Key points are not available for this paper at this time.
While Generative AI (GenAI) holds transformative potential for Interdisciplinary Collaborative Design (ICD), its practical deployment is critically hindered by an inability to ensure engineering feasibility, reliability, and compliance, compromising the trustworthiness of its black-box outputs. To surmount this challenge, this paper proposes a mechanistic-priority framework wherein GenAI contributions are contingent upon satisfying three core engineering invariants: Representation, Reasoning, and Realisation (R3). This R3 theory underpins a novel Agent-based Workflow Orchestration (AWO) architecture that operationalises a closed-loop process of generation, validation, and governance. The architecture, whose resilience is fortified by proactive Failure Mode and Effects Analysis (FMEA), systematically guides GenAI towards verifiable solutions. Framework efficacy is substantiated through a four-stage process (Work Packages A-D), highlighted by a micro-benchmark experiment that quantitatively assesses the impact of applying R3 invariants. Key performance indicators are synthesised into a holistic symbiosis pilot scorecard, which serves as a reference model for interdisciplinary projects. This research thus delivers both the theoretical foundations and an operational framework for governable GenAI in complex engineering, reframing human-AI collaboration as an engineering problem predicated on verifiability and accountability to advance the paradigm of synergistic symbiosis.
Gao et al. (Thu,) studied this question.