Abstract

Current large language models (LLMs) embody a positivist epistemology that treats knowledge as an external regularity latent in data. Yet in the human sciences meaning is enacted, embodied, and socially negotiated. This research essay (i) diagnoses epistemic blind spots in data‑driven AI, (ii) re‑introduces phenomenology and constructivism as alternative foundations, and (iii) offers an Epistemic Matrix that maps AI techniques to their assumptions and vulnerabilities. Positivist LLMs excel at syntactic patterning yet generate hallucinations, context drift, and accountability gaps whenever second‑order notions such as roles or intentions are relevant. Phenomenology highlights lived experience and structural coupling; constructivism foregrounds community sedimentation of shared meaning. Synthesising these lenses yields six research propositions for mixed human–AI sense‑making. The paper concludes with practical diagnostics managers can apply to decide when dialogical, community‑centric AI design is mandatory.

Download

Share

COinS