There’s an information vacuum when it comes to applications of artificial intelligence in the construction industry. The possibilities are seemingly endless, but the resources demanded to implement new technology and the unknowns surrounding its downstream effects both hamper efforts to move forward.
AXA XL’s Innovator’s Circle -- a group of construction clients and industry leaders – met recently to discuss these issues, including current use cases for AI and best practices. The meeting was kicked of with an overview of , owner of tech startup Placer Solutions, in which more than 100 construction professionals weighed in on the opportunities and risks presented by AI.
The report also includes a use case prioritization matrix, based on level of effort and level of impact on an average construction business.
Top Use Cases in Construction
On the lower effort end of the scale, use cases fell into four buckets:
- Data Management
Fuller highlighted the potential of AI to transition “knowledge workers” into “knowledgeable workers,” pointing out that the ability to use information to advance project goals is a more valuable skill in an AI-enabled world than possessing the information or knowledge itself.
To that end, the more that AI can take over rote, manual tasks around data collection, organization and analysis, the more that human workers can apply their expertise in putting that data to work in meaningful ways. Making data useful takes experience, knowledge, and skill that AI is not yet able to emulate.
- Communication:
Text-based AI tools like Chat GPT can facilitate communication within project teams to keep things moving forward. AI tools could be used to capture action items and main ideas from calls, producing and disseminating a report at the end of a meeting to keep all parties on the same page and clarify goals and responsibilities. AI can also help teams learn various skills that help them in their daily jobs, one example was a project manager who used AI to learn how to write DAX Measures to filter and sort information from project management software to send to stakeholders.
More broadly, AI could also help with drafting of communications in general, easing workload for staff and potentially helping to improve the tone of communications.
- Training
Creating learning curriculums is a common and straightforward application of AI. This can be used to onboard new hires as well as provide upskilling opportunities for current employees. AI can also generate safety talks and help to ensure compliance with safety regulations.
- Marketing
Generative AI can take some of the lift from various teams. According to one group member, AI in one instance produced a 72% boost in efficiency in terms of cost per proposal for their marketing team.
Other similar use cases for AI involve tasks directly related to preconstruction and project management, such as bid generation and drafting of daily logs.
In both cases, AI certainly has the ability to collect all the data needed to put together these documents, but these tasks still require domain expertise and organization-specific knowledge, such as familiarity with a company’s strategy, goals and priorities. In these instances, AI may not be reliable enough to produce a final draft, but can many times get the preparatory work done faster.
At best, outputs produced by AI need refinement, especially more training specific to the construction industry, in order to create more accurate and reliable results. At worst, they open companies up to new liability risks due to lack of consistent rules and governance.
Risks and Roadblocks Presented by AI
Meeting attendees all expressed reservations about the unknowns attached to artificial intelligence. The tools and applications out there are relatively new and untested. At best, outputs produced by AI need refinement, especially more training specific to the construction industry, in order to create more accurate and reliable results. At worst, they open companies up to new liability risks due to lack of consistent rules and governance.
Top challenges discussed include:
- Data security and privacy
AI models learn and improve through the digestion of vast amounts of shared data. Companies may use proprietary data to train their AI models, which in turn generates further confidential or sensitive information. Additionally, most companies will rely on vendors to help with development of AI models, raising questions around the ownership, storage and protection of data as it moves between parties. In this setting, how can private data be protected? Ensuring security will demand investment in IT expertise and infrastructure.
- Specialized knowledge
AI tends to be better at general knowledge than specialized expertise. Training a large model to understand construction terms is time-consuming and costly. There is broad concern that outputs may be skewed by models not understanding construction vernacular.
- Reliability
AI models are not always accurate or reliable, in part because, in addition to internal data, they also draw information from external sources on the Internet. They may generate content that is irrelevant, misleading, or even harmful. Organizations need some way of knowing where data originated in order to discern the likelihood of inaccuracies, but methods for accomplishing this are not exactly clear. In general, there is a lack of standard benchmarks and best practices. AI is a new and evolving technology that does not have clear metrics or guidelines for measuring its performance and value.
For example, over the course of the call, meeting attendees discussed Chat GPT multiple times, while an AI voice-to-text application transcribed. It heard and interpreted “Chat GPT” in several different ways, including “cheat sheet PPT,” “Chat peachy tea,” and “Chachi party.” This shows how even the most basic AI applications need refinement and review by human eyes.
- Legal Risks
There is a lot of uncertainty around how documents generated by AI will be viewed in court. Will reports, daily logs, and other communications created by AI be viewed as admissible evidence, given the above concerns with reliability? Other concerns center on professional liability. If decisions are made based on AI-generated output, which then result in a defect or bodily harm, who is to blame? Meeting attendees compared this scenario to the introduction of BIM models, which were initially not seen as reliable as a stamped drawing. In that instance, there was a parallel path during which all the usual instruments of service were in place alongside the BIM model, and the BIM process was viewed as for information only for a time. AI may take a similar path.
If one of the benefits of AI is taking some mundane work off the plates of construction professionals, the payoff may take some time, as we develop confidence and comfort with the results it produces.
Best Practices
Going forward, the construction industry will need clear protocols around use of AI in order to balance the risks and benefits. Early preparation and responsible governance are key to leveraging A.I. as a strategic advantage in construction. Some best practices discussed by meeting attendees included:
- Treating AI as a tool that augments human expertise, not replaces it. While text based A.I. can automate mundane tasks, the output still requires human validation.
- Keep both internal and external users educated on the risks and opportunities of AI. Everyone must be aware of what is considered a safe and appropriate use of AI tools. Outline clear protocols for acceptable use – which may be extensions of current internet and social media policies. Investing in expert support to set up secure, customized internal applications can be a game-changer.
- Firms must focus on communication, governance, and strategy. Once the guidelines are established, keep your teams updated through multi-media educational initiatives.
- Keep lines of communication and collaboration open. Invite team members to share their experiences, insights and ideas on how AI can enhance their processes and improve outcomes. Generative AI is highly accessible and easy to use. If people don’t have somewhere to take their ideas within the bounds of your policy, they may go rogue, and proprietary information might be shared, or due diligence not done.
While use cases for AI are just taking shape, it is important that construction companies jump in now, and dedicate resources to better understand how AI can be meaningful and useful for their business without creating unmanageable risk.
Cheri Hanes is Head of Innovation and Sustainability for AXA XL’s North America Construction insurance business. She can be reached at cheri.hanes@axaxl.com.
Damon Ranieri is Construction Innovation and Sustainability Partner for AXA XL’s North America Construction insurance business. He can be reached at damon.ranieri@axaxl.com.
