Repository logo
 

A Cognitive Engineering Approach to Transparency of Contrastivity of AI Algorithms

Files

Thesis(3.81 MB)

Date

2025

Authors

Obolonkova, Xeniya

Supervisor

Wong, William

Item type

Thesis

Degree name

Master of Computer and Information Sciences

Journal Title

Journal ISSN

Volume Title

Publisher

Auckland University of Technology

Abstract

The increasing adoption of Artificial Intelligence (AI) within critical decision-making domains has intensified the need for transparency, fairness, and explainability in model design and operation. While technical methods for post-hoc explainability have advanced, their integration into system architectures capable of addressing societal, psychological, and governance concerns remains limited. This paper proposes a conceptual framework for AI model transparency that integrates post-hoc interpretability techniques within the principles of Ecological Interface Design (EID)(Vicente, 1995). We validate the applicability of a cognitive engineering approach - specifically, Cognitive Work Analysis (CWA) (Rasmussen, 1985) and Work Domain Analysis (WDA) - to achieve greater model transparency in the area of textual analysis. The framework leverages abstraction hierarchy modelling and constraint visualisation to connect lower-level elements- such as features and coefficients to higher-order functional and relational representations, enabling multi-level reasoning about model behaviour. The approach addresses fairness assessment, bias mitigation, and reasoning quality evaluation for both individual and group predictions, incorporating reasoning in model explanations (Miller, 2018) into “Explanation Contrastivity” metric to make causal reasoning explicit.

Description

Keywords

Source

DOI

Publisher's version

Rights statement

Permanent link

http://hdl.handle.net/10292/20106

Collections

Masters Theses