Reproduce “Epistemological Fault Lines Between Human and Artificial Intelligence” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Walter Quattrociocchi, Valerio Capraro, Matjaž Perc
This research formalizes the fundamental epistemological disparities between human cognition and large language models (LLMs) by introducing the concept of "Epistemia," where linguistic plausibility substitutes for genuine epistemic evaluation. The paper identifies seven specific "epistemic fault lines" that delineate these structural differences, aiming to provide a critical framework for AI evaluation and governance.
Key insights
- LLMs are statistical pattern-completion systems that operate solely on linguistic regularities, fundamentally lacking grounding in real-world experience, internal motivation, or causal models.
- The "Epistemia" framework highlights how LLM fluency exploits human cognitive heuristics (like fluency-as-credibility) to create an illusion of veracity, making epistemic evaluation optional for the user.
- Increased scale in LLMs (more data, parameters) refines linguistic automation but does not bridge the gap from statistical correlation to genuine cognition, belief, or metacognition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.