Development of a hierarchical fuzzy assessment model for quantifying GenAI risks in sustainable construction projects

Abstract

Purpose:

The purpose of this study is to identify and evaluate the significance of emerging risks associated with the integration of Generative Artificial Intelligence (GenAI) into risk management (RM) within sustainable construction projects (SCPs).

Methodology:

The research followed a four-stage methodology to collect and analyse data, including: (1) a systematic literature review to identify a comprehensive list of GenAI-related risks in the context of SCPs; (2) the development of a multi-criteria risk assessment model to determine key evaluation criteria; (3) the distribution of a structured survey to 80 construction experts to assess the identified risks based on these criteria; and (4) the development of a fuzzy-based model to quantify and rank the significance of the risks.

Findings:

A total of 30 risk factors were identified and subsequently classified into five categories: input quality, technological adaptability, ethical and governance, information integrity, and financial-related risks. Furthermore, the fuzzy analysis revealed that the most significant risk factors were human error, data unavailability, insufficient training, data breaches, and lack of awareness.

Implications:

The study provides both theoretical and practical contributions by developing a novel risk assessment framework tailored to GenAI integration in sustainable construction. The fuzzy set theory approach enhances the accuracy of decision-making in high-uncertainty environments and aids project managers and policymakers in prioritising critical risks. The findings also offer actionable insights for developing mitigation strategies and fostering responsible GenAI implementation.

Originality/value:

This study makes a unique contribution by being among the first to systematically analyse the risks of GenAI integration in construction RM using fuzzy logic. It advances the understanding of AI-driven risks in the built environment and provides a replicable framework for future risk assessments. Moreover, it encourages further exploration of regional and contextual variations in GenAI risk perceptions and supports the broader digital transformation agenda in sustainable construction.

1 Introduction

Artificial Intelligence (AI) has become deeply embedded in project management, influencing both the discipline and the various stages of the project life cycle in multiple ways (Bento et al., 2022; Holzmann et al., 2022; Müller et al., 2024). Continuous advances in AI have facilitated the use of sophisticated computational tools and extensive datasets to uncover previously undetected patterns and establish precise relationships between input and output variables, thereby enhancing the quality and robustness of project-related decision-making (Lachhab et al., 2018; Ojiako et al., 2025). These developments have ushered in a new era of Generative Artificial Intelligence (GenAI), characterized by complex algorithmic architectures and models that exhibit human-like creativity, adaptability, and reasoning for insight generation and problem-solving (He et al., 2025). GenAI utilizes a combination of neural networks, natural language processing (NLP), reinforcement learning, generative adversarial networks (GANs), and transformer architectures to achieve these capabilities (Bengesi et al., 2024). As transformative technology, GenAI has been increasingly applied across various domains, notably within construction management, where it has demonstrated significant potential in advancing risk management practices (Chenya et al., 2022). A growing number of construction and project-based organizations now employ GenAI-driven tools to enhance various elements of their practices (Ghimire et al., 2023; Onatayo et al., 2024; Taiwo et al., 2024), including their risk management processes (Mohamed et al., 2025; Qiao et al., 2025). Consequently, project managers are better positioned to make informed, data-driven decisions, optimize risk mitigation strategies, and identify opportunities to improve project performance and outcomes (Regona et al., 2022; Chenya et al., 2022).

Despite these advancements, drawing from (Zhu et al., 2025), there appears to be a paucity in research on the notable risks associated with integrating GenAI into the risk management of Sustainable Construction Projects (SCPs). SCPs are defined as those designed, planned, and executed to minimize adverse environmental, social, and economic impacts while maximising long-term societal and ecological benefits (Kibert, 2016). On this basis, establishing a coherent scholarly consensus on the risks associated with integrating GenAI into the risk management of SCPs is of critical importance. The absence of such agreement brought several challenges. It can result in inconsistencies and a fragmented approach to developing appropriate risk frameworks, thereby impeding the formulation and adoption of universally accepted best practices. Moreover, it constrains the replication and validation of related empirical research, which in turn weakens the credibility and generalisability of the academic evidence base. Ultimately, these uncertainties are likely to discourage industry practitioners from adopting GenAI-driven tools for risk management, potentially undermining the very sustainability objectives that such technologies are intended to support.

Noting these considerations, this study investigates the key risks associated with integrating GenAI into risk management for SCPs and quantitatively assesses their significance using a fuzzy-based multi-criteria model. To this end, this study seeks to: (1) identify and categorize the key risks that construction organizations may encounter when implementing GenAI in risk management for SCPs; and (2) develop a multi-criteria risk analysis model, based on fuzzy set theory, to quantitatively assess the significance of the identified risks.

2 Literature review

2.1 Sustainable construction

Sustainable construction involves the development, assembly, and creation of structures or infrastructure using materials, labour, equipment, and technology, with a strong focus on minimising environmental impact and conserving resources while meeting project objectives. In addition to these considerations, it aims to deliver economic and social benefits. A key feature of sustainable construction is the integration of sustainability principles throughout all stages of the construction lifecycle, from initial design and planning to operation and eventual decommissioning (Zuo and Zhao, 2014). According to Giesekam et al. (2014) and Azhar (2011), core aspects of sustainable construction include the use of eco-friendly materials, energy-efficient systems such as solar panels and smart technologies, water conservation strategies, waste reduction, green building design, promotion of occupant health and wellbeing, reduction of carbon footprints, and sustainable site management that preserves biodiversity and minimises land disturbance. Risk management in sustainable construction differs significantly from traditional construction due to its broader scope, higher complexity, and additional objectives. Whereas traditional construction primarily addresses operational, financial, and technical risks, such as cost overruns, schedule delays, and design errors, sustainable construction places greater emphasis on environmental, social, and governance (ESG) factors. This includes risks related to regulatory compliance, energy consumption, community impact, and long-term environmental sustainability. Projects in this context may face challenges such as supply chain disruptions for eco-friendly materials, technological uncertainties in green building systems, labour skill shortages, and unforeseen environmental impacts. Consequently, managing risks in sustainable construction is inherently more complex. Mitigation strategies extend beyond conventional approaches to focus on enhancing environmental performance, promoting social benefits, and ensuring the long-term resilience and sustainability of the project.

2.2 GenAI in risk management for construction

A growing body of research has examined the capabilities, benefits, and comparative performance of GenAI across various dimensions of construction management. However, most of these studies have primarily emphasised the advantages of GenAI, particularly its ability to improve predictive accuracy, automate data-heavy processes, and support real-time decision-making, while giving comparatively little attention to its associated risks and vulnerabilities. For instance, Pan and Zhang (2021) investigated the use of GenAI to automate project documentation and forecast potential delays, highlighting its effectiveness in managing unstructured data and enhancing project responsiveness. Similarly, Prebanic and Vukomanovic (2021) and Nyqvist et al. (2024) explored GenAI-powered chatbots as communication tools for project stakeholders, reporting notable improvements in stakeholder engagement and information exchange (Al-Mhdawi et al., 2023). compared GenAI-based risk identification systems with traditional expert-driven models and found that GenAI exhibited superior capability in detecting early-stage risks and uncertainties (Prieto et al., 2023). demonstrated that ChatGPT can generate preliminary project schedules, reducing the need for extensive manual input from project managers and streamlining the scheduling process. Likewise, Omotayo et al. (2024) observed that integrating GenAI with Building Information Modelling (BIM) enhances digital cost management and promotes continuous cost optimisation within construction firms.

Research focusing specifically on the application of GenAI for risk management in traditional (non-sustainability-oriented) construction projects remains limited, with notable contributions from (Mohamed et al., 2025). In their first study, Mohamed et al. (2025) conducted a systematic literature review of publications between 2014 and 2024 using the Scopus database and identified five primary categories of risks in construction projects: “financial”, “technological adaptability”, “information integrity”, “input quality”, “ethical and governance risks”. In a subsequent study (Mohamed et al., 2026) expanded their review by drawing on a broader range of databases (i.e., ASCE Library, Emerald Insight, Google Scholar, IEEE Xplore, ScienceDirect, Scopus, Springer, and Taylor and Francis), covering the same time frame. This extended analysis led to the identification of nine families of risks relevant to the application of GenAI in risk management: “social”, “security”, “data”, “integration”, “performance”, “legal”, “resource”, “efficiency”, and “cross-domain (knowledge area) impact risks”.

Despite limited GenAI risk frameworks applicable to construction, the reviewed studies highlight GenAI’s growing influence in risk management, emphasising its strengths in automation, prediction, and decision support.

2.3 GenAI in risk management for sustainable construction

GenAI is increasingly being recognised as a significantly important tool for managing risks in SCPs, offering innovative approaches to identifying, evaluating, and mitigating threats that could affect environmental outcomes and long-term project sustainability (Zhu et al., 2025). Unlike conventional AI, which mainly depends on historical data for predictive modelling, GenAI can learn from complex, multidimensional datasets to generate new scenarios, insights, and solutions (Epstein et al., 2023; Feuerriegel et al., 2024; Fui-Hoon Nah et al., 2023). This capability is particularly valuable in sustainable construction, where risks are shaped by environmental, social, and economic factors.

GenAI can be applied across multiple areas of sustainable construction risk management (Zhu et al., 2025). For example, it can generate diverse scenarios and simulations, revealing risks that may not be immediately apparent to project teams. It is also useful for modelling extreme conditions, such as the effects of severe weather on project schedules or resource availability. Additionally, GenAI can optimise the allocation of resources, helping reduce environmental impacts like carbon emissions and energy consumption. It can support predictive risk analytics by identifying early warning signs of potential issues, and it can improve communication with stakeholders by producing clear, data-driven visualisations and risk assessments, enhancing transparency and accountability.

Despite these advantages, the integration of GenAI into sustainable construction risk management encounters obstacles, such as worries regarding data reliability, algorithmic bias, and a lack of qualified professionals to manage AI-driven processes. Nevertheless, when applied judiciously and with appropriate oversight, GenAI could transform risk management, facilitating a more proactive, precise, and efficient approach to identifying, assessing, and mitigating intricate environmental, social, and governance-related risks.

2.4 Research gap

There are expected opportunities for integrating GenAI into the risk management processes of SCPs (Zhu et al., 2025). However, efforts to do this are likely to be accompanied by a spectrum of inherent risks. These may include, among others, (i) data input quality risks, (ii) technological adaptability risks, (iii) ethical and governance risks, (iv) information integrity risks, and (v) financial risks. For instance, data input quality risks arise when GenAI models are trained on inaccurate, biased, or incomplete datasets, potentially leading to flawed or unreliable risk evaluations (Afzal et al., 2021). Technological adaptability risks concern the challenges of embedding GenAI tools within existing risk management frameworks, particularly issues of system compatibility, interoperability, and the demand for costly infrastructure upgrades (An et al., 2021). Ethical and governance risks emanate from ongoing debates about transparency, accountability, data security, and compliance with relevant regulatory and legal requirements (Aladag, 2023). Information integrity risks involve the potential for GenAI to produce misleading, inconsistent, or fabricated results, which could compromise the credibility and dependability of decision-making processes. Lastly, financial risks are associated with the substantial costs of implementation and long-term maintenance, as well as the potential inefficiencies that may arise if GenAI technologies fail to deliver the anticipated performance benefits (Regona et al., 2022).

Despite these developments in the literature, a significant gap persists in the literature concerning the risks associated with integrating GenAI into the risk management processes of SCPs. Although a few recent studies, such as (Mohamed et al., 2026), have begun to draw attention to this relatively underexplored dimension, such works remain the exception rather than the norm. The emerging research highlights a variety of concerns, including dependence on data quality, ethical and legal ambiguities, and the technical challenges of integrating GenAI with existing risk management systems. Nevertheless, few studies have attempted to systematically categorise these risks into coherent thematic groups or to assess their relative importance within the specific context of sustainable construction. This gap underscores the pressing need for a comprehensive, empirically informed investigation into GenAI-related risk management, one that not only identifies and characterises the nature of these risks but also quantifies their impact to enable more informed and evidence-based decision-making. Therefore, this study advanced the literature and moved beyond general discussions of GenAI adoption to provide a risk-specific framework for RM within SCPs through identifying a structured GenAI risk taxonomy for SCPs and combining this with a fuzzy-based model that quantifies risk significance under conditions of uncertainty.

3 Methodology

The research adopted a four-stage, multi-method approach to systematically identify, assess, and quantify the risks associated with integrating GenAI into the risk management of SCPs, as illustrated in Figure 1. Details of each of the adopted methodological steps are presented in the following subsections.

FIGURE 1

3.1 Stage one: identification of the key risks

This stage adopts a structured, three-step methodology for literature collection and analysis, designed to thoroughly examine existing studies and identify the key risks associated with integrating GenAI into risk management in SCPs. The first step involves identifying relevant databases and journals to establish a robust foundation for the literature search. The second step entails strategically selecting articles using targeted keywords to ensure the inclusion of the most pertinent studies. Lastly, the third step focuses on conducting systematic content analysis to extract valuable insights. This approach is guided by frameworks outlined in several risk management studies, particularly those by (Al-Mhdawi et al., 2025; Al-Mhdawi et al., 2024b).

3.1.1 Database and journal identification

In this research, the Scopus database was selected due to its extensive coverage of relevant research disciplines and its established use in similar literature-based studies within the field of construction management. The target journals for this study were chosen based on the following criteria: (1) they must be published in English, (2) they must have a minimum impact factor of 1.0, and (3) they must be ranked in the top quartile of the Scopus database, reflecting their substantial influence on advancements in construction management research.

3.1.2 Keyword identification and article selection

In this step, a comprehensive search was performed using the title, abstract, and keyword (T/A/K) fields within the Scopus search engine. The search keywords included “GenAI risks”, “GenAI”, “GenAI in RM”, and “GenAI in sustainable construction project management”. Articles containing these terms in the title, abstract, or keywords were considered to meet the preliminary inclusion criteria for further analysis. These keywords were carefully chosen to encompass a wide range of studies addressing the risks and applications of GenAI in risk management for SCPs and related fields. The search results were then refined by removing duplicate entries, irrelevant studies, and papers that did not provide a substantive focus on the intersection of GenAI and risk management in SCPs.

3.1.3 Content analysis

(Drisko and Maschi, 2016) outline three approaches to content analysis: conventional, directed, and summative. This study adopted the conventional approach, a data-driven, open-ended method that allows categories to emerge organically without the constraints of predefined frameworks (Blomkvist, 2015). This approach, suitable for both qualitative and quantitative analysis, is particularly well-suited for investigating the emerging integration of GenAI into risk management for SCPs, as it facilitates the extraction of detailed themes directly from the data (Kibiswa, 2019). Unlike directed analysis, it avoids limitations imposed by existing theories, fostering a rich and context-specific understanding (Krippendorff, 2018). Using this method, the study systematically refined an initial pool of 471 papers to 55, identifying key risks and categories associated with integrating GenAI into risk management for SCPs. Accordingly, the five categories emerged through conventional content analysis of the reviewed studies by grouping risks that shared a common source and mechanism of influence on GenAI-enabled risk management. Risks were clustered according to whether they primarily originated from data inputs, technological adaptation, governance and ethics, information integrity, or financial implementation conditions. This structure was considered appropriate because it reflects the main domains through which GenAI-related vulnerabilities materialise in SCPs.

3.2 Stage two: development of a multi-criteria risk analysis model

Failure Mode and Effects Analysis (FMEA) is a widely applied risk assessment technique that evaluates the significance of risks by analysing three key factors: the probability level (PL) of occurrence, the impact level (IL) on project objectives, and the detectability level (DL) of the risk (Kritsky et al., 2018). This approach is extensively used across various engineering and construction fields (Alvand et al., 2023; Zhang et al., 2024). To develop the proposed multi-criteria risk analysis model incorporating IL, PL and DL, a structured Delphi technique was undertaken among the authors. In this study, the Delphi stage was designed as an internal expert-based Delphi process involving the author team. This approach was adopted to iteratively refine the model structure, variable grouping, and rule logic based on the authors’ combined expertise in construction risk management and fuzzy-based modelling, rather than establishing broad external consensus. This methodological positioning is consistent with construction risk research emphasising that the selection and application of risk techniques should reflect the context and purpose of the study (Kıral, 2025). This process drew on the team’s combined expertise in construction risk management and their previous work on fuzzy-based risk assessment frameworks (Al-Mhdawi et al., 2024a; Mahammedi et al., 2020). Two Delphi rounds were implemented to allow iterative refinement of the model structure, ensuring conceptual clarity, logical coherence and alignment with domain-specific risk assessment practices.

Both rounds followed established Delphi procedures featuring anonymous individual responses and controlled feedback. In Round 1, each author independently evaluated the relevance and clarity of the proposed input variables, the suitability of their grouping under IL, PL and DL and the adequacy of the initial fuzzy rule structure, using a five-point rating scale. Medians and interquartile ranges (IQRs) were calculated for all items, and anonymised qualitative comments were compiled to highlight areas needing clarification or adjustment. In Round 2, authors received a structured summary of Round 1 results, including medians, IQRs and aggregated comments, and were invited to revise their ratings considering the group feedback. Consensus thresholds were predefined as median scores of at least 4 and IQRs of 1 or lower at the item level, consistent with accepted Delphi practice. Kendall’s W was monitored to assess the overall level of agreement across rounds (McMillan et al., 2016). The process concluded after Round 2 once stability in ratings was observed and all items met the consensus criteria.

The impact dimension was categorised into two main groups: project-related impacts, which include the impact on project cost (ipc), schedule (ips), and quality (ipq); and organizational impacts, which include the impact on decision-making (idm) and organisational reputation (ior). A similar structuring was applied to the probability criterion, which was assessed based on the availability of expertise (ae), availability of assets (aa), and the organisation’s technological maturity (otm). For detectability, the relevant factors included ea, the presence of continuous monitoring mechanisms (acm), and the level of exposure, which was further informed by the risk duration (rd) and frequency of occurrence (fo). This methodological framework aligns with validated practices from previous fuzzy risk assessment studies (e.g., Al-Mhdawi and O’Connor, 2026; Al-Mhdawi et al., 2024c). The model ensures that the developed Risk–Probability, Impact, and Detectability (R-PID) model offers a context-specific and reliable tool for evaluating GenAI-related risks in SCPs. The overall fuzzy risk number (F-RN) is derived as shown in Equation 1:

3.3 Stage three: rating risks for GenAI integration

In this stage, the authors employed a questionnaire survey instrument for data collection, as it enables respondents to provide information in a structured and standardised format, facilitating systematic data analysis and comparison while also offering a level of anonymity and confidentiality, which can encourage more honest and candid responses (Chartres et al., 2019).

3.3.1 Survey development

The authors developed a questionnaire survey to evaluate the significance of risks associated with integrating GenAI into risk management for SCPs, as identified through the SLR. To quantify risk significance, respondents assessed each risk based on four key criteria: (1) the probability of occurrence, which measured the likelihood of the risk arising based on factors such as ea, aa, and otm; (2) the impact on project and organization, including ipc, ips, ipq, idm, and ior; (3) the detectability of the risks, based on ea, acm and the exposure level which could be measured through evaluating risk fo and rd, which considered the duration and frequency of the risk during the project lifecycle. The survey consisted of two sections. The first section collected information from respondents, including their roles in construction field, years of experience, educational background and level of experience with GenAI in project management. The second section required respondents to assign relative weights to the identified risks. To facilitate this assessment, a five-point Likert scale was adopted, ranging from 1 (Very Low - V.L) to 5 (Very High - V.H). This scale enabled respondents to systematically evaluate the significance of each risk based on predefined assessment criteria, ensuring consistency and comparability in the collected data. Experts were recruited through purposive sampling through professional networks in construction project management and risk. Eligibility requires a current or recent role in construction projects or risk management and at least 3 years of relevant experience. Prior exposure to GenAI in construction was recorded but not required. At enrolment, participants completed screening questions on role, experience, sector, prior GenAI use, and self-rated expertise level as beginner, intermediate, or expert. We applied quality checks, including one attention item and internal consistency checks, and removed incomplete or contradictory responses. The author team conducted two Delphi rounds to refine the instrument and assess the face validity of responses. We did not collect document-based credentials; this is noted as a limitation.

3.3.2 Survey pilot testing

Pilot testing is a critical preliminary step in survey-based research used to evaluate the reliability, validity, and overall quality of a questionnaire before deploying it on a larger scale. It ensures that survey items are clearly understood, appropriately worded, and capable of capturing the intended data, thereby improving the instrument’s overall robustness and response accuracy (Bryman, 2016). Conducting a pilot also helps identify potential ambiguities that may affect data collection and respondent experience. In this study, a pilot test was conducted with 10 construction experts from the United Kingdom to assess the clarity and effectiveness of the developed questionnaire. Table 1 presents participants’ years of experience, industry roles, and educational backgrounds. Based on the feedback received, minor revisions were made to enhance the precision and relevance of the survey. These included adding a new question to assess respondents’ experience with using GenAI in project management and renaming certain criteria to improve clarity and ease of understanding during the risk evaluation process.

3.3.3 Survey administration

The final version of the survey was distributed to 136 construction management experts in the United Kingdom, selected based on two criteria: (1) holding a professional role in the construction industry or academia related to construction education and (2) demonstrating proficiency in applying GenAI in construction projects. The respondent group included project managers, consultants, academics, and other project management professionals. Of the 136 experts surveyed, 101 responded, with a response rate of 74.3%. However, 21 responses were incomplete, leaving 80 valid responses for analysis. A more detailed discussion of the respondents’ profiles is provided in the results and analysis section.

3.4 Stage four: development of a fuzzy-based risk quantification model

Fuzzy Set Theory (FST), introduced by (Zadeh, 1965), extends classical set theory by providing a framework for addressing uncertainty and imprecision in decision-making (Gholamizadeh et al., 2022; Lauron and Abellana, 2024). Unlike traditional binary sets, FST allows for partial membership, making it particularly effective for representing and analysing linguistic variables such as “very low”, “low”, “moderate”, “high”, and “very high” (Akram et al., 2024). This flexibility enables the handling of vagueness and subjectivity inherent in human judgments (Chrysostom and Dwivedi, 2016; Mahmood et al., 2024). A key advantage of FST is its ability to formalise and quantify human knowledge by converting qualitative linguistic assessments into fuzzy numerical values, thus managing imprecise or incomplete information and reconciling conflicting expert opinions (Adak et al., 2024; Al-Mhdawi et al., 2023). This makes it particularly valuable in scenarios where precise numerical data are unavailable or insufficient (Al-Mhdawi et al., 2024c). Therefore, Fuzzy logic was selected because the present study relies on subjective expert judgements expressed through linguistic scales rather than large probabilistic datasets. Unlike Bayesian approaches, it does not require stable conditional probability distributions, and unlike AHP, it can better accommodate ambiguity and non-linear interactions among risk dimensions.

Given these strengths, FST was employed in this study to assess the significance of risks associated with integrating GenAI into risk management of SCPs. Its capacity to accommodate ambiguity and incorporate expert judgment provided a systematic approach to quantifying risks, even in cases of uncertain or subjectively defined data.

Fuzzy logic was selected because the risk information in this study is expressed through linguistic expert judgements, which require a modelling approach capable of handling subjectivity without imposing probabilistic assumptions that the available data cannot support. Bayesian networks and similar probabilistic methods depend on conditional probability distributions that require large, high-quality datasets not available for emerging GenAI risks. AHP and other weighting-based MCDM techniques provide ordinal rankings but cannot represent the non-linear interactions among IL, PL and DL that form the core of the R-PID structure. In contrast, Mamdani fuzzy inference allows experts’ qualitative inputs to be mapped into interpretable IF–THEN rules while preserving ambiguity through membership functions. This makes fuzzy inference well aligned with early-stage risk domains where empirical data are limited and expert reasoning plays a central role, while also offering transparency in how risk components interact—an essential requirement for explaining GenAI-related risk pathways.

To operationalise this approach, a fuzzy-based model was developed for quantitatively assessing the significance of identified risks. The model utilised multiple input variables and a single output and was implemented using MATLAB R2024b (version 24.2). In specific, the model utilised six fuzzy controllers to assess risk significance. The first controller evaluated the IL of each risk based on its components: ipc, ips, ipq, idm, and ior. The second controller measured the probability of risk occurrence using ea, aa, and otm. While the third controller assessed the DL of each risk, incorporating aa, acm, and the output of the fourth controller, which determined the exposure level based on rd and fo. Finally, the fifth and main controller computed the overall risk significance by integrating the three primary assessment criteria (i.e., impact, probability, and detectability). Additionally, in each controller, the risk assessment criteria and their respective components were transformed into fuzzy sets using predefined membership functions. These fuzzy inputs were then processed within a fuzzy inference engine to estimate the level of each criterion and determine the overall significance of identified risks.

The architecture of the proposed GenAI risk assessment model is structured around three essential processes: fuzzification, fuzzy inference, and defuzzification. In the fuzzification process, both input and output variables were represented using triangular membership functions. This choice was guided by their simplicity, computational efficiency and suitability for modelling expert linguistic judgements, which aligns with their widespread adoption in fuzzy risk-assessment applications (Gerla, 2013; Yager and Zadeh, 2012). Triangular functions were selected because they require fewer parameters than trapezoidal or Gaussian alternatives, which reduces the cognitive burden on experts during elicitation and avoids imposing distributional assumptions that the available data cannot support (Liang et al., 2013).

Gaussian functions, for example, assume smooth probabilistic transitions that are not appropriate for the discrete linguistic scales used in this study, while trapezoidal functions add additional degrees of freedom that offer limited interpretive benefit for early-stage, expert-driven risk modelling. Triangular membership functions therefore provide a transparent and parsimonious representation of subjective assessments and facilitate efficient computation across all fuzzy controllers.

For the inference stage, Mamdani’s Fuzzy Inference System (MFIS) was employed due to its intuitive reasoning capabilities, ability to handle linguistic variables, and strong prevalence in engineering and decision-making literature (Mamdani and Assilian, 1975).

Finally, the defuzzification process was carried out using the centroid of area method, which is commonly preferred in fuzzy systems for its accuracy in aggregating fuzzy sets into a single representative crisp output. This method is particularly useful in modelling expert judgments, as it balances multiple overlapping membership functions to produce a meaningful outcome (Kaynak et al., 2012). Collectively, these techniques form a cohesive framework suitable for modelling the uncertainty and subjectivity inherent in assessing the risks of integrating Generative AI into construction risk management.

Scalability and extensibility were also considered in the design of the fuzzy architecture. The proposed model scales linearly with the number of risks because each risk is processed once through a fixed sequence of six Mamdani controllers, each with a constant computational structure. As construction projects become more complex or involve a broader set of risks, the modular structure enables straightforward extension. Additional factors can be incorporated in three ways: (i) integrating new indicators into an existing controller when conceptual alignment exists, (ii) combining related inputs into a composite variable to prevent exponential growth of the rule base, or (iii) introducing supplementary controllers that feed into the final RL controller without altering the core R-PID structure. This modularity ensures that the framework remains computationally tractable and adaptable for larger or more diverse datasets typically encountered in complex SCPs.

To this end, a five-point Likert scale (ranging from Very Low (V.L) to Very High (V.H)) was used to define the inputs and outputs of the assessment model. Accordingly, five membership functions were established for the assessment components, criteria, and output variables.

When input values lie within overlapping linguistic regions, several rules may activate at the same time, including rules that assign different consequents. Each rule’s activation level is calculated using the minimum operator for the AND condition. During implication, the consequent membership function is clipped at this activation level using the minimum operator. All activated consequents for the output variable are then aggregated using the maximum operator, which retains the influence of each rule instead of overriding conflicting values. The final crisp output is generated through centroid defuzzification, which resolves contradictory activations by producing a value that reflects the weighted balance of all fired rules. This configuration follows the standard Mamdani structure implemented in MATLAB (AND = min, OR = max, implication = min, aggregation = max, defuzzification = centroid).

To ensure reproducibility and to clarify the implementation environment, the fuzzy controllers were developed and executed using MATLAB R2024b with the Fuzzy Logic Toolbox (version 2.10). All membership functions, rule bases, and inference settings were created within the toolbox using the standard Mamdani configuration (AND = min, OR = max, implication = min clipping, aggregation = max, defuzzification = centroid). The six controllers were implemented as separate. fis files linked sequentially so that outputs such as IL, PL, DL, and exposure were passed into subsequent controllers as crisp inputs. The computational performance of the fuzzy inference system was also reviewed to support practical scalability. The proposed model is computationally lightweight (e.g., required approximately 120 min for execution) because each Mamdani controller performs closed-form membership evaluations and min/max operations across a fixed rule base with no iterative optimisation. The computational cost per risk item is linear in the number of controllers and constant per controller, involving at most 5n rule evaluations and a single centroid defuzzification step. In practice, this makes inference effectively instantaneous, even for datasets far larger than those used in this study, indicating that scalability does not represent a barrier to industrial adoption.

To complement the methodological description, the computational performance of the fuzzy inference system was also examined. Executing all six Mamdani controllers for the full dataset of 30 risks required approximately 75 min on a standard workstation (Intel i7 processor, 16 GB RAM) when running the full pipeline including fuzzification, inference, defuzzification, entropy weighting and result recording. The inference mechanism itself remains computationally lightweight because each controller performs closed-form membership evaluations and min/max operations over a fixed rule base without iterative optimisation. The computational cost per risk item is linear with respect to the number of controllers and constant within each controller, requiring at most rule evaluations and a single centroid calculation. This structure makes model execution effectively instantaneous when scaled to larger datasets, supporting industrial applicability where rapid or repeated evaluations may be required.

Since the Fuzzy Logic Toolbox enforces fixed operator semantics and inference procedures, the full pipeline can be reproduced by loading the provided membership function definitions and rule tables into MATLAB and executing the evalfis () function for each controller. The controller architecture therefore follows the toolbox’s native execution structure, making additional pseudocode unnecessary, as it would reproduce MATLAB’s standard pipeline without introducing study-specific implementation choices.

Table 2 presents a worked example of the fuzzification and inference process for R01 using the PL controller. It provides a concise illustration of how triangular fuzzy numbers (TFNs) for each linguistic term (VL, L, M, H, VH) are defined using the a–b–c parameters on a 1 to 5 universe, and how the crisp inputs for ea, aa, and otm for R01 are fuzzified, the relevant IF–THEN rule is activated, and the final defuzzified PL value is obtained. The same TFNs and procedures apply to all variables and controllers in the model.

The fuzzy IF-THEN conditional statements were developed by adapting rule structures from previous fuzzy risk assessment models in construction (Al-Mhdawi et al., 2024a; Al-Mhdawi et al., 2024bAl-Mhdawi and O’Connor, 2026). These studies constructed fuzzy rule bases through systematic analysis of risk factors, linguistic categorization, and domain knowledge. Similarly, this study formulated fuzzy rules by aligning with validated methodologies used for cost overrun risk assessment, oil and gas construction risk modelling, and emerging risk evaluation in construction, ensuring the model reflects best practices in fuzzy-based risk assessment. Consistent with this literature, each controller employs five linguistic terms (Very Low, Low, Medium, High, Very High) with triangular membership functions under a Mamdani inference scheme, and experts provided ratings on these five levels for every input variable in each controller. A complete rule grid was generated per controller to ensure full input–output coverage.

A total of 550 rules were developed for the model, including 125 for impact on project insights, 25 for impact on the organisation, 125 for probability measurement, 25 for exposure level assessment, 125 for detectability, and 125 for overall risk significance. These counts arise from 5n combinations of five linguistic terms across the n inputs of each controller (for instance, 5³ = 125, 5² = 25). To support rule-base transparency, a small set of representative IF–THEN rules from each controller is provided in Table 3.

While the fuzzy rule-base defines the structural relationships among variables, variability in expert judgments introduces additional uncertainty that must be measured to ensure robustness. Therefore, uncertainty quantification was incorporated using a fuzzy-entropy approach. This method captures the degree of ambiguity in expert linguistic evaluations by analysing the spread of the aggregated membership functions associated with each linguistic term. Fuzzy entropy was computed using the De Luca and Termini formulation, where entropy represents the aggregated membership grade for linguistic terms (Very Low, Low, Medium, High, Very High) for each risk. Low entropy values indicate a sharply concentrated membership distribution and strong expert agreement, while high values reflect greater ambiguity or diversity in the assessments. Entropy values were calculated for all risks across the four evaluation criteria: probability, impact, detectability and exposure. Risks such as Data Unavailability (R01) and Lack of Awareness (R02) exhibited higher entropy, indicating broader variation in linguistic assessments. In contrast, Inconsistent Connectivity (R08) showed lower entropy, suggesting stronger consensus among experts. To ensure that high-uncertainty items did not disproportionately influence the aggregated scores, an entropy-based reliability weight was incorporated. This weighting preserves the original fuzzy-inference structure while introducing a transparent uncertainty quantification mechanism. The approach aligns with established fuzzy-modelling practices and is consistent with possible future extensions using interval type-2 fuzzy sets or possibility-distribution-based uncertainty analyses.

The use of a rule-based fuzzy inference system in this study provides a structured representation of how expert evaluations interact across multiple criteria through an interpretable set of IF–THEN relationships. This approach differs from ranking-oriented fuzzy multi-criteria decision-making methods such as fuzzy AHP, fuzzy TOPSIS and fuzzy CoCoSo, which rely on predefined weights and generate direct ordinal rankings. The present study adopts a rule-based fuzzy framework because its objective is to capture complex causal interactions among probability, impact, detectability and exposure rather than to produce a purely rank-based prioritisation. Although alternative fuzzy MCDM techniques can offer useful comparative perspectives, their integration would require recalibrating weights, consistency checks and parallel modelling pipelines. These approaches therefore represent promising future extensions that can complement, rather than replace, the causal reasoning provided by the current fuzzy-inference model.

A sensitivity analysis was then conducted to evaluate the model’s robustness and identify key risk drivers, following the approach proposed by (Rathore et al., 2021). Key input variables were systematically varied by ±10%, and the corresponding changes in F-RN were monitored. The objective was to determine whether small variations in inputs would cause significant shifts in model outputs or rankings, thereby validating the model’s stability. This approach aligned with the sensitivity analysis standards adopted in previous studies (Jain et al., 2018) and ensured that the model provided reliable insights for GenAI in risk management for SCPs.

4 Results

4.1 Risks identification and classification

The SLR identified 30 key risks associated with integrating GenAI into risk management for SCPs. These risks were categorised into five main groups, namely: Input quality risks, technological adaptability risks, ethical and governance risks, information integrity risks, and financial risks, as outlined in Figure 2, along with their respective sources. The identification methods varied across studies, including GenAI model training and testing, case studies, interviews, questionnaire surveys, and focus group sessions (Mohamed et al., 2025). In addition, research suggests that employing multiple methodologies to identify risks in construction projects is generally more effective than relying on a single approach, as it enhances the depth and reliability of findings (Sharma and Gupta, 2019). However, using a single method provides advantages such as simplicity, consistency, efficiency, and a more focused approach, facilitating detailed insights and improving replicability (Runeson et al., 2012). Despite these benefits, a single-method approach may introduce bias and the risk of overlooking critical factors, potentially limiting the comprehensiveness of risk identification. Therefore, integrating multiple identification methods is essential to ensure a robust and holistic assessment.

FIGURE 2

4.2 Profile of survey respondents

The survey respondents represented a range of roles within the construction industry, with the majority being Project Managers (62%), followed by other project management roles (23%), Academics (10%), and Consultants (5%). This distribution indicates a significant bias towards project management professionals, suggesting that most respondents were directly involved in supervising and managing construction projects. However, the relatively small proportion of consultants and academics may limit the diversity of perspectives, particularly in terms of expert advice and theoretical insights. In terms of professional experience, the respondents demonstrated a range of project management backgrounds. Most had between 1 and 5 years of experience (37%), followed by those with 6–15 years (32%) and 16–25 years (24%), whilst only a small portion reported having more than 25 years (6%). This indicates that the sample is predominantly composed of early to mid-career professionals, reflecting the views of those who are actively engaged in contemporary project management practices. The range of experience levels helps to provide a balanced understanding of challenges across different stages of professional development.

Educational qualifications among respondents were generally high, with 51% holding bachelor’s degrees, 42% holding master’s degrees, and 6% having completed doctorate degrees. This suggests that most participants were well educated, with a significant proportion possessing postgraduate qualifications. The strong educational background among the respondents enhances the reliability of the insights gathered, particularly in discussions related to advanced construction and risk management practices. Regarding familiarity with GenAI in construction management, the survey revealed varying levels of experience. Most participants identified themselves as having Intermediate experience (51%), followed by Beginner level (41%), and only a small group being classified as Experienced (9%). This distribution shows that whilst GenAI is gaining traction within the field, it remains relatively new, with most professionals still at the early or developing stages of adoption. These findings highlight opportunities for further training and capacity building to enhance the effective integration of GenAI into construction management practices. Figures 3–6 provide detailed illustrations of the respondents’ profiles based on their roles, years of experience, educational qualifications, and experience levels with GenAI in construction management.

FIGURE 3

FIGURE 4

FIGURE 5

FIGURE 6

4.3 Architecture of the developed GenAI risk assessment model and its outputs

As described in stage three of the research methodology, the proposed risk assessment model was designed using six fuzzy controllers via MATLAB R2024b, with each controller dedicated to evaluating a specific risk dimension. The first controller assessed the impact on project insights, using three input variables: ipc, ips, and ipq. The second controller evaluated the impact on the organisation, incorporating idm and ior; the summation of the outputs of first and second controller represented the overall IL. Moreover, the third controller measured the probability of risk occurrence, with inputs ae, aa, and otm, producing the PL as the output. The fourth controller assessed the DL, utilising ae, acm, and the output from the fifth controller, which measured exposure level based on fo and rd.

Finally, the sixth controller integrated the outputs of the IL, PL, and DL, generating the overall risk significance level for integrating GenAI into risk management for SCPs. This structured approach ensured a comprehensive and systematic evaluation of risk factors, enabling a more robust and data-driven assessment of potential risks.

The fuzzy controllers were designed using the IF–THEN rules presented in Stage Four of the methodology. The model is organised into six linked controllers (refer to Figure 7). The model is organised into six linked controllers. The first two controllers estimate project-level and organisation-level impacts, which are combined into total impact level. The third controller estimates probability level, while the fourth estimates exposure level from risk duration and frequency of occurrence. The fifth controller uses exposure and monitoring-related inputs to estimate detectability level. The sixth controller then combines total impact, probability, and detectability to generate the final risk severity level. This workflow is intended to clarify how the intermediate outputs are sequentially propagated through the model. To visualise the relationships between fuzzy controllers’ input and output variables, three-dimensional mappings were generated using the Fuzzy Logic Surface Viewer. These graphical representations illustrate how the output variables vary in response to changes in the input variables, enhancing interpretability. Furthermore, Figures 8–11 depict the dependencies for each controller, where each surface plot includes two input variables and one output variable, specifically, for impact, probability, detectability, exposure, and overall significance level. For instance, in the surface viewer, the detectability input was held constant at its default value to visualise the joint effect of total impact and probability on risk severity. Figure 11 illustrates the resulting fuzzy risk severity surface, showing that risk severity generally increases as both inputs rise. The variation in colour intensity reflects the magnitude of risk significance, with stronger gradients indicating higher severity levels. The non-linear shape of the surface reflects the influence of the fuzzy rule base and overlapping membership functions, providing a clear visual representation of the interaction between these key risk factors.

FIGURE 7

FIGURE 8

FIGURE 9

FIGURE 10

FIGURE 11

For the results presented in Figures 12, 13 illustrate the distribution of fuzzy occurrence probabilities across the five risk significance levels. The scatter plot (Figure 12) shows that the High-risk significance level (Level 4) is the most frequently activated classification, as indicated by its relatively higher and more consistent probability values across the sampling runs. This pattern suggests that the fuzzy inference system predominantly interprets the evaluated GenAI-related risks as having a relatively elevated level of significance within the construction context. The Moderate (Level 3) and Low (Level 2) levels are also assigned with noticeable frequency, indicating their role as intermediate classifications that capture transitional risk conditions. These levels reflect situations where the risk is neither negligible nor critically severe, highlighting the system’s ability to represent gradual variations in risk significance. In contrast, the Very Low (Level 1) and Very High (Level 5) levels appear less dominant in the scatter distribution, suggesting that extreme classifications are assigned only under more specific input conditions. This indicates a balanced behaviour of the fuzzy model, where boundary levels are not excessively triggered. The heatmap (Figure 13) further reinforces these observations by providing a consolidated visual representation of the probability distribution. The higher intensity of colours observed in Level 4 confirms its dominance across the sampling iterations, while Levels 2 and 3 exhibit moderate intensity patterns. Lower intensity regions corresponding to Levels 1 and 5 indicate their relatively limited contribution to the overall classification outcomes. To this end, the consistent dominance of the High-risk significance level, supported by the moderate presence of intermediate levels and limited activation of extreme levels, demonstrates the stability and realistic behaviour of the fuzzy inference system in capturing the underlying risk dynamics. Table 4 shows the fuzzy analysis outputs.

FIGURE 12

FIGURE 13

4.4 Sensitivity analysis

To evaluate the robustness of the fuzzy risk assessment model, a sensitivity analysis was conducted in accordance with the methodology. Each input parameter was varied by 10%, and the resulting changes in the F-RN were examined. The findings indicate that the variation in F-RN values remained within a ±6% to ±6.5% range across the top risk categories, demonstrating strong model stability. Importantly, these variations did not affect the relative ranking of the risks. Figure 14 illustrates the effect of varying one of the inputs on the F-RN.

FIGURE 14

5 Discussion

This section discusses the most significant risk categories associated with integrating GenAI into RM for SCPs, based on the outputs of the developed fuzzy-based assessment model. Building on the earlier methodological stages, the analysis highlights the risks that emerged as most critical within each category. In practical terms, high-significance risks require immediate attention and targeted mitigation, while moderate risks require closer review and low risks can be monitored periodically.

5.1 Input quality-related risks

Input quality risks encompass factors related to the accuracy, completeness, consistency, and contextual relevance of data used in training and deploying GenAI models (Chenya et al., 2022; Holzmann and Lechiara, 2022). These risks are particularly salient in SCPs, where GenAI-enabled RM relies heavily on clean, diverse, and current datasets to generate reliable predictions and inform effective decision-making (Wijayasekera et al., 2022). Findings from the fuzzy analysis highlight the criticality of input quality risks. Specifically, R02 (data unavailability) was ranked 2nd overall (F-RN = 3.63), R03 (data bias) ranked 4th (F-RN = 3.59), and R01 (inaccurate or incomplete data) ranked 8th (F-RN = 3.52). These high rankings clearly demonstrate that substandard data inputs can severely undermine the predictive validity and reliability of GenAI models used in SCP risk assessments.

The effective deployment of GenAI in SCPs hinges critically on the quality and availability of input data, where inaccuracies, omissions, or biases can significantly compromise the integrity of risk identification and mitigation strategies (Aldoseri et al., 2023). Data unavailability, often due to fragmented data silos or inaccessible historical records—restricts GenAI’s ability to learn from past risk occurrences, thereby limiting its capacity to generalize effectively and produce reliable forecasts for future project scenarios (Mahmood et al., 2024). Similarly, data bias introduces systematic distortions in model outputs, potentially misclassifying emerging threats or ignoring contextual factors unique to SCPs, such as regulatory, environmental, or stakeholder-driven complexities (Chenais et al., 2023).

Inaccurate or incomplete data further compounds the challenge by introducing noise, uncertainty, and missing contextual signals. These deficiencies not only degrade model performance but also jeopardize decision outcomes across project planning, risk assessment, and stakeholder communication (Kumar et al., 2024; Delello et al., 2025). Erroneous data points or poorly annotated training samples can lead to biased parameter estimation, flawed mitigation strategies, and ineffective prioritization of risks. As (Zhang and Zhang, 2023) note, poor data annotation practices significantly reduce GenAI system reliability, especially in dynamic and risk-sensitive environments such as SCPs. Moreover, the implications of data quality issues extend beyond technical performance. As (Steimers and Schneider, 2022) highlight, data governance also encompasses ethical and operational dimensions, especially in SCPs where RM decisions often have long-term sustainability implications, including financial losses, safety breaches, or environmental degradation. Addressing these challenges demands a multi-pronged strategy involving stringent data validation protocols, robust data governance frameworks, and cross-stakeholder collaboration to enable seamless data sharing and integration (Adesina et al., 2024; Ambasht, 2023).

5.2 Technological adaptability-related risks

Technological adaptability risks refer to the challenges involved in embedding GenAI into existing construction management workflows, systems, and decision-making processes. These risks are especially significant in the construction industry, where human judgment, domain expertise, and contextual interpretation remain indispensable to project delivery (Adekunle et al., 2022; Hu and Castro-Lacouture, 2019). Fuzzy analysis highlights human error (R11) as the most critical risk in this category, ranking 1st overall with an F-RN of 3.68. This underscores a key vulnerability: while GenAI can produce sophisticated and valuable insights, its effectiveness is ultimately constrained by the competence and attentiveness of the human actors interpreting and applying those outputs (Grewal et al., 2024). This finding highlights the critical importance of the human–machine interface in construction workflows, where the quality of decision-making is only as strong as the human ability to engage meaningfully with GenAI systems, as emphasised by Epstein et al. (2023) and Ghimire et al. (2023).

The second most significant risk in this category is insufficient training (R07), which ranks 3rd overall (F-RN = 3.60). This reflects a pervasive lack of readiness among construction professionals to effectively utilize and collaborate with GenAI technologies, leading to potential inefficiencies, poor system adoption, and suboptimal integration outcomes (Taiwo et al., 2024). Inadequate exposure to digital tools or unfamiliarity with GenAI functionality can result in dependency without understanding, weakening confidence in AI-assisted decisions and increasing the likelihood of operational errors. Closely related is the risk of misinterpretation of GenAI results (R10), which ranks 10th overall (F-RN = 3.46). Limited GenAI literacy can precipitate misguided conclusions, particularly when outputs are taken at face value without a thorough understanding of underlying assumptions, model limitations, or contextual nuances (Grewal et al., 2024; Hassoun et al., 2024). These misjudgments can undermine project outcomes and erode stakeholder trust, especially when decisions based on flawed interpretations lead to delays, cost overruns, or misaligned risk responses.

These findings are echoed in broader literature emphasising the need for not only technical training but also the cultivation of higher-order cognitive skills. Construction professionals must be equipped not just to operate GenAI tools, but also to critically evaluate, contextualise, and apply GenAI-derived insights within complex, dynamic project environments (Lee and Shin, 2020; Pan and Zhang, 2021). This requires the development of GenAI-focused curricula and comprehensive training programs that bridge the gap between algorithmic output and practical project requirements. Proactively managing these technological adaptability risks is essential for unlocking the full potential of GenAI in construction. It calls for a multidimensional strategy that includes continuous workforce upskilling, intuitive system design, and strong digital leadership to reduce friction and resistance (Ghimire et al., 2023). Such efforts foster innovation while minimising the risk of errors and misapplications that could otherwise undermine project success (Reis and Melão, 2023).

5.3 Ethical and governance-related risks

Ethical and governance risks encompass the legal, regulatory, and ethical challenges arising from the deployment of GenAI in SCPs. These risks are particularly significant in data-sensitive, high-stakes environments where decisions influenced by GenAI may have far-reaching consequences (Rane, 2023; Regona et al., 2024). As GenAI technologies become more integrated into project decision-making, questions of fairness, transparency, and accountability become increasingly pressing. Based on the fuzzy analysis, the most critical risk in this category is unclear responsibility and accountability (R20), which ranks 1st within the category and 7th overall with an F-RN of 3.55. This highlights a major concern in GenAI-assisted decision-making: determining who is liable when AI-generated insights contribute to negative project outcomes (Evans and Pasquale, 2022). As construction projects become more digitised and AI-reliant, ambiguity in accountability can create legal grey areas—particularly in hybrid decision-making settings where responsibilities are shared between human actors and GenAI systems (Hendrycks et al., 2023).

Building on this concern, confidentiality breaches (R19) and noncompliance with organisational data privacy policies (R22) are also ranked prominently at 2nd within the category and 12th overall (F-RN = 3.45). These risks reflect heightened anxieties about the misuse of sensitive data, unauthorised access, and the potential exposure of confidential project information. The reliance of GenAI models on large and often sensitive datasets exacerbates these concerns, making data protection a central issue (Palaniappan et al., 2024; Stahl, 2021). Beyond the risk of regulatory penalties, such breaches can significantly damage the reputational standing of construction firms, particularly those involved in publicly funded or regulated SCPs. Closely related are the risks associated with non-transparent decision-making processes, namely, R18 and R21, both of which rank 3rd in the category and 16th overall with an F-RN of 3.41. These risks highlight the intrinsic opacity of many GenAI systems, especially deep learning models, that often function as “black boxes,” where the internal logic driving decisions are not easily understandable to users. This lack of explainability can diminish trust, hinder stakeholder engagement, and limit the practical adoption of GenAI in critical project functions (Chandrika Kandasamy, 2024).

Together, these findings reinforce the broader call in the literature for the development of robust legal, ethical, and governance frameworks tailored to the unique context of GenAI in the built environment (Parveen, 2018; Pillai and Matus, 2020; Regona et al., 2022). The absence of clearly delineated roles, enforceable standards, and transparent auditing mechanisms presents a substantial barrier to responsible GenAI deployment. Legal uncertainty over liability, combined with ethical challenges surrounding data use and algorithmic fairness, necessitates proactive governance approaches capable of managing these evolving risks. Integrating ethical and regulatory considerations into GenAI adoption strategies therefore demands more than baseline compliance (Zhang and Zhang, 2023; Raza et al., 2025). It requires the establishment of clear accountability structures, alignment with data governance policies, and the embedding of responsible GenAI practices into construction project workflows (Xue and Pang, 2022). These efforts are essential for building stakeholder confidence and ensuring that GenAI contributes meaningfully and sustainably to risk management in SCPs.

5.4 Information integrity-related risks

Information integrity risks encompass threats related to data security, authenticity, and system reliability in GenAI-powered RM. As construction firms increasingly digitise workflows and integrate AI-driven tools, maintaining the integrity of the information supporting these systems becomes essential (Rane, 2023). These risks are particularly pertinent in SCPs, where risk assessments often rely on complex, multi-source data environments. The broader literature emphasises that vulnerabilities in data handling not only undermine trust in GenAI systems but also increase the likelihood of flawed project decisions, cost escalations, and reputational damage (Gupta et al., 2023; Omotayo et al., 2024).

The fuzzy analysis identifies data breach (R23) as the most critical risk in this category, ranking 1st within the group and 4th overall with an F-RN of 3.59. This finding aligns with prior research that highlights cybersecurity as a foundational challenge in GenAI integration, particularly in sectors handling sensitive, high-value data (Jada and Mayayise, 2024). In construction, where data-sharing across partners, contractors, and regulatory bodies is routine, the threat of unauthorised access or malicious exploitation is significantly amplified. Compounding this concern, earlier studies have noted the lack of sector-specific cybersecurity protocols as a key barrier to the safe deployment of GenAI in construction (Ghimire et al., 2023), reinforcing the urgency of addressing this risk.

The second highest risk in this category is data fabrication or manipulation (R25), which ranks 14th overall with an F-RN of 3.44. This concern is well-documented in the GenAI ethics literature, where tampered or falsified data is known to compromise model outputs by introducing bias or misleading patterns (Chandrika Kandasamy, 2024). In the context of SCPs—where project conditions are often dynamic, localised, and non-standardised—unverified or manipulated data can distort GenAI’s ability to accurately assess risk exposures. These results echo previous calls for implementing data provenance systems and real-time validation mechanisms as critical safeguards in GenAI-driven decision environments (Longpre et al., 2024). The third major concern is overdependence on synthetic data (R24), which ranks 19th overall with an F-RN of 3.39. While synthetic data offers scalability and addresses data privacy constraints, it may fall short in capturing the complexity, variability, and contextual nuances inherent in real-world construction projects (van Breugel et al., 2023). Prior studies have cautioned that an excessive reliance on synthetic datasets may result in blind spots during risk prediction, especially in industries like construction where workflows are heterogeneous and non-standardised (Stahl, 2021).

These findings reinforce a growing body of literature advocating for comprehensive strategies to ensure information integrity in GenAI implementation. This includes the development of robust cybersecurity infrastructure, data authenticity protocols, and balanced data sourcing practices (Jallan and Ashuri, 2020; Regona et al., 2022; Yao and Soto, 2024). Cybersecurity must go beyond basic protections to encompass AI-specific safeguards such as encrypted model pipelines and context-sensitive intrusion detection systems (Singh et al., 2023). Simultaneously, ensuring data authenticity through validation tools, audit trails, and provenance tracking is essential to prevent flawed or manipulated inputs from undermining trust in AI-generated outputs. Moreover, the risks associated with synthetic data highlight the need for thoughtful integration of real-world data to maintain model reliability. While synthetic data can supplement scarce datasets, it should not substitute the richness and unpredictability of actual project conditions (Meiser and Zinnikus, 2024). Without these safeguards in place, the use of GenAI in RM remains susceptible to both technical failures and ethical breaches, which can significantly erode stakeholder trust and jeopardise project success (Stanovsky et al., 2025). Ultimately, aligning technical measures with construction-specific standards and ethical guidelines is vital to ensure responsible and effective GenAI integration in risk management.

5.5 Financial-related risks

Financial risks represent the economic uncertainties and cost-related concerns tied to the adoption and integration of GenAI technologies into risk management practices for SCPs. These risks are especially significant in a sector where project budgets are tightly managed and investments in emerging technologies are often scrutinised for their long-term value (Regona et al., 2024; Salzano et al., 2024). In such settings, the perceived financial viability of GenAI becomes a key factor influencing its acceptance among construction stakeholders (Ghimire et al., 2023), particularly when the return on investment (ROI) is not immediately evident.

In this study, the fuzzy analysis identified ROI outcome discrepancies (R28) as the most critical risk within the financial category. It ranked 1st in this domain and 11th overall, with an F-RN of 3.47. This finding underscores the prevalent uncertainty regarding the actual value GenAI may deliver over time (Pashazanous, 2025). While GenAI has the potential to enhance decision-making accuracy, reduce exposure to risk, and streamline operational processes, many project managers remain cautious (Sai et al., 2025). A major contributor to this caution is the gap between the anticipated benefits and the realised outcomes, which discourages resource allocation in the absence of reliable financial forecasting and ROI evaluation tools (Fabricius and Büttgen, 2015).

Following closely is the risk of high initial investment costs (R27), ranked 2nd in the financial category and 28th overall, with an F-RN of 3.21. Although high implementation costs are often seen as a barrier to digital innovation, the relatively lower overall ranking suggests that stakeholders might be open to absorbing these upfront costs—provided there is a well-defined path to long-term value. However, the magnitude of investment required for GenAI infrastructure, licensing, workforce training, and integration poses a substantial challenge, especially for small to medium-sized firms with limited financial flexibility and digital maturity (Gurjar et al., 2024). The third-highest risk in this category, customisation and integration expenses (R30), ranks 3rd within the financial domain and 29th overall, with an F-RN of 3.17. This risk highlights the financial burden associated with tailoring GenAI tools to specific projects or organisational needs. Ensuring system compatibility, modifying workflows, and training personnel all incur additional costs (Ghimire et al., 2023). These often hidden or underestimated expenses can further complicate investment planning and hinder the scalability of GenAI implementation, particularly in resource-constrained construction environments.

These findings are consistent with prior studies that underline the importance of strategic financial planning when introducing advanced digital technologies into traditional project environments. As Xu and Cho (2025) and Cooper (2026) argue, the lack of structured cost–benefit frameworks and performance-tracking mechanisms can obscure the financial justification for GenAI investment. Developing transparent ROI assessment tools, aligning GenAI adoption with broader business objectives, and setting realistic performance expectations are necessary steps to ensure financially sound and sustainable implementation in SCP risk management contexts.

6 Contributions

The contributions of this research are twofold. First, the study develops a structured taxonomy of 30 GenAI-related risk factors for SCPs, grouped into five thematic categories: input quality, technological adaptability, ethical and governance, information integrity, and financial risks. Second, it proposes a hierarchical fuzzy assessment model that integrates probability, impact, detectability, and exposure to quantify the significance of these risks under conditions of uncertainty.

6.1 Theoretical contribution

This study contributes to theoretical understanding of GenAI-enabled risk management in SCPs by clarifying how distinct categories of risks emerge from interactions between data quality, human interpretative capacity and organisational governance structures. The findings show that GenAI-related risks cannot be understood solely as technical issues. They must be conceptualised as socio-technical phenomena shaped through the alignment, or misalignment, between data ecosystems, human expertise and organisational procedures (Adekunle et al., 2022; Chenya et al., 2022). This expands existing discussions that highlight GenAI’s dependence on high-quality, complete and contextually relevant data inputs (Holzmann and Lechiara, 2022; Wijayasekera et al., 2022; Mahmood et al., 2024). The prominence of input-related risks reinforces the theoretical view that data governance, annotation accuracy and dataset representativeness remain foundational determinants of GenAI reliability in construction environments (Zhang and Zhang, 2023; Delello et al., 2025).

The analysis also strengthens theoretical knowledge by illustrating how persistent human-centred risks, including human error, limited GenAI literacy and misinterpretation of GenAI outputs, influence the adoption of AI-enabled risk management tools. The findings align with literature emphasising that construction professionals draw heavily on higher-order cognitive skills and contextual judgement when evaluating AI-generated insights (Epstein et al., 2023; Grewal et al., 2024). The significance of these risks suggests that theories of digital transformation and GenAI adoption in construction should give greater weight to human–GenAI interaction, recognising that interpretability, user competence and cognitive readiness shape the actual value derived from GenAI tools (Lee and Shin, 2020; Pan and Zhang, 2021; Taiwo et al., 2024).

The fuzzy assessment model developed in this study introduces a theoretical–methodological bridge between subjective expert judgement and quantifiable risk significance. The inclusion of fuzzy entropy strengthens this bridge because it formalises the representation of ambiguity and variability in expert evaluations. This responds to established limitations in traditional risk assessment methods that struggle to capture uncertainty in expert judgement (Adesina et al., 2024; Ambasht, 2023). This approach reinforces the theoretical premise that uncertainty is a central component of construction risk modelling, particularly when evaluating emerging technologies such as GenAI. The study therefore provides a foundation for future work seeking to integrate socio-technical, cognitive and governance dimensions within holistic models of GenAI-driven risk management.

6.2 Methodological contribution

This study offers methodological contributions that enhance transparency, reliability and interpretive depth in GenAI-related risk assessment for SCPs. The fuzzy-based model enables a structured analysis of expert judgements, which often contain ambiguity, subjectivity and inconsistency by providing a systematic mechanism for converting linguistic ratings into quantifiable expressions, allowing expert perceptions to be compared and aggregated more meaningfully across different risk categories (Holzmann and Lechiara, 2022; Chenya et al., 2022).

The incorporation of fuzzy entropy adds another methodological refinement because it detects variability and disagreement in expert evaluations. This reduces the influence of low-consensus inputs and improves overall robustness (Adesina et al., 2024; Ambasht, 2023). Uncertainty is therefore treated as a measurable dimension of expert knowledge rather than an analytical weakness, which supports more reliable decision-making. The structured evaluation across input quality, technological adaptability, ethical and governance conditions, information integrity and financial dimensions represents an additional methodological advancement. Existing GenAI studies in construction typically prioritise predictive accuracy or algorithmic optimisation. The current framework instead captures socio-technical interactions that emerge at human, organisational and data governance levels (Grewal et al., 2024; Steimers and Schneider, 2022). This broader analytical scope enables the identification of vulnerabilities that are often overlooked in conventional models yet remain central to responsible GenAI integration in construction risk management.

6.3 Managerial implications

Managing risk in SCPs presents continuing challenges as GenAI systems become more embedded in project decision-making. Project conditions shift, data availability fluctuates and expectations for transparency and accountability continue to rise. These conditions create a setting in which GenAI-generated insights must be interpreted carefully to avoid decisions that unintentionally introduce new risks or amplify existing weaknesses (Grewal et al., 2024; Palaniappan et al., 2024). GenAI tools can support more informed risk identification and improve analytical precision. However, the effective implementation of the proposed framework may face practical barriers, including limited organisational expertise, uneven understanding of GenAI-specific impacts, and the absence of fully developed regulatory guidance for GenAI use in RM for SCPs.

The findings highlight the importance of organisational readiness in areas such as data governance, training and ethical oversight. Human error, insufficient training and misinterpretation of GenAI results emerged as critical risks, indicating that managers should cultivate environments where GenAI literacy, critical interpretation and reflective analysis are encouraged (Adekunle et al., 2022; Taiwo et al., 2024). Project managers, engineers and technical staff need systematic procedures that question how data were collected, how models interpret information and whether outputs align with project context (Epstein et al., 2023). This strengthens confidence in AI-assisted decisions and reduces the likelihood of accepting algorithmic outputs without proper evaluation. Data governance and information integrity also present important managerial implications. Risks associated with data breaches, data manipulation and improper handling of sensitive information reinforce the need for strong cybersecurity protocols, data validation mechanisms and transparent data management practices (Ghimire et al., 2023; Jada and Mayayise, 2024). Senior managers should ensure that organisational policies are aligned with the specific sensitivities of GenAI-enabled workflows, particularly where decisions involve legal, financial or reputational consequences (Regona et al., 2024). Addressing these issues proactively improves trust among internal and external stakeholders. Financial risks linked to ROI uncertainty, integration costs and customisation requirements highlight the need for rigorous financial planning. Managers should assess not only initial GenAI investment, but also ongoing operational costs associated with training, integration and system maintenance (Gurjar et al., 2024). These assessments help organisations balance expected efficiency gains against the resources required for sustainable and responsible GenAI implementation.

Although the developed model provides a clearer picture of the risk landscape surrounding GenAI adoption, it also indicates areas where human judgement and organisational learning remain central. GenAI cannot substitute professional expertise or contextual interpretation. Managers must therefore adopt a well-informed and critical stance toward GenAI outputs, recognising both their strategic value and their limitations within complex sustainable construction environments.

7 Conclusion

This study sets out to evaluate the risks associated with integrating GenAI into risk management practices for SCPs. The research followed a structured multi-stage design that included a systematic literature review, Delphi-based model refinement, expert elicitation through a survey, and fuzzy-based multi-criteria risk quantification using probability, impact and detectability. Although fuzzy logic has been applied previously in construction risk modelling, existing studies have not examined GenAI-specific risks nor integrated a structured hierarchy of input variables tailored to GenAI-enabled decision processes. The current work therefore extends the literature through a dedicated GenAI risk taxonomy and a fuzzy inference architecture that incorporates uncertainty quantification via fuzzy entropy.

The developed model enabled the identification and evaluation of 30 risks categorised into five domains: input quality, technological adaptability, ethical and governance, information integrity, and financial risks. Human error, data unavailability and insufficient training emerged as the most influential risks, reflecting the strong dependence of GenAI performance on both technical and human-centred factors. The model outputs indicate that misinterpretation of GenAI results, fragmented datasets and limited digital readiness can significantly disrupt GenAI-enabled decision-making in sustainable construction contexts. These findings reinforce growing evidence that successful GenAI integration requires more than technical deployment; it requires organisational capacity building, enhanced data governance and clearer decision accountability mechanisms.

The study contributes methodologically through the development of a modular fuzzy inference system that formalises expert judgement, incorporates entropy-based uncertainty weighting and allows transparent traceability of how linguistic assessments propagate through the risk significance computation. The approach aligns with established fuzzy risk modelling principles while introducing a GenAI-focused structure that captures socio-technical interactions across project environments. Sensitivity analysis confirmed the numerical stability of the model, with no ranking changes observed under ±10 percent variation in inputs, supporting its suitability for early-stage decision support.

The practical implications of the findings highlight the need for construction organisations to prioritise data quality infrastructure, invest in GenAI-related upskilling programmes and develop governance frameworks that ensure responsible and accountable use of AI-generated insights. The prominence of ethical, interpretative and information integrity risks suggests that organisations should implement clearer protocols for data sharing, transparency and model explainability to maintain trust and reduce operational vulnerabilities.

7.1 Research limitations

This study offers valuable insights into the risks associated with integrating GenAI into sustainable construction project risk management; however, several limitations should be acknowledged. The sample of 80 respondents provides a suitable foundation for exploratory analysis, but it cannot be considered fully representative of the wider construction sector. Although many participants possessed considerable experience in construction management, only a limited number demonstrated advanced expertise in GenAI applications. As a result, the findings primarily reflect the current level of industry maturity rather than perspectives from a highly experienced GenAI user base. Consequently, the findings should be interpreted as an informed sectoral assessment, and future research would benefit from a larger and more diverse sample, including a greater proportion of respondents with advanced GenAI expertise.

The model depends on expert linguistic ratings, which inherently introduce perceptual variability. These ratings were processed through a fixed Mamdani fuzzy-inference structure that supports transparency and interpretability, though it does not adapt to new data dynamically. The use of crisp survey averages instead of real-time project data means the model reflects stable expert assessments rather than evolving site conditions. Future work could explore adaptive, data-driven fuzzy systems to extend applicability in dynamic project environments.

Although the variable set and controller structure were consolidated through a two-round Delphi process to ensure parsimony and conceptual coherence, a formal multicollinearity diagnostic was not undertaken because the FMEA triad inputs (probability, impact and detectability) are theoretically distinct dimensions expected to show constructive, non-redundant association. Nonetheless, correlation or PCA-based diagnostics will be included in future validation studies once a larger dataset becomes available.

Finally, the sensitivity analysis used in this study followed a deterministic ±10% variation approach, which is standard for fuzzy-inference model validation. More advanced probabilistic stress-testing methods such as Monte Carlo fuzzification or Sobol sensitivity could yield additional insights, yet they require assumptions about input distributions that extend beyond the scope of the present dataset. These techniques represent a logical direction for future research as more granular input data become accessible.

7.2 Future research directions

Future research should validate the proposed model using data from real construction projects, by comparing fuzzy-generated risk levels with observed project outcomes. It should also test the framework across different national and organisational contexts to examine how regulatory conditions, digital maturity, and industry practices shape risk perception. In addition, comparative studies using methods such as AHP, Bayesian approaches, or probabilistic sensitivity analysis could help assess the relative strengths of the proposed fuzzy model where sufficient data are available. Further work may also link risk scores to project performance indicators such as cost overruns, schedule delays, and safety incidents, while exploring more adaptive fuzzy systems for dynamic risk environments (Dikmen et al., 2007).

References

• 1

  AdakS.KarT.JanaS. (2024). A fuzzy inference system for predicting outbreaks in emerging infectious diseases. Decis. Anal. J.10, 100436. 10.1016/j.dajour.2024.100436

  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  AdekunleP.AigabvboaC.ThwalaW.AkinradewoO.OkeA. (2022). Challenges confronting construction information management. Front. Built Environ.8, 1075674. 10.3389/fbuil.2022.1075674

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  AdesinaA. A.IyeloluT. V.PaulP. O. (2024). Leveraging predictive analytics for strategic decision-making: enhancing business performance through data-driven insights. World J. Adv. Res. Rev.22, 1927–1934. 10.30574/wjarr.2024.22.3.1961

  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  AfzalF.YunfeiS.NazirM.BhattiS. M. (2021). A review of artificial intelligence based risk assessment methods for capturing complexity-risk interdependencies: cost overrun in construction projects. Int. J. Manag. Proj. Bus.14, 300–328. 10.1108/ijmpb-02-2019-0047

  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  AkramF.AhmadT.SadiqM. (2024). An integrated fuzzy adjusted cosine similarity and TOPSIS based recommendation system for information system requirements selection. Decis. Anal. J.11, 100443. 10.1016/j.dajour.2024.100443

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  Al-MhdawiM. K. S.O'ConnorA. (2026). Impact of sustainable safety climate on poor compliance with personal protective equipment in oil and gas construction projects: quantitative analysis and best sustainable practices. Environ. Impact Assess. Rev.117, 108211. 10.1016/j.eiar.2025.108211

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  Al-MhdawiM. K. S.QaziA.AlzarradA.DacreN.RahimianF.BuniyaM. K.et al (2023). Expert evaluation of ChatGPT performance for risk management process based on ISO 31000 standard. SSRN Electron. J., 1–6. 10.2139/ssrn.4504409

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  Al-MhdawiM. K. S.BritoM.OnggoB. S.QaziA.O'ConnorA. (2024a). COVID-19 emerging risk assessment for the construction industry of developing countries: evidence from Iraq. Int. J. Constr. Manag.24, 693–706. 10.1080/15623599.2023.2169301

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  Al-MhdawiM. K. S.O'ConnorA.QaziA. (2024b). Structural equation modeling and fuzzy set theory: advancing risk assessment in oil and gas construction projects. Environ. Impact Assess. Rev.109, 107622. 10.1016/j.eiar.2024.107622

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  Al-MhdawiM. K. S.QaziA.KarakhanA. A.RahimianF. P.AbualqumbozM.Al MhdawiA. K.et al (2024c). Implementation of a combined fuzzy controller model to enhance risk assessment in oil and gas construction projects. IEEE Access12, 68319–68333. 10.1109/access.2024.3399129

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  Al-MhdawiM. K. S.O'connorA.QaziA.RahimianF.DacreN. (2025). Review of studies on risk factors in critical infrastructure projects from 2011 to 2023. Smart Sustain. Built Environ.14, 342–376. 10.1108/SASBE-09-2023-0285

  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  AladagH. (2023). Assessing the accuracy of ChatGPT use for risk management in construction projects. Sustainability15, 16071. 10.3390/su152216071

  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  AldoseriA.Al-KhalifaK. N.HamoudaA. M. (2023). Re-thinking data strategy and integration for artificial intelligence: concepts, opportunities, and challenges. Appl. Sci.13, 7082. 10.3390/app13127082

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  AlmashhourR.Al-MhdawiM. K. S.DaghfousA.QaziA.OjiakoU. (2025). Traditional to sustainable risk management in the construction industry: a systematic literature review. Int. J. Manag. Proj. Bus.18 (3), 528–565. 10.1108/ijmpb-01-2025-0021

  • CrossRef
  • Google Scholar
  • View reference in article

• 15

  AlvandA.MirhosseiniS. M.EhsanifarM.ZeighamiE.MohammadiA. (2023). Identification and assessment of risk in construction projects using the integrated FMEA-SWARA-WASPAS model under fuzzy environment: a case study of a construction project in Iran. Int. J. Constr. Manag.23, 392–404. 10.1080/15623599.2021.1877875

  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  AmbashtA. (2023). Real-time data integration and analytics: empowering data-driven decision making. Int. J. Comput. Trends Technol.71, 8–14. 10.14445/22312803/ijctt-v71i7p102

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  AnY.LiH.SuT.WangY. (2021). Determining uncertainties in AI applications in AEC sector and their corresponding mitigation strategies. Autom. Constr.131, 103883. 10.1016/j.autcon.2021.103883

  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  ArunrajN. S.MandalS.MaitiJ. (2013). Modeling uncertainty in risk assessment: an integrated approach with fuzzy set theory and monte carlo simulation. Accid. Anal. Prev.55, 242–255. 10.1016/j.aap.2013.03.007

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  AzharS. (2011). Building information modeling (BIM): trends, benefits, risks, and challenges for the AEC industry. Leadersh. Manag. Eng.11, 241–252. 10.1061/(asce)lm.1943-5630.0000127

  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  AzizF. (2025). Generative artificial intelligence in AEC organizations: a literature and SWOT analysis. J. Intell. Constr.3, 1–23. 10.26599/jic.2025.9180094

  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  BengesiS.El-SayedH.SarkerM. K.HoukpatiY.IrunguJ.OladunniT. (2024). Advancements in generative AI: a comprehensive review of GANs, GPT, autoencoders, diffusion model, and transformers. IEEE Access. 12, 69812–69837. 10.1109/ACCESS.2024.3397775

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  BentoS.PereiraL.GonçalvesR.DiasÁ.CostaR. L. D.Costaa. (2022). Artificial intelligence in project management: systematic literature review. Int. J. Technol. Intell. Plan.13, 143–163. 10.1504/ijtip.2022.126841

  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  BlomkvistJ. (2015). “Understanding the results of conventional qualitative content analysis for design research,” in 11th international european academy of design conference.

  • Google Scholar
  • View reference in article

• 24

  BrymanA. (2016). Social research methods. Oxford University Press.

  • Google Scholar
  • View reference in article

• 25

  Chandrika KandasamyU. (2024). Ethical leadership in the age of AI challenges. Oppor. Framew. Ethical leadersh.

  • Google Scholar
  • View reference in article

• 26

  ChartresN.BeroL. A.NorrisS. L. (2019). A review of methods used for hazard identification and risk assessment of environmental hazards. Environ. Int.123, 231–239. 10.1016/j.envint.2018.11.060

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  ChenaisG.LagardeE.Gil-JardinéC. (2023). Artificial intelligence in emergency medicine: viewpoint of current applications and foreseeable opportunities and challenges. J. Med. Internet Res.25, 40031. 10.2196/40031

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  ChenyaL.AminudinE.MohdS.YapL. S. (2022). Intelligent risk management in construction projects: systematic literature review. IEEE Access10, 72936–72954. 10.1109/ACCESS.2022.3189157

  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  ChrysostomS.DwivediR. (2016). A state of the art review of fuzzy approaches used in the failure modes and effects analysis: a call for research. Int. J. Ind. Syst. Eng.23, 351–369. 10.1504/ijise.2016.076868

  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  CooperR. G. (2026). Driving business value: a strategic framework for AI adoption and deployment success. IEEE Eng. Manag. Rev. pp(99), 1–19. 10.1109/EMR.2026.3660823

  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  Cox JrL. (2008). What's wrong with risk matrices? Risk anal. Int. J.28, 497–512. 10.1111/j.1539-6924.2008.01030.x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 32

  DelelloJ. A.SungW.MokhtariK.HebertJ.BronsonA.GiuseppeT. (2025). AI in the classroom: insights from educators on usage, challenges, and mental health. Educ. Sci.15, 113. 10.3390/educsci15020113

  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  DikmenI.BirgonulM. T.HanS. (2007). Using fuzzy risk assessment to rate cost overrun risk in international construction projects. Int. J. Proj. Manag.25, 494–505. 10.1016/j.ijproman.2006.12.002

  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  DriskoJ. W.MaschiT. (2016). Content analysis. New York, NY:Oxford University Press.

  • Google Scholar
  • View reference in article

• 35

  ElmontsriM. (2014). Review of the strengths and weaknesses of risk matrices. J. Risk Anal. Crisis Response4, 49–57. 10.2991/jrarc.2014.4.1.6

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  ElseknidyM.Al-MhdawiM. K. S.QaziA.OjiakoU.MahammediC.RahimianF. P. (2025). Developing a sustainability-driven risk management framework for green building projects: a literature review. J. Clean. Prod.519, 145891. 10.1016/j.jclepro.2025.145891

  • CrossRef
  • Google Scholar
  • View reference in article

• 37

  EpsteinZ.HertzmannA.Human CreativityI.AktenM.FaridH.FjeldJ.et al (2023). Art and the science of generative AI. Science380, 1110–1111. 10.1126/science.adh4451

  • CrossRef
  • Google Scholar
  • View reference in article

• 38

  EvansB. J.PasqualeF. (2022). Product liability suits for FDA-Regulated AI/ML software. Future Med. Device Regul. Innov. Prot.656, 22–35. 10.1017/9781108975452.003

  • CrossRef
  • Google Scholar
  • View reference in article

• 39

  FabriciusG.BüttgenM. (2015). Project managers' overconfidence: how is risk reflected in anticipated project success?Bus. Res.8, 239–263. 10.1007/s40685-015-0022-3

  • CrossRef
  • Google Scholar
  • View reference in article

• 40

  FeuerriegelS.HartmannJ.JanieschC.ZschechP. (2024). Generative ai. Bus. Inf. Syst. Eng.66, 111–126. 10.1007/s12599-023-00834-7

  • CrossRef
  • Google Scholar
  • View reference in article

• 41

  Fui-Hoon NahF.ZhengR.CaiJ.SiauK.ChenL. (2023). Generative AI and ChatGPT: applications, challenges, and AI-human collaboration. J. Inf. Technol. Case Appl. Res.25, 277–304. 10.1080/15228053.2023.2233814

  • CrossRef
  • Google Scholar
  • View reference in article

• 42

  GerlaG. (2013). Fuzzy logic: mathematical tools for approximate reasoning. Springer Science and Business Media.

  • Google Scholar
  • View reference in article

• 43

  GhimireP.KimK.AcharyaM. (2023). Generative ai in the construction industry: opportunities and challenges.

  • Google Scholar
  • View reference in article

• 44

  GholamizadehK.ZareiE.OmidvarM.YazdiM. (2022). “Fuzzy sets theory and human reliability: review, applications, and contributions,” in Linguistic methods under fuzzy information in system safety and reliability analysis, 91–137.

  • Google Scholar
  • View reference in article

• 45

  GiesekamJ.BarrettJ.TaylorP.OwenA. (2014). The greenhouse gas emissions and mitigation options for materials used in UK construction. Energy Build.78, 202–214. 10.1016/j.enbuild.2014.04.035

  • CrossRef
  • Google Scholar
  • View reference in article

• 46

  GrewalD.SatorninoC. B.DavenportT.GuhaA. (2024). How generative AI is shaping the future of marketing. J. Acad. Mark. Sci.53, 1–21. 10.1007/s11747-024-01064-3

  • CrossRef
  • Google Scholar
  • View reference in article

• 47

  GuptaM.AkiriC.AryalK.ParkerE.PraharajL. (2023). From chatgpt to threatgpt: impact of generative ai in cybersecurity and privacy. IEEE Access11, 80218–80245. 10.1109/access.2023.3300381

  • CrossRef
  • Google Scholar
  • View reference in article

• 48

  GurjarK.JangraA.BaberH.IslamM.SheikhS. A. (2024). An analytical review on the impact of artificial intelligence on the business industry: applications, trends, and challenges. IEEE Eng. Manag. Rev.52 (2), 84–102. 10.1109/EMR.2024.3355973

  • CrossRef
  • Google Scholar
  • View reference in article

• 49

  HassounA.AbonizioA.OsbornK.WuC.GoldbergB. (2024). The influencer next door: how misinformation creators use GenAI.

  • Google Scholar
  • View reference in article

• 50

  HeR.CaoJ.TanT. (2025). Generative artificial intelligence: a historical perspective. Natl. Sci. Rev.12, 050. 10.1093/nsr/nwaf050

  • CrossRef
  • Google Scholar
  • View reference in article

• 51

  HendrycksD.MazeikaM.WoodsideT. (2023). An overview of catastrophic AI risks.

  • Google Scholar
  • View reference in article

• 52

  HolzmannV.LechiaraM. (2022). Artificial intelligence in construction projects: an explorative study of professionals' expectations. Eur. J. Bus. Manag. Res.7, 151–162. 10.24018/ejbmr.2022.7.3.1432

  • CrossRef
  • Google Scholar
  • View reference in article

• 53

  HolzmannV.ZitterD.PeshkessS. (2022). The expectations of project managers from artificial intelligence: a Delphi study. Proj. Manag. J.53, 438–455. 10.1177/87569728211061779

  • CrossRef
  • Google Scholar
  • View reference in article

• 54

  HuY.Castro-LacoutureD. (2019). Clash relevance prediction based on machine learning. J. Comput. Civ. Eng.33, 04018060. 10.1061/(asce)cp.1943-5487.0000810

  • CrossRef
  • Google Scholar
  • View reference in article

• 55

  JadaI.MayayiseT. O. (2024). The impact of artificial intelligence on organisational cyber security: an outcome of a systematic literature review. Data Inf. Manag.8, 100063. 10.1016/j.dim.2023.100063

  • CrossRef
  • Google Scholar
  • View reference in article

• 56

  JainV.SangaiahA. K.SakhujaS.ThodukaN.AggarwalR. (2018). Supplier selection using fuzzy AHP and TOPSIS: a case study in the Indian automotive industry. Neural comput. Appl.29, 555–564. 10.1007/s00521-016-2533-z

  • CrossRef
  • Google Scholar
  • View reference in article

• 57

  JallanY.AshuriB. (2020). Text mining of the securities and exchange commission financial filings of publicly traded construction firms using deep learning to identify and assess risk. J. Constr. Eng. Manag.146, 04020137. 10.1061/(asce)co.1943-7862.0001932

  • CrossRef
  • Google Scholar
  • View reference in article

• 58

  KaynakO.ZadehL. A.TürksenB. (2012). Computational intelligence: soft computing and fuzzy-neuro integration with applications. Springer Science and Business Media.

  • Google Scholar
  • View reference in article

• 59

  KibertC. (2016). Sustainable construction: green building design and delivery. John Wiley and Sons.

  • Google Scholar
  • View reference in article

• 60

  KibiswaN. K. (2019). Directed qualitative content analysis (DQlCA): a tool for conflict analysis. Qual. Rep.24, 2059–2079. 10.46743/2160-3715/2019.3778

  • CrossRef
  • Google Scholar
  • View reference in article

• 61

  KiralI. A. (2025). Contextual evaluation of risk identification techniques for construction projects: comparative insights and a decision-support model. Buildings15 (20), 3806. 10.3390/buildings15203806

  • CrossRef
  • Google Scholar
  • View reference in article

• 62

  KrippendorffK. (2018). Content analysis: an introduction to its methodology. Thousand Oaks, CA:Sage publications.

  • Google Scholar
  • View reference in article

• 63

  KritskyD.DruzhininE.PogudinaO.KritskayaO. (2018). “Decision making by the analysis of project risks based on the FMEA method,” in 2018 IEEE 13th international scientific and technical conference on computer sciences and information technologies (CSIT) (IEEE), 187–190.

  • Google Scholar
  • View reference in article

• 64

  KumarP.ChoubeyD.AmosuO. R.OgunsujiY. M. (2024). AI-enhanced inventory and demand forecasting: using AI to optimize inventory management and predict customer demand. World J. Adv. Res. Rev.23, 1931–1944. 10.30574/wjarr.2024.23.1.2173

  • CrossRef
  • Google Scholar
  • View reference in article

• 65

  LachhabM.BélerC.CoudertT. (2018). A risk-based approach applied to system engineering projects: a new learning based multi-criteria decision support tool based on an ant colony algorithm. Eng. Appl. Artif. Intell.72, 310–326. 10.1016/j.engappai.2018.04.001

  • CrossRef
  • Google Scholar
  • View reference in article

• 66

  LauronH.AbellanaD. (2024). An integrated fuzzy multi-criteria decision-making model for determining the interdependencies among the African swine fever spread factors. Decis. Anal. J.11, 100454. 10.1016/j.dajour.2024.100454

  • CrossRef
  • Google Scholar
  • View reference in article

• 67

  LeeI.ShinY. J. (2020). Machine learning for enterprises: applications, algorithm selection, and challenges. Bus. Horiz.63, 157–170. 10.1016/j.bushor.2019.10.005

  • CrossRef
  • Google Scholar
  • View reference in article

• 68

  LiangD.LiuD.PedryczW.HuP. (2013). Triangular fuzzy decision-theoretic rough sets. Int. J. Approx. Reason.54, 1087–1106. 10.1016/j.ijar.2013.03.014

  • CrossRef
  • Google Scholar
  • View reference in article

• 69

  LongpreS.MahariR.Obeng-MarnuN.BrannonW.SouthT.KabbaraJ.et al (2024). “Position: data authenticity, consent, & provenance for ai are all broken: what will it take to fix them?”, in proceedings of the 41st international conference on machine learning. PMLR. 235, 32711–32725. 10.48550/arXiv.2404.12691

  • CrossRef
  • Google Scholar
  • View reference in article

• 70

  MahammediC.MahdjoubiL.BoothC.AkramH.ButtT. (2020). A systematic review of risk assessment tools for contaminated sites–current perspectives and future prospects. Environ. Res.191, 110180. 10.1016/j.envres.2020.110180

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 71

  MahmoodY.ChenJ.YodoN.HuangY. (2024). Optimizing natural gas pipeline risk assessment using hybrid fuzzy bayesian networks and expert elicitation for effective decision-making strategies. Gas. Sci. Eng.125, 205283. 10.1016/j.jgsce.2024.205283

  • CrossRef
  • Google Scholar
  • View reference in article

• 72

  MamdaniE.AssilianS. (1975). An experiment in linguistic synthesis with a fuzzy logic controller. Int. J. Man-Mach. Stud.7, 1–13. 10.1016/s0020-7373(75)80002-2

  • CrossRef
  • Google Scholar

• 73

  MasjedyH.AdelS. M. R. (2025). Fuzzy logic theory: a unifying pathway to intelligent accountability in addressing methodological challenges of applied linguistics. Res. Methods Appl. Linguist.4, 100278. 10.1016/j.rmal.2025.100278

  • CrossRef
  • Google Scholar

• 74

  McMillanS. S.KingM.TullyM. P. (2016). How to use the nominal group and Delphi techniques. Int. J. Clin. Pharm.38, 655–662. 10.1007/s11096-016-0257-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 75

  MeiserM.ZinnikusI. (2024). A survey on the use of synthetic data for enhancing key aspects of trustworthy AI in the energy domain: challenges and opportunities. Energies17, 1992. 10.3390/en17091992

  • CrossRef
  • Google Scholar

• 76

  MohamedM. A. H.Al-MhdawiM. K. S.OjiakoU.DacreN.QaziA.RahimianF. (2025). Generative AI in construction risk management: a bibliometric analysis of the associated benefits and risks. Urbanization, Sustain. Soc.2 (1), 196–228. 10.1108/uss-11-2024-0069

  • CrossRef
  • Google Scholar

• 77

  MohamedM. A. H.Al-MhdawiM. K. S.OjiakoU.QaziA.MahammediC.DacreN. (2026). A GenAI-driven risk management framework for sustainable development projects. Proj. Leadersh. Soc.7, 100217. 10.1016/j.plas.2026.100217

  • CrossRef
  • Google Scholar

• 78

  MüllerR.LocatelliG.HolzmannV.NilssonM.SagayT. (2024). Artificial intelligence and project management: empirical overview, state of the art, and guidelines for future research. Proj. Manag. J.55, 9–15. 10.1177/87569728231225198

  • CrossRef
  • Google Scholar

• 79

  NajiK. K.GunduzM.MohamedA.AlomariA. (2025). Generative AI for sustainable project management in the built environment: trends, challenges, and future directions. Sustainability17, 9063. 10.3390/su17209063

  • CrossRef
  • Google Scholar

• 80

  NyqvistR.PeltokorpiA.SeppänenO. (2024). Integration of generative artificial intelligence across construction management. IOP Conf. Ser. Earth Environ. Sci.1389, 012011. 10.1088/1755-1315/1389/1/012011

  • CrossRef
  • Google Scholar

• 81

  OjiakoU.WongT.SmithC.ChipuluM.Al-MhdawiM. K. S.OyewoB.et al (2025). Advancing resilience in infrastructure projects through machine learning-driven models. Prod. Plan. Control. 1–25. 10.1080/09537287.2025.2583300

  • CrossRef
  • Google Scholar

• 82

  OmotayoT.DengJ.HossainS.KhanS.ParmarM. (2024). “Generative AI for BIM-Based digital construction cost management: a qualitative sentiment analysis approach,” in Proceedings of the 14th annual international conference on industrial engineering and operations management, 12–14.

  • Google Scholar

• 83

  OnatayoD.OnososenA.OyediranA. O.OyediranH.ArowoiyaV.OnatayoE. (2024). Generative AI applications in architecture, engineering, and construction: trends, implications for practice, education and imperatives for Upskilling—A review. Architecture4, 877–902. 10.3390/architecture4040046

  • CrossRef
  • Google Scholar

• 84

  PalaniappanK.LinE. Y. T.VogelS.LimJ. (2024). Gaps in the global regulatory frameworks for the use of artificial intelligence (AI) in the healthcare services sector and key recommendations. Healthcare12, 1730. 10.3390/healthcare12171730

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 85

  PanY.ZhangL. (2021). Roles of artificial intelligence in construction engineering and management: a critical review and future trends. Autom. Constr.122, 103517. 10.1016/j.autcon.2020.103517

  • CrossRef
  • Google Scholar

• 86

  ParveenR. (2018). Artificial intelligence in the construction industry: legal issues and regulatory challenges. Int. J. Civ. Eng. Technol.9, 957–962.

  • Google Scholar

• 87

  PashazanousE. (2025). Exploring the evolution of project management: harnessing the potential of artificial intelligence for future success. Politecnico di Torino: Doctoral dissertation.

  • Google Scholar

• 88

  PillaiV. S.MatusK. J. (2020). Towards a responsible integration of artificial intelligence technology in the construction sector. Sci. Public Policy47, 689–704. 10.1093/scipol/scaa073

  • CrossRef
  • Google Scholar

• 89

  PrebanicK. R.VukomanovicM. (2021). Realizing the need for digital transformation of stakeholder management: a systematic review in the construction industry. Sustainability13, 12690. 10.3390/su132212690

  • CrossRef
  • Google Scholar

• 90

  PrietoS. A.MengisteE. T.SotoB. (2023). Investigating the use of ChatGPT for the scheduling of construction projects. Buildings13, 857. 10.3390/buildings13040857

  • CrossRef
  • Google Scholar

• 91

  QiaoS.ZhiweiL.JieW.YuxiM.GuoZ.HanW. (2025). Generative AI on innovation performance of construction enterprises: the role of knowledge-based dynamic capabilities and enterprise AI capabilities.

  • Google Scholar

• 92

  RaneN. (2023). Integrating building information modelling (BIM) and artificial intelligence (AI) for smart construction schedule, cost, quality, and safety management: challenges and opportunities. Cost. Qual. Saf. Manag. Chall. Oppor.10.2139/ssrn.4616055

  • CrossRef
  • Google Scholar

• 93

  RathoreR.ThakkarJ. J.JhaJ. K. (2021). Evaluation of risks in foodgrains supply chain using failure mode effect analysis and fuzzy VIKOR. Int. J. Qual. Reliab. Manag.38, 551–580. 10.1108/ijqrm-02-2019-0070

  • CrossRef
  • Google Scholar

• 94

  RazaS.QureshiR.ZahidA.KamawalS.SadakF.FioresiJ.et al (2025). Who is responsible? The data, models, users or regulations? A comprehensive survey on responsible generative ai for a sustainable future.

  • Google Scholar

• 95

  RegonaM.YigitcanlarT.XiaB.LiR. (2022). Opportunities and adoption challenges of AI in the construction industry: a PRISMA review. J. Open Innov. Technol. Mark. Complex.8, 45. 10.3390/joitmc8010045

  • CrossRef
  • Google Scholar

• 96

  RegonaM.YigitcanlarT.HonC.TeoM. (2024). Artificial intelligence and sustainable development goals: systematic literature review of the construction industry. Sustain. Cities Soc.108, 105499. 10.1016/j.scs.2024.105499

  • CrossRef
  • Google Scholar

• 97

  ReisJ.MelãoN. (2023). Digital transformation: a meta-review and guidelines for future research. Heliyon9, e12834. 10.1016/j.heliyon.2023.e12834

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 98

  RunesonP.HostM.RainerA.RegnellB. (2012). Case study research in software engineering: guidelines and examples. John Wiley and Sons.

  • Google Scholar

• 99

  SaiS.ArunakarK.ChamolaV.HussainA.BishtP.KumarS. (2025). Generative AI for finance: applications, case studies and challenges. Expert Syst.42, 70018. 10.1111/exsy.70018

  • CrossRef
  • Google Scholar

• 100

  SalzanoA.CasconeS.ZitielloE. P.NicolellaM. (2024). Construction safety and efficiency: integrating building information modeling into risk management and project execution. Sustainability16, 4094. 10.3390/su16104094

  • CrossRef
  • Google Scholar

• 101

  SharmaS.GuptaA. K. (2019). Risk identification and management in construction projects: literature review. Int. J. Humanit. Arts Soc. Sci.5, 224–231. 10.20469/ijhss.5.20002-6

  • CrossRef
  • Google Scholar

• 102

  SinghA.DwivediA.AgrawalD.SinghD. (2023). Identifying issues in adoption of AI practices in construction supply chains: towards managing sustainability. Oper. Manag. Res.16, 1667–1683. 10.1007/s12063-022-00344-x

  • CrossRef
  • Google Scholar

• 103

  SirajN. B.FayekA. R. (2019). Risk identification and common risks in construction: literature review and content analysis. J. Constr. Eng. Manag.145, 03119004. 10.1061/(asce)co.1943-7862.0001685

  • CrossRef
  • Google Scholar

• 104

  StahlB. C. (2021). “Ethical issues of AI,” in Artificial intelligence for a better future: an ecosystem perspective on the ethics of AI and emerging digital technologies (Cham: Springer International Publishing), 35–53.

  • Google Scholar

• 105

  StanovskyG.KeydarR.PerlG.HabbaE. (2025). Beyond benchmarks: on the false promise of AI regulation. arXiv.2501.15693. 10.48550/arXiv.2501.15693

  • CrossRef
  • Google Scholar

• 106

  SteimersA.SchneiderM. (2022). Sources of risk of AI systems. Int. J. Environ. Res. Public. Health19, 3641. 10.3390/ijerph19063641

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 107

  TaiwoR.BelloI.AbdulaiS.YussifA.SalamiB.SakaA.et al (2024). Generative AI in the construction industry: a state-of-the-art analysis.

  • Google Scholar

• 108

  TamN. (2025). How generative AI reshapes construction and built environment: the good, the bad, and the ugly. Build. Environ.284, 113526. 10.1016/j.buildenv.2025.113526

  • CrossRef
  • Google Scholar

• 109

  TianK.ZhuZ.MbachuJ.GhanbaripourA.MoorheadM. (2025). Artificial intelligence in risk management within the realm of construction projects: a bibliometric analysis and systematic literature review. J. Innov. Knowl.10, 100711. 10.1016/j.jik.2025.100711

  • CrossRef
  • Google Scholar

• 110

  van BreugelB.QianZ.SchaarM. (2023). “Synthetic data, real errors: how (not) to publish and use synthetic data,” in International conference on machine learning (Honolulu, HI:PMLR), 34793–34808.

  • Google Scholar
  • View reference in article

• 111

  WijayasekeraS.HussainS.PaudelA.PaudelB.SteenJ.SadiqR.et al (2022). Data analytics and artificial intelligence in the complex environment of megaprojects: implications for practitioners and project organizing theory. Proj. Manag. J.53, 485–500. 10.1177/87569728221114002

  • CrossRef
  • Google Scholar

• 112

  XuC.ChoS. E. (2025). Factors affecting Human–AI collaboration performances in financial sector: sustainable service development perspective. Sustainability17, 4335. 10.3390/su17104335

  • CrossRef
  • Google Scholar

• 113

  XueL.PangZ. (2022). Ethical governance of artificial intelligence: an integrated analytical framework. J. Digit. Econ.1, 44–52. 10.1016/j.jdec.2022.08.003

  • CrossRef
  • Google Scholar

• 114

  YagerR.ZadehL. (2012). An introduction to fuzzy logic applications in intelligent systems (Springer Science and Business Media).

  • Google Scholar

• 115

  YaoD.SotoB. (2024). Cyber risk assessment framework for the construction industry using machine learning techniques. Buildings14, 1561. 10.3390/buildings14061561

  • CrossRef
  • Google Scholar

• 116

  ZadehL. A. (1965). Fuzzy sets. Inf. Control8, 338–353. 10.1016/s0019-9958(65)90241-x

  • CrossRef
  • Google Scholar

• 117

  ZhangJ.ZhangZ. (2023). Ethics and governance of trustworthy medical artificial intelligence. BMC Med. Inf. Decis. Mak.23, 7. 10.1186/s12911-023-02103-9

  • CrossRef
  • Google Scholar

• 118

  ZhangP.ZhangZ. J.GongD. Q. (2024). An improved failure mode and effect analysis method for group decision-making in utility tunnels construction project risk evaluation. Reliab. Eng. Syst. Saf.244, 109943. 10.1016/j.ress.2024.109943

  • CrossRef
  • Google Scholar

• 119

  ZhuH.FengY.HwangB. G.FanX. (2025). Generative AI for sustainable development in the construction industry: exploring organizational adoption patterns. Sustain. Develop.10.1002/sd.70246

  • CrossRef
  • Google Scholar

• 120

  ZuoJ.ZhaoZ. (2014). Green building research–current status and future agenda: a review. Renew. Sustain. Energy Rev.30, 271–281. 10.1016/j.rser.2013.10.021

  • CrossRef
  • Google Scholar