Human-Centred AI in Architecture, Engineering and Construction: Oversight, Explainability, and Organisational Readiness

Abstract

Architecture, engineering and construction have adopted AI faster than they have learned to supervise it. Practices now use generative layout tools, drawing recognition, and schedule prediction on live projects, yet the people accountable for those projects are rarely given the means to interrogate what a model produced or to know when to override it. Progress is measured in model accuracy, while the questions a practising architect asks, whether the output can be checked, who signs it off, and what happens when it is wrong, remain outside the evaluation. This session treats those questions as the research problem rather than as deployment detail. It brings together work on human oversight of AI in design and construction workflows, explainability for spatial and drawing-based data, trust calibration among domain professionals, and organisational readiness for responsible adoption. We welcome empirical studies, system contributions, frameworks, and critical analyses. The aim is a programme that connects the conference’s first topic, trustworthy and explainable human-AI interaction, to a domain that builds the environment everyone else studies interaction within, and where the consequences of uncalibrated automation are physical and long-lived.

Acronym: HCAI-AEC

Scope and Topics

Submissions are invited on, but not limited to:

• Human oversight and function allocation in AI-assisted design and construction workflows
• Explainability and interpretability for spatial data: floor plans, technical drawings, point clouds, BIM models
• Trust calibration and appropriate reliance among architects, engineers and site personnel
• Organisational readiness, governance and maturity for responsible AI adoption in practice
• Evaluation methods for AI tools embedded in long-horizon, multi-stakeholder professional work
• Accountability, auditability and sign-off when AI contributes to a regulated deliverable
• Human-AI collaboration in generative and computational design
• Failure, error and recovery in AI-assisted technical work
• Field studies of AI adoption in practices, contractors and public clients
• Inclusive and accessible design practice supported by AI

Rationale

Three things make this timely, and none of them is that AI is popular.

The adoption curve has outrun the oversight curve. RIBA’s own surveys put AI use in architectural practice at 41% of firms in 2024 and 74% in 2026, while the share of practices investing in AI research and development stayed in the low twenties. Tools are arriving in offices that have no internal capacity to evaluate them. This is precisely the condition human-centred AI research exists to address, and it is happening at scale in a profession HCI rarely studies.

The domain breaks assumptions the HCI literature is built on. Much human-AI interaction work assumes a single user, a short decision horizon, and a reversible outcome. AEC offers none of these. A design decision is made by a team over months, is legally attributable to a named person, and becomes a building that stands for fifty years. Trust calibration, function allocation and explainability all need reworking under those constraints, and that reworking is a contribution to HCI, not merely an application of it.

The technical literature and the interaction literature are not talking. Floor-plan understanding, scan-to-BIM and generative layout have strong benchmark cultures and almost no evaluation of the human workflows they enter. Meanwhile human-centred AI work on oversight and explanation rarely takes spatial or drawing-based data as its object. A session is the right instrument for putting the two in one room.

IHCI is the right venue rather than an AEC conference because the questions are interaction questions. An AEC audience will ask whether the tool works. This audience will ask who it works for, and that is the harder and more useful question.

Mapping to the Conference Topic List

The session sits primarily under Human-AI interaction and explainability; trustworthy, ethical, and responsible AI, and draws on Computer vision and vision-based interaction, Design methods, prototyping, evaluation, and reproducible HCI, Human-environment interaction and situated interfaces, and Education, healthcare, sustainability, transportation, and other application domains.

Organiser

Vivek Kumar Chenna · National Institute of Technology Calicut, India · vivekkumarchenna@nitc.ac.in

Assistant Professor in Construction Engineering and Management, and Associate Dean (Planning and Development), where he heads campus engineering and capital projects. He directs an engineering unit of forty professionals delivering a capital portfolio of about US$500M across more than eighty concurrent projects. Before joining the Institute he served as Architect and Project Manager in the Central Public Works Department, Government of India, selected through the UPSC national civil service examination, with owner-side accountability for over seventy buildings and 7.4 million square feet. His doctoral research is on AI for the built environment, and he has designed and deployed production AI software for construction operations, code-compliance checking and project documentation.

The organiser will serve as session chair on site and will attend for the full duration of the conference.

Expected Submissions and Session Structure

Target: 5 papers, within the 4–6 the call specifies.

Seed submissions. Two papers submitted to IHCI 2026 by the organiser and colleagues fall within the session scope and would be requested for it if accepted:

1. A maturity framework for AI readiness in architectural practice, covering eight dimensions including governance and human oversight, with non-compensatory aggregation so that a governance deficit cannot be offset by technical capability.
2. A human-centred taxonomy of floor-plan understanding research, mapping the technical literature against oversight, explainability and trust requirements, and deriving design principles and a research agenda.

These are declared here for transparency. Both will be reviewed under the same process, by the same standards, as every other submission, and the organiser will not handle the review of his own work.

Structure (60 minutes):

If 40 minutes is allocated instead, the format compresses to four presentations at 8 minutes plus a 5-minute close, and the framing is folded into the chair’s introduction of the first paper.

We would prefer the moderated close over an invited talk. An invited talk adds a voice; a structured close makes the session produce something the five separate papers do not, which is the point of running it as a session rather than as five slots.

Dissemination Plan

• Direct invitation to authors publishing on AI in AEC at CAADRIA, eCAADe, ACADIA, CIB W78 and Automation in Construction, a community that is productive, adjacent, and largely absent from HCI venues. This is the main channel, and the one that justifies the session.
• Circulation through the architectural computing and construction informatics groups at Indian NITs and IITs, and through the RIBA and CIOB digital practice networks.
• Posting to the CHI-ANNOUNCEMENTS and BCS Interaction lists, and to the relevant special interest groups on ResearchGate and LinkedIn.
• Direct approach to authors of recent work on explainability for spatial data and on trust in AEC automation, identified from the citation graphs of the two seed papers.

Chairs

• Vivek Kumar Chenna, National Institute of Technology Calicut, India