A Risk-Based Framework for Managing Errors in AI-Based Engineering Chatbots: Project Consequences, Control Gates, and Accountability
A Risk-Based Framework for Managing Errors in AI-Based Engineering Chatbots: Project Consequences, Control Gates, and Accountability
Hanieh Rezaeian1 Amirhossein Rezaeian2
1) M.Sc. Student in Engineering Education, Faculty of Engineering Science, University of Tehran, Tehran, Iran Email:
2) M.Sc. in Mechatronics Engineering, University of Sistan and Baluchestan, Zahedan, Iran Email:
IOI Code
:
ABEB-ACAFZF
Publication :
4th International Conference on Recent Advances in Engineering, Innovation and Technology - Belgium(eitconf.com)
Abstract :
AI-based chatbots are increasingly used to support engineering design, project planning, technical documentation, risk analysis, and operational decisions. Their outputs may contain fabricated sources, incorrect calculations, obsolete requirements, omitted assumptions, or recommendations that do not fit project conditions. This paper develops a risk-based conceptual explanation of how such errors travel from model output to project consequence and how control and accountability can interrupt that process. The study uses a critical integrative review and theory-synthesis approach, combining research on generative-AI reliability, human reliance on automation, engineering project risk, and professional governance. The analysis identifies a multi-stage propagation mechanism and derives a six-gate governance architecture, E-TRACE, comprising task-risk evaluation, input traceability, constrained generation, independent audit, accountable release, and incident learning. Its theoretical contribution is to explain the interaction among task criticality, evidence quality, human reliance, control-gate effectiveness, and organizational accountability, rather than merely listing chatbot risks. An error-propagation model, a responsibility ledger, boundary conditions, and testable propositions are developed. No experimental, survey, interview, or field data are claimed; E-TRACE is a theoretically derived framework requiring subsequent empirical evaluation.
Keywords : engineering chatbots generative ai project risk human oversight error propagation accountability