From Data to Wisdom: Rethinking AI Through an Epistemic Lens

Afonso Eduardo

TL;DR

This paper introduces the Data-to-Wisdom Epistemic Architecture for AI (EAAI), a framework that transforms the classic "Knowledge Pyramid" from a linear progression into a set of five recursively interacting functions. By distinguishing between model prediction (Knowledge), reliability appraisal (Understanding), and justified action (Wisdom), EAAI provides a rigorous vocabulary for evaluating AI systems, particularly through the lens of Bayesian methodology.

What Problem Does This Paper Address?

In both AI research and information science, the "Knowledge Pyramid"—Data, Information, Knowledge, (sometimes Understanding), and Wisdom—is often treated as a one-way street where data is magically distilled into wisdom. The author argues this obscures the reality of how inquiry works: a scientist might start with a decision (Wisdom) that dictates what data to collect, or a model failure (Understanding) might force a reinterpretation of existing facts (Information).

Furthermore, the paper identifies a terminological gap in Bayesian literature. While "prior distributions" are mathematically precise, terms like "prior information" and "prior knowledge" are often used interchangeably. This lack of clarity makes it difficult to communicate why a model was built a certain way or how its outputs should inform high-stakes decisions in fields like finance or healthcare.

Methods

The paper is primarily a conceptual and methodological contribution. The author develops the EAAI framework by:

  1. Redefining the DIKUW Categories: Treating them as distinct epistemic functions (e.g., Data as observation, Knowledge as modeling/prediction).
  2. Targeted Textual Review: Analyzing ten foundational Bayesian reference books to identify ambiguities in how "prior" context is categorized.
  3. Bayesian Mapping: Formally mapping Bayesian statistical tools (likelihood, priors, posterior predictive checks, and decision theory) onto the EAAI functions.
  4. Worked Applications: Applying the framework to three complex scenarios: quantitative financial risk, actuarial health insurance modeling, and agentic AI for medical reimbursement.
The EAAI framework reinterprets the DIKUW hierarchy as a set of interacting functions rather than a fixed sequence.

Datasets

Not discussed in the paper. As a conceptual framework paper, it does not introduce new training datasets or benchmarks. Instead, it uses domain-specific "worked examples" to demonstrate the framework's utility in structuring real-world AI problems.

Key Insights and Ideas

  • Understanding is the "Missing Link": Many modern DIKW versions omit "Understanding." EAAI restores it as the function of critical appraisal—asking why a model succeeds or fails, rather than just looking at its accuracy.
  • Recursive, Not Linear: Inquiry can start at any level. A failed backtest (Understanding) can prompt new data collection (Data) or model revision (Knowledge).
  • Five Kinds of Prior Context: The paper extends the concept of "prior" beyond just a probability distribution. It distinguishes between prior data, prior information, prior knowledge, prior understanding, and prior wisdom, each playing a different role in shaping a model.
  • Agentic AI and Error Propagation: In multi-step "agentic" systems (like an AI reviewing medical claims), the output of one model (Knowledge) becomes the "Data" for the next. EAAI highlights the need to treat these intermediate outputs as evidence open to challenge, rather than established facts.
The recursive nature of EAAI shows how functions like Understanding and Wisdom guide and constrain Data and Knowledge.

Key Results

The paper’s primary "result" is the successful mapping of Bayesian methodology to the EAAI functions, providing a "shared vocabulary" that makes complex statistical tools more accessible to non-statisticians.

  • Bayesian Correspondence: The author shows that Likelihood corresponds to Information (contextualizing data), Priors/Posteriors to Knowledge (the model's beliefs), Model Criticism to Understanding (evaluating reliability), and Decision Analysis to Wisdom (justifying action).
  • Functional Evaluation: The framework argues that a model's predictive performance (Knowledge) does not establish its reliability (Understanding) or its appropriateness for a specific decision (Wisdom). These require separate evidence and oversight.
Mapping Bayesian tools to EAAI functions provides a disciplined account of belief revision and decision theory.

Limitations and Future Work

  • Conceptual Scope: The author explicitly states that EAAI is an analytical framework and does not claim that AI systems currently possess human-like "Understanding" or "Wisdom."
  • Bounded Review: The textual review of Bayesian literature was targeted and not an exhaustive survey of the entire field.
  • Human Oversight: The framework places a heavy emphasis on human review at the "Understanding" and "Wisdom" stages, which may be a bottleneck for fully autonomous systems.
  • Future Directions: The author suggests that EAAI could be extended with more functions and that its relationship to prior work (like cognitive architectures) warrants a more systematic, large-scale review.

Summary updated

Abstract

This paper proposes the Data-to-Wisdom Epistemic Architecture for AI (EAAI), a conceptual framework treating Data, Information, Knowledge, Understanding and Wisdom as five distinct but recursively interacting epistemic functions: observation, contextual interpretation, modeling and prediction, critical evaluation and informed judgment. Two observations motivate it: sequential readings of the Knowledge Pyramid obscure how these functions interact, and a targeted review of ten Bayesian reference books finds that prior distributions are defined mathematically while "prior information" and "prior knowledge" are not consistently distinguished as epistemic categories. A Bayesian interpretation relates the likelihood, prior and posterior distributions, posterior prediction, model checking and decision analysis to the five functions, without limiting EAAI to Bayesian systems. Extending "prior" in an epistemic sense, the paper distinguishes five kinds of prior context that can shape the model, the likelihood, the prior, evaluation and action. Three worked applications -- quantitative financial risk, actuarial modeling in health insurance and agentic AI for medical reimbursement claims -- show how EAAI structures the analysis of model assumptions, evidence interpretation, validation, dependencies and decision consequences, separating the prediction of a risk, the appraisal of its reliability and the justification of action. It draws implications for AI and machine learning, including evaluation by function, treating intermediate model outputs as evidence open to challenge and separating generation from appraisal, and compares EAAI with work on DIKW, Bayesian workflow, AI evaluation and risk governance. Its contributions are an epistemic vocabulary with recursive relationships, a more accessible reading of Bayesian methodology, a typology of prior context and clearer methodological terminology.
BibTeX
@misc{eduardo2026from,
  author = {Afonso Eduardo},
  title = {{From Data to Wisdom: Rethinking AI Through an Epistemic Lens}},
  year = {2026},
  eprint = {2611.16726},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.16726},
}
Figure