AI for Computational Design Science: A Responsible Human-AI Framework and Case Study on Short-Form Video Safety Surveillance

The provided search context does not contain information regarding an "AI4CDS" framework or a case study on "Short-Form Video Safety Surveillance."

The available sources describe alternative frameworks for responsible AI and computational design:

• GeDSR Framework: A generative design science research approach emphasizing responsible autonomy, AI self-explainability, AI bootstrapping, and knowledge-informed machine learning (KIML) to enable self-improving AI systems.
• Human-Centered AI (HCAI): Focuses on integrating empathy, user inclusion, and continuous feedback to ensure AI augments human capabilities rather than replacing them.
• SoS-ML Blueprint: A System-of-Systems Machine Learning framework that integrates social-cognitive frameworks to enhance explainability and functional interpretability through modular agents.
• Responsible AI System (RAIS) Framework: Builds societal trust through domain definition, trustworthy AI design, auditability, accountability, and governance.

No evidence supports the existence or details of the AI4CDS framework in the provided text.

AI4CDS framework. The paper develops AI4CDS, a responsible human-AI framework for Computational Design Science (CDS). Rather than treating AI as a downstream optimization tool, it positions AI as an active participant across the design-science cycle: helping formulate ambiguous problems, search over candidate designs, evaluate alternatives, and abstract reusable design knowledge. The framework is organized around five phases, with explicit human oversight to preserve accountability, interpretability, and ethical responsibility in AI-assisted design.

Case study on short-form video safety surveillance. The material applies the framework to short-form video safety surveillance, a high-stakes platform-governance domain where content risk, user safety, and operational trade-offs intersect. In this setting, AI supports the design and evaluation of safety-related monitoring or moderation mechanisms—such as identifying risk signals, proposing intervention designs, and assessing outcomes against criteria like safety, fairness, privacy, and feasibility—while humans retain authority over value judgments and deployment decisions.

Why it matters. The main contribution is methodological: it offers a transferable template for integrating AI into design-science research without reducing the process to opaque automation. This is important because safety-critical digital platforms pose complex design problems where naive AI use can obscure assumptions, amplify harms, or produce unaccountable recommendations. By making the human-AI division of labor explicit, the work provides a basis for more rigorous, auditable, and socially responsible AI-augmented design.