform the expert, or does it reason differently from the expert?) and governance design (who adjudicates conflicts between human and algorithmic judgement?) are not implementation details but the central product decisions.