Human-in-the-loop (HITL) as a Verification and Validation Strategy for Knowledge Generated by Generative artificial intelligence

Resumen

The Human-in-the-Loop (HITL) approach describes human participation in various stages of artificial intelligence system development. This research identifies the methods employed by end users for verifying and validating knowledge generated by generative artificial intelligence (GAI). A systematic literature review was conducted following the PRISMA protocol to analyze the methods used for knowledge verification and validation in the context of the HITL approach. The search equation, developed using a generative AI tool, was applied to the Scopus database and the AI-powered search engine Undermind, retrieving a total of 95 documents. After applying inclusion and exclusion criteria, 19 articles were selected for analysis. The findings allowed for the categorization of the identified methods into two groups: those used in the design and implementation stages of GAI systems and those employed by end users. However, persistent challenges remain, particularly the lack of detailed specification and formalization of knowledge verification and validation methods at the end-user level, which impacts the accuracy of responses and the control of generated knowledge creativity. Future research should focus on specifying, testing, and formalizing these methods to optimize their application within the HITL framework. This study contributes to the field by providing a set of methods for verifying and validating knowledge generated by GAI, thereby improving response accuracy and control over creativity.