Human Reasoning Module

Abstract

This paper introduces a framework of human reasoning and its ACT-R based implementation called the Human Reasoning Module (HRM). Inspired by the human mind, the framework seeks to explain how a single system can exhibit different forms of reasoning ranging from deduction to induction, from deterministic to probabilistic inference, from rules to mental-models. The HRM attempts to unify previously mentioned forms of reasoning into a single coherent system rather than treating them as loosely connected separate subsystems. The validity of the HRM is tested with cognitive models of three tasks involving simple casual deduction, reasoning on spatial relations and Bayesian-like inference of cause/effect. The first model explains why people use an inductive, probabilistic reasoning process even when using ostensibly deductive arguments such as Modus Ponens and Modus Tollens. The second model argues that visual bottom-up processes can do fast and efficient semantic processing. Based on this argument, the model explains why people perform worse in a spatial relation problem with ambiguous solutions than in a problem with a single solution. The third model demonstrates that statistics of Bayesian-like reasoning can be reproduced using a combination of a rule-based reasoning and probabilistic declarative retrievals. All three models were validated successfully against human data. The HRM demonstrates that a single system can express different facets of reasoning exhibited by the human mind. As a part of a cognitive architecture, the HRM is promising to be a useful and accessible tool for exploring deeps of human mind and modeling biologically inspired agents.