AI as Deep Rationality Prosthetic: Evidence Processing, Delegated Credence, and Core Rationality in AI-Mediated Inquiry
An AI-mediated inquiry paper arguing that better evidence processing does not automatically produce better assigned credence.
Source Paper
This page is a simplified guide to the paper. The source document is linked below; the guide here keeps the main idea accessible and tied back to the site’s model.
Open on Academia.eduLow-Register Abstract
This paper treats AI as a possible prosthetic for Deep Rationality. AI can widen search, summarize complex material, generate rival hypotheses, identify base rates, compare likelihoods, and find contradictions. Used well, it can help a user perceive the evidence more clearly.
But AI cannot by itself secure Core Rationality. The final assignment of confidence still passes through human or institutional uptake. A user can use AI to see better, or use AI to make a preferred conclusion look better than the evidence warrants.
How It Fits Credencing
This paper ties the AI Alignment application directly into the two-gap model. AI can reduce E0-EP gaps while leaving EP-CA gaps untouched, hidden, or amplified.
It also clarifies the site’s AI warnings: the danger is not only hallucination. The danger includes fluent false confidence, false precision, motivated prompting, automation bias, source opacity, delegated credence, and institutional laundering.
How to Read It
Read this when using AI for research, teaching, policy, diagnosis, writing, or institutional review. Ask what the AI changed in your perceived evidence, what uncertainty remains, and who is assigning the final credence.
Adjacent References
These are starting points rather than exhaustive citations. They connect this page to nearby work in probability, calibration, cognitive bias, and social epistemology.
- |Thomas Bayes, "An Essay towards solving a Problem in the Doctrine of Chances" (1763).
- |E. T. Jaynes, Probability Theory: The Logic of Science (2003).