Costanza Larese presented her work at the 10th International Conference on Model-Based Reasoning (MBR026_Rome): Epistemology, Artificial Intelligence, Cognitive Science.
The conference was held at the Department of Philosophy Sapienza University of Rome, from 17 June to 19 June 2026.
Speaker: Costanza Larese (joint work with Hykel Hosni and Francesco Ponti)
Title: Polya’s fundamental inductive pattern for medical diagnostics
Abstract: In Mathematics and Plausible Reasoning (1954), George Polya offered one of the earliest systematic accounts of plausible, non-deductive reasoning patterns, among which the fundamental inductive pattern (FIP) stands out for its simplicity and generality. Polya begins from a familiar situation: a conjecture A implies a consequence B, while neither is yet known to be true. If B is shown to be false, classical logic (via modus tollens) allows us to conclude definitively that A is false. But if B is found to be true, the result is different: A is not proven, yet it becomes more credible. The FIP thus captures a form of plausible rather than demonstrative reasoning: confirming a predicted consequence strengthens confidence in a hypothesis without establishing it as true. Despite its relevance to scientific reasoning, this pattern has received comparatively little attention within the logical community.
In this paper, we propose a formal reconstruction of the FIP by introducing a variant within the framework of adaptive logics, a family of systems specifically designed to model non-monotonic and dynamically revisable inference. Formally, our system, ALrP⋆ is built on the monotonic lower-limit logic P⋆, a first-order preferential logic with an object-language default connective, and incorporates a set of abnormalities and an adaptive strategy that regulate the defeasible application of the FIP. The behaviour of the system is illustrated through non-trivial examples from medical diagnosis, where typical clinical signs increase the plausibility of a disease without ever entailing it, and where alternative explanations or incompatible background information may suspend or defeat the inference. In this way, we demonstrate that Polya’s reasoning schemas can be integrated into a precise formal framework capable of representing central features of medical diagnosis.