Introduction
In the landscape of formal epistemology and mathematical economics, a sharp distinction exists between uncertainty and unawareness. Standard models of knowledge, such as Aumann’s state-space models, excel at representing uncertainty: an agent may not know whether a coin landed heads or tails, but they possess the concepts of both outcomes. Unawareness, however, represents a more fundamental "lack of conception." It is the state where an event is not merely unknown, but entirely absent from the agent's mind.
Historically, modeling unawareness has proven difficult due to the DLR Impossibility Theorem (Dekel, Lipman, and Rustichini, 1998), which demonstrated that non-trivial unawareness is logically impossible within standard state-space frameworks if certain intuitive axioms are maintained. This paper introduces a novel "partial-state space" model that utilizes possibility semantics to bypass these theoretical hurdles and provide a more robust representation of an agent's mental state.
The Limitations of Complete-World Paradigms
Traditional models rely on "total states"—complete descriptions of the world that settle every possible truth. This approach creates two primary issues:
- The DLR Constraint: In a standard state space, if an agent is unaware of an event, the logic of the model often collapses into a state where the agent is either aware of everything or the model becomes trivial.
- The Knowledge-Awareness Link: Many existing frameworks define unawareness as a derivative of knowledge (e.g., "I am unaware of $E$ if I don't know $E$ and don't know that I don't know $E$"). This definition fails to account for overconfident agents. For example, an investor who falsely believes they understand a complex financial instrument is psychologically "aware" of it, even if their knowledge is incorrect. Traditional definitions would incorrectly label such an agent as unaware.
Methodology: Possibility Semantics and Partial Specifications
The author departs from the "complete world" paradigm in favor of Possibility Semantics, a framework derived from modal logic and set-theoretic "forcing."
Partial Specifications
Instead of states that define every variable, the model uses a partially ordered set (poset) of "possibilities." A possibility is a partial specification; it might determine that "Ann plays Up" while leaving the actions of other players undetermined. This allows the model to represent different levels of conceptual granularity.
Regular Open Sets and Awareness Correspondences
Events in this model are represented as regular open sets within a downset topology. This mathematical choice allows for a Boolean algebra of events where negation is not a simple set-theoretic complement, providing the flexibility needed to model complex epistemic states. The author introduces an Awareness Correspondence ($A$) on the state space, treating awareness as a primitive operator rather than a derivative of knowledge. This decoupling is essential for representing agents who lack information without necessarily lacking the underlying concepts.
Key Theoretical Results
The core of the paper is the Representation Theorem (Theorem 4.6). The author defines "epistemic awareness algebras"—abstract algebraic structures that capture the logic of knowledge and awareness—and proves they are isomorphic to the proposed partial-state space structures. This result provides a solid mathematical foundation for the framework.
Furthermore, the paper demonstrates that these "epistemic possibility frames" satisfy the essential axioms of the literature while successfully bypassing the DLR impossibility result. The author identifies that the DLR result hinges on a "Plausibility" axiom that conflates awareness of events with awareness of sentences. By distinguishing between the two, the model turns a previous "no-go" theorem into a viable "possibility" result.
Insights and Implications
Non-Monotonicity of Awareness
A significant insight of the model is that awareness is not necessarily monotonic. An agent might be aware of a broad event (e.g., "The market is volatile") without being aware of the specific sub-events that compose it (e.g., specific derivative failures) if they lack those specific concepts. This mirrors real-world cognitive limitations where understanding a general concept does not grant automatic awareness of all its implications.
Modeling Overconfidence
By treating awareness as a primitive, the model can represent an investor who is aware of "Profitability" but entirely unaware of the possibility of "Fraud." Such an agent can be overconfident about the former precisely because they cannot conceive of the latter. This provides a more realistic tool for behavioral economics than models that link awareness strictly to knowledge.
Conclusion and Future Directions
While this partial-state space model provides a powerful alternative to traditional frameworks, it remains a "non-hyperintensional" event-based model. Future research may focus on creating a "bridge" to sentence-based awareness, where the model could distinguish between an agent being aware of "A and B" versus "B and A."
Additionally, the current framework is qualitative. Integrating quantitative probability (multi-prior representations) and modeling the dynamics of awareness—how agents are "made aware" of new risks or opportunities—remain vital areas for the evolution of decision theory and game theory.
Summary updated