Trustworthy Generative AI for Enterprise Software Quality Engineering: Evaluation, Traceability, and Human Oversight

Main Article Content

Brinda Pradeep

Abstract

A variety of software quality engineering tasks such as test design, defect analysis and technical documentation are currently aided by generative artificial intelligence. Outputs produced usually will appear right on the surface but may also have gaps, incorrect logic, or material that may not be able to be referenced to actual requirements. The paper has created a model within which businesses can implement the usage of generative AI in the quality engineering practice in a regulated and reliable way. Within the framework, an evaluation model generates scores that are given to generated artifacts as correct, complete, traceable, reproducible and confident. It introduced provenance mechanisms to relate the artifact produced to its requirement in the source and its supporting evidence. The output can be reviewed on a risk-based review process, when a human is required to check the output before it is used. The paper presents the findings of an experimental test on generated test artifacts in a number of software modules. The results confirm the argument that combined with structured assessment, traceability, and human control, generative AI can be considered reliable.

Article Details

Section
Articles