TLDR: A new research paper introduces ‘conjectural logics’, a novel framework for formalizing hypothetical reasoning. Unlike traditional belief or knowledge systems, it uses a principle called Axiom C (if a fact, then true in all hypothetical scenarios) without leading to ‘modal collapse’ by employing paracomplete semantics (Weak Kleene logic, Description Logic) where statements can be ‘undefined’. The paper defines new logical systems (KC, KDC, etc.) that are complete and decidable, and introduces an ‘inclusion theorem’ to classify different cognitive states (epistemic, conjectural, delusional). It also proposes ‘epistemic settlements’ as a dynamic operation to transition conjectures into accepted facts, with potential applications in AI and knowledge representation.
In the realm of artificial intelligence and philosophy, understanding how we reason about beliefs, knowledge, and hypotheses is crucial. Traditional approaches have focused on ‘doxastic logics’ for beliefs (which can be incorrect) and ‘epistemic logics’ for knowledge (which must be true). However, a new research paper introduces a third, distinct category: ‘conjectural logics’.
Authored by Fabio Vitali from the University of Bologna, Italy, the paper titled “From Knowledge to Conjectures: A Modal Framework for Reasoning about Hypotheses” delves into a novel way to formalize hypothetical reasoning. Imagine a system where an agent starts with known facts and then adds unverified assumptions to explore their potential consequences. This is the core idea behind conjectural logics.
The Core Idea: Axiom C and Avoiding Collapse
At the heart of this new framework is a principle called Axiom C, expressed as φ → □φ. In simple terms, this means “if something is a fact, then it is also considered true within all hypothetical scenarios.” Historically, this axiom was often dismissed because, when combined with certain classical assumptions, it led to a ‘modal collapse’ – a situation where all truths become necessarily true, erasing the distinction between what is simply true and what is necessarily true. This would eliminate the concept of possibility or contingency, which is problematic for many philosophical and logical systems.
Vitali’s paper demonstrates that this collapse only occurs under specific conditions, particularly when paired with another principle called Axiom T (□φ → φ, meaning “if something is considered true in all hypothetical scenarios, then it is actually true”) and when using classical, two-valued logic (where every statement is either true or false). To prevent this collapse, the new framework avoids Axiom T and adopts a ‘paracomplete semantic framework’.
Embracing Uncertainty: Paracomplete Semantics
Instead of the usual true/false dichotomy, conjectural logics use a system where propositions can also be ‘undefined’. This is achieved through two main approaches:
-
Weak Kleene Logic: This logic allows for three truth values: true, false, and undefined. A key feature is that if any part of a complex statement is undefined, the entire statement becomes undefined. This is particularly useful for modeling incomplete knowledge, where not all information is available or evaluable.
-
Description Logic: Often used in knowledge representation, Description Logic operates on an ‘open-world assumption’. This means that if something isn’t explicitly known to be true, it’s not assumed to be false; instead, its truth value remains undetermined. This mirrors the ‘undefined’ concept of Weak Kleene logic, allowing for reasoning with incomplete knowledge bases.
By using these paracomplete semantics, the logic can maintain a clear distinction between factual statements and conjectural ones, even with Axiom C in play. This allows for a layered approach to reasoning, where established facts are preserved while exploring hypothetical assumptions.
New Logical Systems and Their Properties
Within this framework, the paper defines new modal systems like KC and KDC, and their extensions KC45 and KDC45. These systems are shown to be complete (meaning all valid formulas can be proven), decidable (meaning there’s an algorithm to determine if a formula is valid), and robust even with partial knowledge. They provide a structured way for agents to build multi-step chains of hypothetical reasoning while maintaining awareness of their commitments and uncertainties.
The Inclusion Theorem and Cognitive Worlds
A fascinating aspect of this research is the ‘inclusion theorem’, which offers a classification of different ‘cognitive worlds’ based on how their knowledge relates to a fixed ‘reality world’ and a ‘shared knowledge world’. This allows for a unified interpretation of various cognitive states:
-
Epistemic Worlds: Where an agent’s knowledge is a subset of reality (knowledge implies truth).
-
Conjectural Worlds: Where an agent’s assumptions extend reality (hypotheses are consistent with known facts).
-
Delusional Worlds: Where an agent’s beliefs contradict reality.
-
Opinion Worlds: Where an agent’s beliefs are disconnected from reality beyond shared knowledge.
This classification provides a powerful tool for representing uncertainty, disagreement, and conflicting opinions within a single, coherent framework.
Dynamic Epistemic Settlements
The paper also introduces a dynamic operation called ‘settle(φ)’. This formalizes the transition from a conjecture to an accepted fact. It’s not about changing an agent’s beliefs, but about updating ‘reality’ itself by assigning a definite truth value to a previously undefined statement. This models real-world scenarios like the outcome of a decision or the discovery of a new empirical fact. As long as independent conjectures exist, cognitive diversity is preserved, but eventually, all conjectures could be settled, leading to a fully defined logic.
Also Read:
- Defining Data Preferences for Inconsistent Knowledge Bases
- Understanding AI Hallucinations and How Models Can Overcome Them
Implications and Future Directions
This new framework has significant potential for practical applications. It could be invaluable in areas like the Semantic Web, where reasoning about conflicting claims and their sources is common. In knowledge provenance, it could help track the credibility of assertions. Furthermore, in neuro-symbolic AI, which combines neural learning with formal logic, this conjectural framework aligns well with systems that share partial knowledge, make conjectures, and dynamically update them as new information becomes definite.
This research offers a unified and formally grounded approach to cognitive logic, bridging the gap between belief, knowledge, and conjecture, and handling partial information without logical collapse. For more details, you can read the full paper here.
