PaperlayerView exact revision on arXiv

Source-linked AI summary

Generative AI in the Construction Industry: Opportunities & Challenges

Prashnna Ghimire, Kyungki Kim, Manoj Acharya

arXiv:2310.04427v1cs.AIcs.LG

TL;DR

Construction lags in AI adoption, and the opportunities and challenges of implementing GenAI in the sector remain insufficiently studied. The paper reviews literature, analyzes construction practitioners’ perceptions, and integrates the authors’ expertise to synthesize applications, barriers, and future research needs. It proposes a conceptual implementation framework while emphasizing risks involving hallucinations, accuracy, outdated models, and broader adoption challenges.

  • Problem

    Research has not adequately investigated GenAI opportunities, applications, and implementation challenges in construction, despite the sector’s slower adoption and the need to guide early integration.

  • Method

    The study combines a non-systematic literature review, snowballing, LinkedIn practitioner-perception analysis, and the authors’ expertise.

  • Results

    The study synthesizes perceived opportunities and challenges, identifies potential applications, and proposes a conceptual GenAI implementation framework for construction.

  • Takeaways & Limitations

    The paper provides foundational guidance for early GenAI adoption and identifies future research needs in construction and allied architecture and engineering domains.

  • Takeaways & Limitations

    GenAI outputs may be convincing but false, inaccurate, and disconnected from real-world construction complexities, requiring validation and verification.

Abstract

from arXiv · show

In the last decade, despite rapid advancements in artificial intelligence (AI) transforming many industry practices, construction largely lags in adoption. Recently, the emergence and rapid adoption of advanced large language models (LLM) like OpenAI's GPT, Google's PaLM, and Meta's Llama have shown great potential and sparked considerable global interest. However, the current surge lacks a study investigating the opportunities and challenges of implementing Generative AI (GenAI) in the construction sector, creating a critical knowledge gap for researchers and practitioners. This underlines the necessity to explore the prospects and complexities of GenAI integration. Bridging this gap is fundamental to optimizing GenAI's early-stage adoption within the construction sector. Given GenAI's unprecedented capabilities to generate human-like content based on learning from existing content, we reflect on two guiding questions: What will the future bring for GenAI in the construction industry? What are the potential opportunities and challenges in implementing GenAI in the construction industry? This study delves into reflected perception in literature, analyzes the industry perception using programming-based word cloud and frequency analysis, and integrates authors' opinions to answer these questions. This paper recommends a conceptual GenAI implementation framework, provides practical recommendations, summarizes future research questions, and builds foundational literature to foster subsequent research expansion in GenAI within the construction and its allied architecture & engineering domains.

1. Introduction

Construction has adopted AI more slowly than other industries, while GenAI creates new opportunities that remain insufficiently studied. This study frames that gap through existing evidence, industry perspectives, and two research questions about applications, challenges, and future research.

  • 1. Introduction: Traditional machine-learning models remain constrained by limited precision–generality trade-offs and difficulty accounting for human errors and unforeseen construction conditions.Examples include balancing project-completion prediction accuracy against applicability and predicting fall risk without representing every accident factor.
  • 1. Introduction: GenAI and LLMs can synthesize novel text, images, and audio from learned representations, motivating their investigation in construction.The paper identifies GPT, PaLM, and Llama as advanced LLM examples and describes GenAI as a deep-learning subset using supervised, unsupervised, and semisupervised approaches.
  • 1. Introduction: Construction trails other industries in AI adoption, despite AI applications spanning safety, cost, scheduling, monitoring, quality, logistics, risk, sustainability, and BIM.Existing work also links robotics and AI with construction quality, safety, project acceleration, and labor-shortage mitigation.
  • 1. Introduction: The study addresses a research gap by asking about current evidence, opportunities, applications, challenges, and priority questions for GenAI in construction.Its planned sections cover methodology, model structures and related work, opportunities and an implementation framework, challenges, recommendations, and future research.

2. Methodology

The study combines a non-systematic literature review, snowballing, and analysis of recent LinkedIn posts to examine GenAI in construction. It also incorporates the authors’ combined expertise in AI, automation, and generative AI for the built environment.

  • 2. Methodology: The study searched Scopus and Google Scholar using GenAI-and-construction keywords, then snowballed through references and citations.The review was non-systematic and intended to capture the limited and emerging literature on GenAI in construction.
  • 2. Methodology: Recent construction professionals’ perceptions were collected from LinkedIn posts published during the three months preceding August 20, 2023.The search used three keyword combinations covering generative AI, construction, and architecture, engineering, and construction.
  • 2. Methodology: The research followed a framework that combines literature evidence, practitioner perceptions, and author perspectives to investigate GenAI opportunities and challenges.The authors contributed more than a decade of combined experience across AI in construction, construction automation, and generative AI in the built environment.

3. Various GenAI Model Structures and Related Work in Construction

The paper surveys five GenAI model structures and relates their mechanisms, strengths, limitations, and construction applications. It emphasizes matching model choice to task while addressing construction’s limited data availability.

  • Five GenAI model types are reviewed: GANs, VAEs, autoregressive models, diffusion models, and flow-based models.Their generative mechanisms differ, so model selection depends on the application.
  • Generative Adversarial Networks: GANs use adversarial generator–discriminator training and can synthesize plausible architectural drawings and facade views.Reported applications include floorplans, mechanical/electrical/plumbing diagrams, sectional views, colored plans, style classification, and image restoration.
  • Variational Autoencoders: VAEs learn compressed latent representations through encoder–decoder components, but latent-space assumptions reduce flexibility for complex real-world data.
  • Autoregressive Models: Autoregressive models predict each next token from preceding context and have supported construction cost-index forecasting through temporal-dependency modeling.Their sequential sampling is slow for high-dimensional data and requires a fixed ordering.
  • Diffusion Models: Diffusion models generate synthetic outputs by repeatedly denoising samples after learning to reverse a scheduled noise-adding process.
  • Flow-based Models: Flow-based models transform noise through bijective functions toward a target distribution, using a reversible encoder but requiring a delicate design.Construction data scarcity motivates complementary approaches such as augmentation, synthetic generation, and transfer learning.

4. Opportunities of GenAI in Construction

The paper identifies broad GenAI opportunities across construction workflows, informed by literature, industry perspectives, and lifecycle applications. It proposes a conceptual framework that selects, fine-tunes, and deploys models for construction-specific use.

  • Prior applications: Prior LLM studies, including BIM-GPT, demonstrate opportunities to retrieve, summarize, and answer questions from complex construction information sources.BIM-GPT is described as addressing the engineering burden of automating information extraction from rich BIM models.
  • Industry perspectives: Construction professionals’ perspectives were analyzed through word-frequency, sentiment, and thematic synthesis to identify GenAI opportunities.The analysis combined word-cloud visualization, sentiment analysis, and synthesized opportunity themes.
  • Information workflows: GenAI can transform architecture, engineering, and construction information workflows by extracting insights from unstructured data and automating document review.Examples include analyzing building codes, identifying requirements, and producing project-specific reports.
  • Lifecycle applications: Potential applications span the building lifecycle, including cost estimation, progress and safety monitoring, quality control, scheduling, information retrieval, training, and maintenance.Table 4 organizes applications by beneficiaries and output modalities such as text, image, 3D, video, and task.
  • Lifecycle applications: GenAI applications include knowledge-management chatbots, work-order generation, predictive maintenance, energy reports, occupant assistance, BIM retrieval, natural-language model updates, and robot instruction.These examples associate construction tasks with text-to-text, text-to-image, text-to-3D, and text-to-task outputs.
  • Implementation framework: The conceptual implementation framework comprises selection, fine-tuning, and utilization, producing construction-domain models that users query through prompt engineering.Fine-tuning may include parameter adjustment, rewards, privacy constraints, and noise reduction.

5. Challenges of GenAI Implementation in Construction

GenAI implementation in construction faces challenges arising from the industry’s complex technical knowledge requirements and the limits of AI in physical, spatial, and contextual settings. Hallucination and inaccurate outputs further complicate reliable adoption.

  • Implementation challenges: The paper frames construction GenAI adoption as requiring careful attention to complex implementation challenges before applying the proposed framework.The challenges are presented as spanning areas including domain knowledge and reliability concerns.
  • 5.1. Domain knowledge: Construction requires extensive multidisciplinary domain knowledge, physical situational awareness, and spatial reasoning that challenge current GenAI capabilities.The relevant expertise spans structural, mechanical, electrical, plumbing, and project management disciplines.
  • 5.2. Hallucination: GenAI systems can produce convincing but false outputs because of inadequate or noisy training data, limited contextual understanding, and imposed constraints.Such outputs may appear aesthetically plausible while being inaccurate and expressed with unwarranted confidence.

5.3. Accuracy

Accuracy is a central challenge because construction decisions can be harmed by outputs based on superficial associations rather than engineering fundamentals. Poor generalization also limits whether models trained on historical data remain useful for new real-world cases.

  • 5.3. Accuracy: Inaccurate GenAI outputs can cause major failures because models may lack fundamental construction engineering knowledge and rely on superficial statistical associations.Construction-specific fine-tuning is identified as one technique for enhancing output validity.
  • 5.3. Accuracy: Poor generalization occurs when knowledge learned from in-sample historical data does not transfer effectively to new out-of-sample situations.A model may fit training data well yet remain unsuitable for real-world decision-making.

5.5. Model Updates and Interpretability

GenAI deployment in construction is constrained by rapidly outdated training data, costly model updating, limited interpretability, and substantial operating expenses.

  • 5.5. Model Updates and Interpretability: Construction GenAI models require regular retraining as materials, methods, regulations, and industry conditions change, but updating is costly and complex at scale.Suggested responses include modular architectures, synthetic-data simulations, transfer learning, and human-AI collaboration.
  • 5.5. Model Updates and Interpretability: Black-box decision processes hinder critical construction applications where users need explainable outputs and confidence in model reasoning.The paper identifies explainable AI as important for wider adoption.
  • 5.5. Model Updates and Interpretability: Training and operating generative AI models require substantial computing resources, time, energy, infrastructure, and ongoing usage costs.The paper notes that access fees and generated-content charges add to adoption costs.

5.7. Ethical Challenges

GenAI adoption in construction raises ethical and regulatory concerns involving privacy, bias, accountability, liability, and ownership of AI-generated content.

  • 5.7. Ethical Challenges: Construction GenAI systems may expose sensitive project and personal information, creating risks of confidentiality breaches and intellectual property violations.The paper recommends privacy safeguards and anonymization measures.
  • 5.7. Ethical Challenges: Implementation requires AI skills and prompt-engineering training, while overreliance on automation may reduce human expertise.The paper frames these issues alongside broader regulatory and legal complexities.
  • 5.7. Ethical Challenges: Unclear responsibility for AI failures and ownership of independently generated content create unresolved liability and intellectual-property questions.Potential responsibility may involve developers, construction companies, or safety managers.

5.9. What Challenges are Perceived by Construction Industry Practitioners?

Practitioners perceive GenAI adoption as early-stage and requiring stronger security, governance, awareness, business justification, and risk controls.

  • 5.9. What Challenges are Perceived by Construction Industry Practitioners?: 40% of 48 AEC professionals had never tried GenAI, while 33% used it sometimes, 19% often, and 8% all the time.The poll indicates that most surveyed professionals remained in the early stages of adoption.
  • 5.9. What Challenges are Perceived by Construction Industry Practitioners?: Construction organizations need proactive security and governance, strategic adoption, and stronger awareness to build convincing business cases for GenAI investments.The paper links these measures to responsible use and broader adoption.
  • 5.9. What Challenges are Perceived by Construction Industry Practitioners?: Risk mitigation should include data encryption, strict access controls, secure storage, intellectual-property guidelines, and contractual agreements.The paper also notes that traditional schedulers may hesitate to adopt newer solutions because of GenAI's novelty.

6. Recommendations and Future Directions

The paper recommends staged GenAI adoption through construction-specific fine-tuning, human oversight, business-impact evaluation, and targeted future research.

  • 6. Recommendations and Future Directions: The paper organizes future work around short-term and long-term adaptation strategies intended to facilitate effective GenAI integration in construction.Section 4.3 frames applications as a foundation for these recommendations and research directions.
  • 6. Recommendations and Future Directions: Initial integration should fine-tune powerful pretrained language models on construction-specific data such as design documents, codes, contracts, technical documents, and BIM data.Starting with modest datasets and clearly defined tasks can simplify prompt engineering.
  • 6. Recommendations and Future Directions: Human-in-the-loop review should validate GenAI output quality and accuracy while feedback improves system performance.The recommendation combines automated generation with human judgment.
  • 6. Recommendations and Future Directions: GenAI business impacts should be assessed through experiments that measure key performance indicators and pilot studies.The passage presents evaluation as part of the recommended adoption strategy.
  • 6. Recommendations and Future Directions: Future research should target document and BIM information extraction, feasible requirement-driven building design generation, and conversational assistants across the building lifecycle.These directions are presented as major research questions for improving productivity, time, cost, and stakeholder interaction.

7. Conclusion

The study synthesizes literature and practitioner insights to identify GenAI opportunities and adoption challenges in construction, then proposes a conceptual implementation framework. It also maps GenAI model types to construction tasks and identifies validation and real-world application as next steps.

  • The study combines a literature review with practitioner sentiment and thematic analysis to examine GenAI opportunities and adoption barriers in construction.
  • Text-to-text, text-to-image, text-to-3D/Video, and text-to-task models are mapped to potential applications spanning feasibility, design, procurement, construction, and operations.
  • The proposed conceptual framework is intended to guide researchers and practitioners implementing GenAI in construction.
  • Key implementation challenges include domain knowledge, hallucinations, accuracy, generalizability, interpretability, cost, ethics, and regulation.
  • Validation studies and development of real construction applications are identified as natural extensions of the research.

Funding

The research received no external funding.

  • The research received no external funding.
Loading 2310.04427v1…