Abstract
Generative artificial intelligence (GenAI) offers much promise. However, despite GenAI’s obvious utility, recent tragedies related to inappropriate outputs, including encouragement of self-harm and suicide, highlight the urgency with which developers must seek improvements. This study introduces the ETHOS framework (Evaluation Through Holistic Oversight and Safeguards) to support the responsible integration of GenAI into decision support systems (DSS). Its goal is to offer a transparent, accountable alternative to opaque AI moderation. ETHOS merges ethical and social principles of ancient texts with Floridi and Cowl’s unified framework of five principles for AI in society. We propose operationalizing a human-in-the-loop evaluation approach to ETHOS through Distributed Cognition Theory, enabling iterative, threshold-based assessments. Outputs falling below defined benchmarks are flagged for human review, allowing automation to continue without compromising oversight. ETHOS addresses inherent GenAI risks through culturally grounded, scalable evaluations offering a replicable model for integrating distributed ethical reasoning into DSS while preserving human judgment.
Recommended Citation
Patel, Hrishitva; Menard, Philip; and Abhishek, K.V., "ETHOS Framework: A Human-in-the-Loop Approach to Responsible GenAI Deployment in Decision Support Systems" (2025). WISP 2025 Proceedings. 19.
https://aisel.aisnet.org/wisp2025/19