TL;DR
This paper revisits the problem of modeling "knowledge of unawareness" in epistemic logic. The authors propose a new logic that extends previous work by allowing different languages to be associated with different possible worlds, enabling agents to be uncertain about whether they are aware of all possible formulas. This new framework provides a sound and complete axiomatization and bridges the gap between existing approaches in computer science and economics.
What Problem Does This Paper Address?
Epistemic logic deals with reasoning about knowledge. A key challenge is modeling "unawareness" – situations where an agent doesn't know something simply because they aren't aware of the relevant concepts or propositions. Early work (Fagin & Halpern) introduced an "awareness operator" but struggled to model agents who might not even be aware of all possible formulas. Later work (Modica & Rustichini, Heifetz, Meier, & Schipper) proposed associating different sets of primitive propositions (languages) with each world. While this allowed for more nuanced awareness, a specific logic by Halpern & Rêgo (HR) that extended the syntax to allow quantification over formulas ran into a problem: it was impossible for an agent to be uncertain about whether they were aware of all formulas. This paper addresses this limitation by combining the expressive power of HR's quantified logic with the world-dependent languages of Modica & Rustichini, thereby enabling agents to reason about their own potential unawareness.
Methods
The core contribution is a new modal logic that extends the syntax of the HR logic (allowing quantification over formulas) but modifies the semantics. Instead of a single language for all worlds, each world in the model is now associated with its own set of primitive propositions (a language). This "extended awareness structure" allows an agent's awareness to be a proper subset of the language of a world they consider possible. The authors provide a sound and complete axiomatization for this new logic, carefully adapting existing axioms and introducing new ones to handle the interaction between quantification and world-dependent languages. They also compare their logic to existing approaches, showing how it can capture the essence of earlier work while overcoming their limitations.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- World-Dependent Languages are Key: The crucial insight is that allowing each possible world to have its own language is the key to modeling uncertainty about awareness itself. If an agent considers a world possible where the language is richer than their current awareness, they can naturally reason about what they might be unaware of.
- Bridging CS and Economics: The paper elegantly connects formalisms from computer science (epistemic logic) and economics (awareness structures with world-dependent languages), showing how a unified logical framework can address problems in both fields.
- Quantification and Awareness: The paper demonstrates how to effectively combine quantification over formulas with awareness operators, enabling statements like "an agent knows that there exists a formula they are unaware of."
- Awareness as a "Label": The authors suggest that primitive propositions an agent is unaware of in a given world can be thought of as "labels" for concepts. This perspective helps explain why agents can reason about things they don't explicitly know.
Key Results
- A new modal logic is proposed that allows agents to reason about their own potential unawareness.
- A sound and complete axiomatization is provided for this new logic.
- The paper shows that under natural assumptions, the quantifier-free fragment of their logic is characterized by the same axioms as the approach of Heifetz, Meier, and Schipper (HMS).
- The proposed logic successfully overcomes the limitation of the HR logic where agents could not be uncertain about being aware of all formulas.
Limitations and Future Work
- Computational Notion of Awareness: The authors acknowledge that their model of awareness is syntactic. They suggest that a more "computational" notion of awareness might require different axioms and leave this as a direction for future work.
- Dynamics of Language Change: The paper touches upon the idea of how languages might change but does not address this dynamic aspect, suggesting it as a topic for further research.
- Infinite Languages: The proofs for completeness rely on the assumption of infinite languages, which is a standard technique but might be a limitation in certain practical applications.
Summary updated