Lifecycle Ethical Analysis (LEA) for AI-based Systems

18 Pages Posted: 31 Oct 2022 Last revised: 23 Nov 2022

Date Written: October 21, 2022

Abstract

Deciding to move forward with an AI-based solution requires analysis that combines ethical reasoning with disciplined engineering practices in the AI lifecycle. However, the distinctiveness of these domains is better served by effectively translating between the two than by literally combining them. The Lifecycle Ethical Analysis (LEA) process constitutes a robust approach for effecting this translation while still allowing for the benefits of the more traditional approaches of engineering ethics and social impact assessment. LEA's theoretical foundations, derived from work in communications theory and science and technology studies, enable this framing.

LEA consists of three phases that correspond to key high-level processes in AI-based system development. These are (1) context definition and analysis, (2) data and model development, and (3) solution validation and deployment. For each of these, potential obstacles are discussed and candidate supporting methods are described. Each method results in a distinct artifact that can act as a "boundary object" which serves as a common point of reference for both ethical and technical reasoning. By assigning methods to such coarse buckets, LEA's activities can be factored and mapped to any specific engineering lifecycle, facilitating tailored implementation.

Keywords: Ethical AI, LEA, AI lifecycle

Suggested Citation

Shapiro, Stuart and Petrozzino, Catherine, Lifecycle Ethical Analysis (LEA) for AI-based Systems (October 21, 2022). Available at SSRN: https://ssrn.com/abstract=4255117 or http://dx.doi.org/10.2139/ssrn.4255117

Stuart Shapiro (Contact Author)

MITRE Corporation ( email )

202 Burlington Road
Bedford, MA 01730
United States

Catherine Petrozzino

The MITRE Corporation ( email )

7515 Colshire Blvd.
McLean, VA 22102
United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
34
Abstract Views
248
PlumX Metrics