Towards a generalized Reason model
Year: 2024
Type: article
Abstract: Normative reasoning is the logical study of norms, a concept which is found in a multitude of domains, as norms range from from ethical, legal, social or cultural norms. In computer science, one way norms are used is to specify and regulate the behavior of artificial intelligence (AI) systems. In this context, one of the open problems is that of norm acquisition. The current state of the art consists of two approaches: One uses a declarative representation using a logical language, which is typically encoded by hand. The other uses machine learning to detect and learn norms from training data. Both approaches lack some desired properties like scalability or transparency. The natural idea is therefore to investigate a hybrid approach that combines the two and balances out their strengths and weaknesses. One framework that could be used for such an approach are precedential reasoning models that are developed for legal reasoning. In particular the Reason Model developed by John Horty turns out to be a suitable candidate. The main goal of the developments surrounding the model is to move it away from purely legal reasoning, opening it up to a general application in AI. One of these developments for example is the encoding of the model in a modal logic, which can be used to generate explanations. This work is a step towards generalizing the reason model for the use in AI. The long term goal of this thesis is to develop a framework that provides a transparent, verifiable and explainable representation of normative information that can be acquired automatically.
Source: reposiTUm (TU Wien)
Author: Thölke, Henri
Institution: TU Wien
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6908723153
FWCI:
Cites:
Cited by:
Related to: