On the Interactions of Awareness and Certainty

Hans van Ditmarsch, Tim FrenchAustralasian Conference on Artificial Intelligence 2011 · DOI

TL;DR

This paper delves into the intricate relationship between an agent's knowledge and their awareness in dynamic epistemic logic. It proposes a unified framework to analyze how agents gain knowledge and become aware of new information over time, introducing a novel formalism for "tacit knowledge" and comparing it with existing notions like "latent" and "strong implicit" knowledge.

What Problem Does This Paper Address?

In artificial intelligence and logic, we often model agents' reasoning capabilities. A key challenge is that real-world agents, unlike idealized "logically omniscient" agents, have limited processing power and cannot possibly be aware of every single piece of information or every logical consequence of their knowledge. This paper tackles the problem of formalizing how agents reason when they have incomplete awareness. Specifically, it investigates how an agent's knowledge changes not just when they learn new facts (certainty), but also when they become aware of concepts or propositions they previously weren't (awareness). This is crucial for modeling agents in complex, dynamic environments like the semantic web, where agents need to interact and infer information without necessarily having explicit access to all underlying details.

Methods

The authors extend dynamic epistemic logic by introducing an "awareness function" alongside the standard "accessibility relations" used to model knowledge. This awareness function maps each agent and world state to the set of propositions the agent is aware of in that state. They then define three distinct notions of implicit knowledge:

  • Latent Knowledge: This is a simpler approach where implicit knowledge is essentially knowledge without awareness. It's defined as what an agent would know if they were fully aware of the relevant propositions.
  • Strong Implicit Knowledge: This is a more complex notion where an agent implicitly knows a proposition if it holds true in all possible interpretations of concepts the agent is unaware of. This is formalized using "awareness bisimulations" and "awareness refinements," which are specialized forms of bisimulations that account for awareness states.
  • Tacit Knowledge: This is a novel contribution, described as a middle ground. It captures knowledge that an agent should know or that is intrinsically related to their explicit knowledge, even if they are not directly aware of the concepts involved. It's formalized using "awareness refinements" and allows for situations where an agent might tacitly know something without being able to explicitly articulate it.

The paper also analyzes how these notions behave under "public announcements," which are events where a proposition becomes known to all agents and they become aware of its subformulas.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Awareness is distinct from Certainty: The paper emphasizes that an agent can be certain about something (know it) but unaware of its implications, or aware of a concept but uncertain about its truth.
  • Implicit Knowledge is Crucial for Dynamic Systems: While explicit knowledge suffices for static scenarios, implicit knowledge is essential for modeling how agents' knowledge and awareness evolve over time, especially in dynamic epistemic logic.
  • Three Flavors of Implicit Knowledge: The paper elegantly distinguishes between latent, strong implicit, and tacit knowledge, each capturing different aspects of how agents might reason with incomplete awareness. Tacit knowledge, in particular, offers a way to model "unspoken knowledge" that is hard to communicate.
  • Public Announcements as a Unifying Tool: The framework uses public announcements to demonstrate how agents' knowledge and awareness change, and how these changes interact with the different types of implicit knowledge.
  • Tacit Knowledge as a Compromise: The proposed tacit knowledge formalism offers a more nuanced approach than latent or strong implicit knowledge, allowing for a richer representation of imperfect reasoning.

Key Results

The paper provides formal definitions and semantic frameworks for latent, strong implicit, and tacit knowledge. It establishes relationships between these notions, showing how they differ in their expressivity and implications, particularly in the context of public announcements. For instance, Lemma 2 demonstrates that explicit knowledge implies implicit knowledge, and implicit knowledge implies latent knowledge. The paper also illustrates these concepts with an example of Alice and Bob ordering food, showing how their knowledge and awareness change after an announcement.

Limitations and Future Work

The authors acknowledge that their work presents a comparative analysis and a unified framework. They suggest that different applications might benefit from different types of knowledge and awareness change modeling (e.g., latent for secure communications, tacit for commercial transactions, strong implicit for automated reasoning). The paper highlights that tacit knowledge, while an interesting epistemological concept, has not been as extensively studied in modal logic as strong implicit and latent knowledge. Future work could explore further applications and axiomatizations of these notions, particularly for tacit knowledge.

Summary updated

BibTeX
@inproceedings{ditmarsch2011interactions,
  author = {Hans van Ditmarsch and Tim French},
  title = {{On the Interactions of Awareness and Certainty}},
  booktitle = {Australasian Conference on Artificial Intelligence},
  year = {2011},
  doi = {10.1007/978-3-642-25832-9\_74},
}
Figure