Year: 2023
Type: dissertation
Abstract: In order to meet user demands, engender properly placed trust, and meet the informational demands from accountability and contestability transparency is an essential quality in artificial intelligence systems. Here the semantic transparency framework is presented, a novel approach linking transparency to system structure in a way that allows for opaque components and is agnostic to implementation methods. This is achieved by capturing demands on the system in inquisitive propositions which are linked to the ability of system structure to provide answers by a shared semantic description of both demands and system. The use-case and audience-dependent transparency - which results from this approach - is categorized into distinct levels which show that common stakeholder concerns can be addressed in a transparency-by-design manner in systems with opaque parts without resorting to more complicated explainability techniques.
Author: Brännström, Mattias
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7006427677