Ignorance with(out) Grasping

This paper proposes a topic-sensitive hyperintensional semantics to model ignorance as a content-dependent attitude, offering sound and complete logical systems for three distinct forms of ignorance while resolving the problem of logical omniscience by accounting for an agent's capacity to grasp proposition content.

The Gist

Imagine you are walking through a vast library of ideas. In this library, there are two books that tell the exact same story, word for word, just with different titles. In the world of standard logic, if you know the story in Book A, you automatically know the story in Book B. They are identical in meaning.

But in the real world, humans aren't like that. You might know the story about "The War of 1812" perfectly well, but if someone hands you a book titled "The War of 1812: A Nuclear Perspective," you might be completely lost. Even though the core facts are the same, the topic has shifted, and your brain hasn't grasped the new angle.

This paper, "Ignorance with(out) Grasping," argues that our current way of modeling human ignorance is too simple. It treats ignorance like a light switch: either you know something, or you don't. The authors, Ekaterina Kubyshkina and Mattia Petrolo, want to upgrade this to a dimmer switch with a topic dial. They propose that ignorance isn't just about what you don't know, but what you can even understand (or "grasp").

Here is the breakdown of their ideas using everyday analogies:

1. The Problem: The "Perfect Robot" vs. Real Humans

In traditional logic (called Kripke semantics), agents are like perfect robots. If a robot knows that "2+2=4," and it knows that "2+2=4" is logically the same as "The square root of 16 is 4," the robot instantly knows both. It has Logical Omniscience.

But humans are different.

• The Analogy: Imagine a person who knows how to drive a car. If you tell them, "Driving a car is the same as operating a vehicle with four wheels," they know it. But if you tell them, "Driving a car is the same as performing a complex dance routine involving 500 specific steps," they might be confused. Even though the "truth" is the same, they haven't grasped the concept of the "dance routine."

The authors argue that we need a new kind of logic that accounts for this "grasping." If you don't grasp the topic, you are ignorant, even if the answer is technically right in front of you.

2. The Solution: The "Topic-Sensitive" Lens

To fix this, the authors use a Topic-Sensitive Semantics. Think of this as giving every idea a "subject tag."

• The Tag: Every sentence has a topic (e.g., "War," "Math," "Cooking").
• The Grasp: An agent has a "mental shelf" of topics they understand.
• The Rule: You can only be "not ignorant" about a sentence if you have the topic tag on your shelf AND you have the information. If the topic is missing from your shelf, you are automatically ignorant, no matter how simple the fact is.

3. Three Types of Ignorance

The paper breaks ignorance down into three flavors, like different types of "not knowing":

• Ignorance Whether (The "Maybe" Box):

  • Scenario: You don't know if it's raining outside.
  • The Twist: You might know the concept of "weather" (you grasp the topic), but you just don't have the data. However, if the question was about "Quantum Rain," and you don't grasp quantum physics, you are ignorant because you don't understand the topic, not just because you lack data.
  • The Paper's Fix: The logic checks: "Do you understand the topic? If no, you are ignorant. If yes, do you know the answer?"

• Ignorance as Unknown Truth (The "Hidden Fact" Box):

  • Scenario: A fact is true (e.g., "There is a hidden treasure in the garden"), but you don't know it.
  • The Twist: If the treasure is described using a language you don't speak, you aren't just "unaware"; you are incapable of entertaining the thought.
  • The Paper's Fix: The logic ensures that for you to be "ignorant of a truth," you must first be able to understand the subject of that truth.

• Disbelieving Ignorance (The "Wrong Guess" Box):

  • Scenario: You believe your brother is in Rome, but he is actually in Paris. You are "disbelievingly ignorant" of the truth.
  • The Twist: To hold a wrong belief, you must first understand what the topic is. If you don't know who your brother is, you can't be wrong about his location; you are just completely clueless.
  • The Paper's Fix: This type of ignorance requires that you grasp the topic. You can't be "wrong" about something you can't even conceptualize.

4. The Big Win: Solving the "Omniscience" Bug

The biggest victory of this paper is solving the Logical Omniscience Problem.

In old logic, if an agent knows $A$, and $A$ is the same as $B$, the agent must know $B$. This makes agents seem like gods who know everything instantly.

• The Paper's Fix: By adding the "grasping" rule, the authors show that an agent can know $A$ but be ignorant of $B$, even if $A$ and $B$ are logically identical.
• Why? Because the agent might not have "grasped" the specific topic of $B$.

The William III Example:
The paper uses a historical example: King William III of England.

• He knew England could avoid war with France (Topic: War).
• He did not know about "Nuclear War" (Topic: Nuclear Physics).
• Logically, "Avoiding War" is the same as "Avoiding Nuclear War" (if you assume nuclear war is just a type of war).
• Old Logic: If he knew the first, he must know the second.
• New Logic: He knows the first because he grasps "War." He is ignorant of the second because he does not grasp "Nuclear Physics." The logic perfectly models his human limitation.

Summary

This paper is like upgrading the operating system of human thought. It moves us from a world where "knowing" is a binary switch (On/Off) to a world where "knowing" is a combination of having the information AND understanding the subject.

By doing this, they created three new logical systems that finally admit: It's okay to be ignorant of things that are logically true, if you just don't get the topic yet. This makes our models of human knowledge much more realistic, humble, and accurate.

Technical Summary

Here is a detailed technical summary of the paper "Ignorance with(out) Grasping" by Ekaterina Kubyshkina and Mattia Petrolo.

1. Problem Statement

The paper addresses two interconnected problems in epistemic logic and formal epistemology:

1. The Lack of Hyperintensional Formalization for Ignorance: While knowledge and belief are widely recognized as hyperintensional notions (where logically equivalent propositions can be treated differently by an agent), ignorance is often treated extensionally in standard logical frameworks. Existing formalizations (e.g., Ignorance Whether, Ignorance as Unknown Truth, Disbelieving Ignorance) rely on standard Kripke semantics, which implies that if $\phi \leftrightarrow \psi$, then an agent's ignorance of $\phi$ is equivalent to their ignorance of $\psi$. This fails to capture the intuition that an agent might be ignorant of a proposition because they do not grasp its content or topic, even if that proposition is logically equivalent to one they understand.
2. The Problem of Logical Omniscience: Standard epistemic logic suffers from the logical omniscience problem, where agents are assumed to know all logical consequences of their knowledge. While this is often framed regarding knowledge, the paper argues it must also be addressed regarding ignorance. Specifically, standard systems often fail to distinguish between an agent being ignorant of a complex tautology (due to lack of conceptual grasp) versus a simple tautology.

2. Methodology

The authors employ Topic-Sensitive Semantics to model ignorance as a hyperintensional notion. This approach integrates the "topic" (subject matter) of a proposition into the truth conditions of ignorance operators.

• Topic Models: They utilize a framework where a topic model $\mathcal{T} = (T, \oplus, t, K)$ consists of:
  • $T$: A set of possible topics.
  • $\oplus$: A fusion operator (idempotent, commutative, associative).
  • $t$: A function assigning a topic to each propositional variable.
  • $K$: The designated topic representing the totality of topics grasped by the agent.
• Grasping Condition: An agent "grasps" a proposition $\phi$ if the topic of $\phi$ (denoted $t(\phi)$) is a sub-topic of the agent's grasped topics ($t(\phi) \sqsubseteq K$).
• Semantic Refinement: The truth conditions for ignorance operators are modified to include a "grasping" clause. An agent is ignorant of $\phi$ not only if the truth conditions of the specific ignorance type are met (e.g., $\phi$ is true but unknown) but also if the agent fails to grasp the topic of $\phi$.
• Logical Systems: The authors define three distinct logical systems, each extending a standard ignorance logic with topic sensitivity:
  1. HIW (Hyperintensional Ignorance Whether): Based on Ignorance Whether ($I_w$).
  2. HIU (Hyperintensional Ignorance as Unknown Truth): Based on Ignorance as Unknown Truth ($I_u$).
  3. HDI (Hyperintensional Disbelieving Ignorance): Based on Disbelieving Ignorance ($I_d$).

3. Key Contributions

A. Three New Hyperintensional Systems

The paper formalizes three forms of ignorance, providing sound and complete axiomatizations for each:

1. System HIW (Ignorance Whether):

   • Definition: $M, w \models I_w \phi$ iff $t(\phi) \not\sqsubseteq K$ OR (the agent does not know whether $\phi$).
   • Key Feature: An agent is ignorant whether $\phi$ if they fail to grasp the topic of $\phi$, even if the truth value of $\phi$ is determinate in all accessible worlds.
   • Axioms: Includes refinements of standard axioms (e.g., $A1_{Iw}$) that condition non-ignorance on the agent grasping the topics of the consequences.

2. System HIU (Ignorance as Unknown Truth):

   • Definition: $M, w \models I_u \phi$ iff $t(\phi) \not\sqsubseteq K$ OR ($\phi$ is true and the agent does not know it).
   • Key Feature: Unlike standard Ignorance as Unknown Truth, $I_u$ in HIU is not strictly factive regarding the agent's state; if the agent does not grasp the topic, they are ignorant even if the proposition is false (though the axiom $A1_{Iu}$ preserves a conditional factivity: if they grasp the topic and are ignorant, the proposition must be true).

3. System HDI (Disbelieving Ignorance):

   • Definition: $M, w \models I_d \phi$ iff $t(\phi) \sqsubseteq K$ AND $\phi$ is true AND the agent believes $\neg \phi$.
   • Key Feature: Crucially, for disbelieving ignorance, the agent must grasp the topic. If the agent does not grasp the topic, they cannot be "disbelievingly" ignorant; they are simply completely ignorant. This distinguishes $I_d$ from $I_w$ and $I_u$ in the hyperintensional framework.

B. Solutions to Logical Omniscience

The authors reframe the logical omniscience problem in terms of ignorance:

• The Problem: If an agent is not ignorant of $\phi$, are they not ignorant of all consequences of $\phi$?
• The Result: In standard Kripke semantics (systems $IW, IU, DI$), certain forms of logical omniscience (like closure under equivalence) still hold. However, in the hyperintensional systems ($HIW, HIU, HDI$), all three forms of logical omniscience fail:
  1. Closure under Implication ($LO^\to_I$): Invalid. An agent can be ignorant of a consequence $\psi$ even if they are not ignorant of $\phi$ (if they don't grasp the topic of $\psi$).
  2. Omniscience Rule ($LO^{NEC}_I$): Invalid. An agent can be ignorant of a valid tautology (e.g., $p \lor \neg p$) if they do not grasp the topic of $p$.
  3. Closure under Equivalence ($LO^{RE}_I$): Invalid. An agent can be ignorant of $\phi$ but not ignorant of $\psi$ even if $\phi \leftrightarrow \psi$, provided the topics differ.

C. Theoretical Justification

The paper provides a robust argument for the hyperintensionality of ignorance independent of the "Standard View" (ignorance = lack of knowledge). Using a modified Stalnaker example (William III and nuclear war), they demonstrate that an agent can be ignorant of a complex proposition (involving nuclear war) while not being ignorant of a logically equivalent simpler proposition (war in general), solely due to a lack of conceptual grasp.

4. Results

• Soundness and Completeness: The authors provide rigorous proofs (via canonical models) that systems $HIW$, $HIU$, and $HDI$ are both sound and complete with respect to their topic-sensitive semantics.
• Differentiation of Equivalent Propositions: The models successfully distinguish between logically equivalent propositions based on the agent's cognitive grasp of their topics.
• Avoidance of Omniscience: The systems successfully model "bounded rationality" where agents are not logically omniscient regarding ignorance. An agent can be ignorant of a tautology if the topic is outside their grasp.

5. Significance

• Bridging Epistemology and Logic: The work bridges the gap between philosophical discussions on the nature of ignorance (e.g., Peels' distinction between complete and disbelieving ignorance) and formal logical systems.
• Refining Hyperintensionality: It expands the domain of hyperintensional notions beyond belief and knowledge to include ignorance, arguing that "thought is hyperintensional" applies to the absence of knowledge as well.
• Practical Application: By solving the logical omniscience problem for ignorance, the framework offers a more realistic model for AI agents and human cognitive modeling, where agents may fail to recognize truths or consequences simply because they lack the conceptual vocabulary (topic) to process them.
• Future Directions: The paper opens avenues for modeling other forms of ignorance (e.g., "complete ignorance") where the absence of grasping is the defining feature, suggesting a rich landscape for future research in topic-sensitive epistemic logic.