🤖 AI

Better Understanding, Understanding Better

This paper introduces a comparative epistemic logic framework that models degrees of understanding and inter-agent comparisons by enriching multi-agent epistemic models with graded explanation structures and justification-style term algebras, while establishing soundness, strong completeness, and decidability for its finite-level fragments.

Original authors: Yu Wei (Department of Philosophy, East China Normal University)

Published 2026-07-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yu Wei (Department of Philosophy, East China Normal University)

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a world where "knowing" and "understanding" are treated as two completely different things. In our daily lives, we often say, "I know the answer," but we also say, "I understand the answer." The paper argues that these aren't just synonyms; they are different levels of mental achievement.

Here is the core idea of the paper, broken down into simple concepts using everyday analogies.

1. The Core Problem: Knowing vs. Understanding

The author, Yu Wei, starts with a famous quote attributed to Einstein: "Any fool can know; the point is to understand."

Knowing is like having a map. You know the route from your house to the store. You can get there.
Understanding is like being a tour guide who can explain why the road goes that way, point out the history of the buildings, and explain what would happen if a bridge collapsed.

The paper notes that while logicians (people who study the rules of thinking) have built complex systems to model "knowing," they have largely ignored "understanding." But in real life, understanding isn't just a light switch (on/off); it's a dimmer switch. You can understand something a little bit, a lot, or perfectly.

2. The Solution: A "Graded" Logic

The paper proposes a new mathematical language (a logic) to measure these different levels of understanding. Think of it like a video game with different difficulty settings or a ladder with many rungs.

The Ladder of Understanding:
- Rung 1 (Minimal): You have a basic story. (e.g., "The Earth orbits the Sun because of gravity.")
- Rung 2 (Everyday): You have a slightly deeper explanation.
- Rung 10 (Scientist): You have a complex, mathematically rigorous explanation.
- Rung Infinity (Ideal): You have the perfect, ultimate explanation that covers every possible angle.

The paper creates a system where we can say, "Agent A is on Rung 3," and "Agent B is on Rung 7."

3. The "Better Than" Comparison

The most unique part of this paper is how it handles comparison. In the real world, we often say, "I understand this better than you do."

The paper introduces a special symbol (like a "greater than" sign) to represent this: $i \succ j$ .

This doesn't just mean "I know more." It means "My explanation for this specific thing is deeper and more robust than yours."
The Catch: You can only compare understanding if you are both looking at the same question. You can't say "I understand gravity better than you understand basketball" using this specific logic; it only compares how well two people understand the same topic.

4. How It Works: The "Explanation Backpack"

To make this work mathematically, the author imagines that every person carries a backpack of explanations.

The Items: Inside the backpack are "explanation terms" (like tools or notes).
The Grades: Each tool has a "grade" or quality score. A cheap plastic hammer is Grade 1; a high-tech laser cutter is Grade 10.
The Rules:
- To understand something at a high level (say, Rung 10), you need a Grade 10 tool in your backpack that works in every possible scenario you can imagine.
- Combining Tools: If you have a Grade 5 tool and a Grade 3 tool, you can combine them, but the result is usually a "messier" tool (a lower grade) because you've lost the specific details of which tool was doing the work.
- Reflecting: If you have a tool that explains why you know something, you can create a "reflection" of that tool (a meta-tool) that explains your reasoning. This bumps your understanding up a level.

5. The Two Versions of the System

The author builds two versions of this logic to handle different needs:

The "Bounded" Version (The Finite Ladder):
- This version assumes there is a maximum level of understanding (e.g., we only care about levels 1 through 10).
- Why? It makes the system easy to check. A computer can easily verify if a statement is true or false in this version. It's like checking a finite list of rules.
The "Full" Version (The Infinite Ladder):
- This version allows for "Ideal Understanding" (Level Infinity). It acknowledges that we can always imagine a deeper explanation.
- The Challenge: Because the ladder goes on forever, you can't check every single step with a simple computer program. The paper proves that while this system is logically sound, it requires a more complex, infinite method to verify truths. It's like trying to count all the stars in the universe; you can describe the rules, but you can't finish the count.

Summary

In short, this paper builds a mathematical ruler for understanding.

It acknowledges that understanding comes in degrees (you can understand more or less).
It allows us to compare two people's understanding of the same topic.
It uses a system of graded tools (explanations) to define what it means to "get it."

The author successfully proves that this new system is consistent (it doesn't contradict itself) and that we can use it to solve problems, provided we stick to the "finite" levels of understanding. It's a formal way of saying: "Yes, you can know something without understanding it, and yes, I can understand it better than you, and here is the math to prove it."

Technical Summary: "Better Understanding, Understanding Better"

Problem Statement
Contemporary epistemology and philosophy of science distinguish understanding from knowledge, particularly emphasizing that understanding admits of degrees (one can understand something "better" or "more deeply" than another) and is inherently comparative. While epistemic logic has developed rich tools for knowledge and belief, it has paid relatively little attention to modeling these specific features of understanding. Existing formal approaches, such as the logic of knowing why [35] and its extension to understanding why via higher-order explanation [31], capture the necessity of explanatory support but lack the machinery to represent a hierarchy of explanatory depth or to explicitly compare the understanding levels of different agents regarding the same proposition. Furthermore, standard justification logics often omit the sum operator ( $+$ ) or lack graded structures necessary to model the transition from minimal to ideal understanding.

Methodology
The paper proposes a Comparative Epistemic Logic of Understanding (CELU) that integrates three core components:

Graded Modalities: A family of operators $U^\tau_i \phi$ where $\tau \in \mathbb{N}^+ \cup \{\omega\}$ represents the level of understanding. Finite indices ( $1, 2, \dots$ ) correspond to increasing explanatory standards (from minimal to typical scientific understanding), while $\omega$ represents ideal understanding.
Comparative Connective: A primitive binary operator $(i \succ j)\phi$ meaning "agent $i$ understands why $\phi$ better than agent $j$ ."
Agent-Indexed Graded Semantics: The semantics enriches multi-agent epistemic models with:
- A set of explanation terms $E$ closed under operations ( $\cdot, +, !$ ) and containing a constant $c$ .
- A grade map $gr: E \to \mathbb{N}$ assigning a "strength" to each term. Crucially, the sum operation ( $+$ ) is strictly grade-downgrading (if inputs are positive, the sum has a lower grade), reflecting the loss of specificity when combining explanations.
- Agent-indexed admissible explanation functions $E_i(t, \phi)$ , indicating at which worlds agent $i$ possesses term $t$ as an explanation for $\phi$ .

The truth condition for $U^n_i \phi$ requires the existence of a single explanation term $t$ with grade $\ge n$ that is available to agent $i$ in all accessible worlds. The comparative operator $(i \succ j)\phi$ is defined semantically as: (1) the degree of $i$ 's understanding of $\phi$ (the supremum of levels $n$ where $U^n_i \phi$ holds) is strictly greater than that of $j$ , and (2) $j$ knows $\phi$ .

Key Contributions
The paper makes four primary technical contributions:

Unified Framework: It provides a single logical framework capable of representing minimal, ordinary, demanding, and ideal understanding, alongside comparative judgments between agents.
Bounded Finitary Calculus ( $SCU_\kappa$ ): For any fixed finite level $\kappa$ , the authors define a finitary Hilbert-style system. They establish soundness and strong completeness for this system. Furthermore, they prove that the satisfiability and validity problems for fixed finite-level fragments are decidable, provided the set of self-evident tautologies $\Lambda$ is effectively controllable.
Infinitary Full-Language System ($SCU$): To capture ideal understanding ( $\omega$ ), the authors introduce an infinitary extension of the calculus. This system includes an $\omega$ -introduction rule allowing the derivation of $U^\omega_i \phi$ from the infinite set of premises $\{U^n_i \phi \mid n \in \mathbb{N}^+\}$ . They prove soundness and strong completeness for this system using a canonical model construction based on maximal consistent sets closed under the infinitary rule.
Boundary Analysis: The paper isolates the logical boundary between the bounded and full languages. It demonstrates that in the full language, the comparative connective $(i \succ j)\phi$ is not eliminable by comparative-free formulas, whereas it is eliminable in the bounded language. Additionally, the full language is shown to be non-compact (a set of formulas can be finitely satisfiable but unsatisfiable as a whole), implying that no sound finitary proof system can be strongly complete for the full language.

Results

Completeness: The paper establishes strong completeness for both the bounded finitary system $SCU_\kappa$ and the infinitary system $SCU$.
Decidability: Satisfiability and validity are decidable for any fixed finite-level fragment $CELU_\kappa$ .
Non-Compactness: The full language $CELU$ is non-compact, as demonstrated by a set of formulas requiring an agent to have understanding at every finite level while another agent has ideal understanding, which creates a contradiction in the limit.
Properties of Comparison: The system derives standard properties for the comparative operator, including irreflexivity, transitivity, and asymmetry, derived from the interaction between the comparative connective and the level-indexed understanding operators.

Significance
The paper claims to recover "understanding" as a formally disciplined notion within contemporary epistemic logic, bridging the gap between philosophical distinctions (knowledge vs. understanding, degrees of understanding) and formal logical tools. By moving beyond the "knowing why" operator to a graded and comparative setting, the framework offers a unified meta-theory for analyzing explanatory depth. The authors position this work as a synthesis of epistemic, comparative, and justification-style ideas, providing a foundation for future work on dynamic changes in understanding (e.g., via communication or learning) and further comparisons with other formal accounts of explanation. The work is presented as a technical recovery of a concept historically treated in medieval logic but largely absent from modern formal epistemology until recently.