Skip to content

teletobe/audit-share

Repository files navigation

Master Thesis for Lowering the Threshold for Embedded AI Ethics

Designing Self-Assessment Tools for Stakeholder-Identified Ethical Concerns

Abstract:

Unethical technology, such as systems that discriminate, exploit, cause harm, or violate privacy, often arise from developers' struggles to define, assess and implement ethics during development. The increasing pace of technological advancement, the ubiquitous deployment of AI, and the complexity of shareholder considerations add to this challenge. This thesis seeks to investigate key questions about ethical technology audits: what are the relevant values and ethical concerns involved? And how can these be translated into a testable environment? In particular, it explores the role of stakeholders in identifying relevant values and creates tools that assist in translating them into the auditable artefacts of Ethical Focus Areas (EFAs). The resulting workflow are the Ethics Self-Assessment Tools (ESAT). A mixed-methods approach is employed throughout the thesis to critically analyse the problem, current approaches and their gaps, and to develop a participatory workflow for eliciting values and concerns. Through trial workshops, a streamlined and approachable method that generates ethical concerns as data is established. Additionally, a straight- forward webtool to process and translate this data into testable artefacts is developed, making the ethics assessment more accessible to non-ethical-expert developers. Context- sensitive applications, such as LLM-based advice chatbots, are the focus of this thesis. However, the transferability to other systems is also discussed. The results indicate that by rooting the concerns directly in stakeholders' desires and expectations, the ESAT workflow builds towards a comprehensive toolkit that integrates auditing procedures early on in an AI's development cycle, enabling a continuous, context- sensitive, and adaptable assessment of a technology's ethical concerns.

Project Structure

.
├── README.md                 # This file
├── thesis_.pdf               # Main thesis document
├── informed-consent.pdf      # Informed consent forms
├── data-protection.pdf       # Data protection guidelines
├── activity-materials.pdf    # Activity materials
├── survey.pdf                # Survey materials
├── workflow-diagram1.png     # Project workflow diagram
├── workflow-diagram2.png     # Project workflow diagram
├── workshop-sample-slides.pdf # Workshop presentation slides
└── webtool/                  # Web-based auditing tool
    └── README.md             # Webtool specific documentation

Webtool

The webtool/ folder contains an interactive web-based application for digitising data from workshops to audit AI technologies. See webtool/README.md for detailed documentation on how to run and use the webtool.

Additional Resources

  • Thesis Document: See thesis_.pdf for the complete thesis
  • Chapter 5.4: Contains detailed screenshots and demonstrations of the webtool interface

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors