Wicked problems in construction: how the intelligent use of AI can help

How can artificial intelligence benefit the construction industry? In first of this three-part series, four industry experts examine the considerations necessary for construction to capitalise on the benefits of AI.

Anil Sawhney FRICS, Head of Sustainability, RICS
Mike Hill, Chief Digital Information Officer, RICS
Jugal Makwana, Senior Executive – Industry Transformation and Strategic/Open Ecosystem, Autodesk
James Garner FRICS, Global Head of Data, Insights and Analytics, Gleeds

The construction industry faces complex and multifaceted (or ‘wicked’) problems that are, at times, difficult to define, let alone solve. These include skills shortages, waste, emissions and low productivity. As the sector evolves with technological advancements, the ‘intelligent’ use of artificial intelligence (AI) presents a promising pathway to addressing these formidable issues.

The rapid expansion of technological capabilities is outpacing the regulation and understanding about how these technologies should be used. It is easy to perceive artificial intelligence (AI) as a universal remedy for the construction industry's hurdles.

From mundane and repetitive tasks to the most complex and tedious (such as providing insights by combining large volumes of emails, videos, photos and sensor feeds) – we can outsource to an intelligent machine, but this approach is simplistic. The key lies in leveraging AI ‘intelligently’ rather than indiscriminately deploying it.

Intelligent use of AI is the strategic utilisation of AI to solve specific, well-selected problems efficiently (completing tasks more quickly, with fewer resources, and in fewer steps) and effectively (doing the right things in the right way to deliver value). The intelligent application of AI considers using all AI models, including generative AI (GenAI) frameworks, large language models (LLMs) and computer vision models.

This article discusses guidelines for using AI intelligently and how AI can help tackle wicked problems in the construction industry and pave the way for enhanced productivity and innovation.

Defining AI

AI is a topic of growing interest among industry professionals. However, the term AI and related concepts are often used interchangeably, which can create confusion. To provide clarity, see Table 1 for clear definitions relevant to this topic that will be consistently used throughout this series of articles.

What problems should AI solve?

Complexity, additionality and data availability can determine whether a task fits an AI solution.

• Complexity: Complex tasks, which require working with numerous diverse concepts and sources of information, are good matches for AI solutions. For example, these sources of information could include recording timesheets, weather conditions, or daily logs.
• Additionality: Delegating a problem to AI adds value only if it significantly outperforms human capabilities in terms of efficiency and effectiveness. If humans can solve it just as efficiently, there is no advantage in utilising AI.
• Data availability: AI solutions are most effective when there is a large body of good-quality data to draw on. AI may struggle to perform if the volume of relevant data is low, and this is where human judgement and expertise could prove more reliable.
   

The best types of problems to be solved by AI will be high in all three of these dimensions. These factors are visualised as dimensions in a 3D space in Figure 1. If the value of any dimension in this space is low for a given use case, then it may not be suitable for an AI solution, even if the values of the other dimensions are high.

Generally, AI should be used because it genuinely adds value, not merely because the tool is available or novel. For instance, tasks that must be repeated frequently and are process-driven (rather than human thought-driven) may be well-suited to Robotic Process Automation (an automation process that is strictly workflow-driven without any emergent behaviour used to perform repetitive tasks), but this is not to be mistaken for AI.

How is AI currently being used in construction?

Two well-accepted examples illustrate the potential for AI in construction:

1. AI is being used to review progress and ensure safety on construction sites. This task involves analysing a large corpus of data comprising images, videos and laser scans through computer vision models. A recent study estimates that 72% of construction leaders say these AI models are critical for preventing accidents, and 40% of construction leaders believe AI improves worker safety.
2. AI is being used to analyse, compare and choose design options based on performance outcomes and generate a final design because of this analysis.
    

Both these tasks score highly on all three dimensions shown in Figure 1:

• the amount of data to be analysed is large
• these data and underlying concepts take different forms and the issues are complex and
• due to the volume and complexity of the data, humans would likely find these tasks difficult, meaning the problems are high in additionality as well.
   

While the above-cited advancements have established a strong foundation for using AI in construction, more recently, discussions primarily focus on GenAI. GenAI can improve workflows, decision-making and project and asset outcomes. As shown in a recent report, GenAI holds immense promise for the construction industry, with 78% of industry leaders agreeing that AI will enhance the industry and 66% agreeing that in two to three years, AI will be essential across the board.

Retrieval Augmented Generation: an application framework

The next frontier involves developing company-wide AI applications by integrating construction sector knowledge, organisational data and project information into general-purpose or fine-tuned generative pre-trained transformer (GPT) models. These models aim to address complex problems and enhance operational efficiency. Figure 2 presents a simplified, high-level application framework that uses Retrieval Augmented Generation (RAG) for construction use cases.

RAG combines retrieval and generation to enhance AI responses. It retrieves relevant information from a database, often a company-specific knowledge base, by encoding and matching it with the user’s query. A GenAI model then generates responses using the retrieved data as context along with the prompt. This ensures that outputs are accurate, domain-specific, and aligned with the company’s knowledge and understanding.

The proposed framework can be enhanced further by integrating agentic workflows and compound AI systems.

Agentic workflows involve deploying AI agents to autonomously manage specific tasks, such as monitoring inventory or tracking the status of long-lead items. For example, an AI assistant could check the status of a delivery weekly, interpret emails and messages, and alert the project manager if issues arise. These workflows streamline routine tasks, reduce manual effort and improve decision-making by providing timely updates and alerts. For the foreseeable future, agentic workflows will still operate with a ‘human-in-the-loop’ to ensure accuracy and effective decision-making.

Compound AI systems combine models, encoders, retrievers, external tools and an orchestrator into an integrated system. Together, they enable seamless data interaction and task execution, enhancing efficiency and allowing professionals to focus on strategic, high-value activities.

How can companies capitalise on their investment in AI?

AI offers exciting possibilities, but its success depends on more than individual enablement; it requires a cohesive organisational strategy. While enhancing employees' personal productivity and skills through tailored AI access, ranging from restricted use to open access, is vital, these efforts must align with a broader strategic vision. Without a well-defined organisational strategy supported by clear use cases, a structured data ecosystem, and a thorough cost-benefit analysis, companies risk adopting AI solutions that fail to deliver value or even detract from it (for example, an LLM-enabled app that provides vague or unhelpful answers). As medium to large construction organisations increasingly develop AI strategies, the focus must shift to creating frameworks that integrate individual enablement with organisational objectives.

The next article in this series explores the types of complex, 'wicked' problems that AI is capable of solving. It highlights how these challenges are currently being tackled, and how they could be further addressed in the construction industry.