GAD Editorial · Published · · English study · Chinese summary · Study dated 2026-10-02
An AI-generated material shortlist is easiest to evaluate when the team can follow each suggestion back to a specific layer, requirement and source. A wall assembly offers a useful starting point: changing one component creates questions about the surrounding construction as well as the component itself.
A DIS2025 Companion paper by Shu Zhong and colleagues describes a research prototype that turns wall-assembly sketches into graph representations for material exploration. Its preliminary study involved 11 architects in one-hour sessions comparing typical and AI-assisted tasks. Those findings concern a small workflow study, rather than verified carbon savings across completed buildings. This is an evergreen review of 2025 research, not a new product announcement. Research description · Original paper
Define the function of the layer
GAD proposes beginning a substitution exercise with a simple assembly record. List the layers in order, their intended roles, thickness assumptions and the source of each stated property. Mark unknown information explicitly. Identify the design question: a team may be exploring lower-impact insulation, a different external finish or a more maintainable internal lining.
Set the comparison conditions before requesting alternatives. Ask the responsible designer which performance requirements must remain and which can change. Two materials offered in different quantities or for different service conditions should not become directly comparable merely because a generated table places them in adjacent rows.
Keep the evidence beside the suggestion
For each candidate, require a product identity, a retrievable technical source and the date checked. Record the unit, declared scope and conditions attached to any environmental or performance figure. Where the service supplies a secondary summary, follow it to the underlying document before adopting the value.
Ask the assistant to show missing evidence as missing. A blank field creates a clear research task; a plausible but unsupported figure can contaminate every comparison that follows. Keep a separate note for assumptions introduced by the team so they are not mistaken for manufacturer declarations.
Review the assembly after the substitution
An illustrative insulation change should trigger a review of the proposed build-up, interfaces and detailing questions. Invite the appropriate envelope, fire, acoustic or structural specialist according to the project and the changed component. This article does not prescribe a compliant assembly or substitute for that review.
Use a before-and-after record that identifies the exact layer changed, the evidence supporting it and the consequences still unresolved. Avoid calling the candidate “equivalent” until the relevant requirements have been checked. If several layers must change together, treat them as one coordinated alternative and explain the dependency.
Measure a useful outcome
For an office trial, count the time required to locate reliable sources, reject unsuitable candidates and prepare an alternative for expert review. Also record how often the system supplied an unusable source or omitted a critical condition. A shorter shortlist is valuable when its reasoning can be inspected.
The research provides an interesting interface direction, but its prototype status should remain explicit. GAD has not tested the system. The practical deliverable is a reviewable assembly comparison with named evidence gaps, rather than a material recommendation detached from the construction it is supposed to serve. Institutional publication record