The recent launch of ChatGPT thrust the conversation surrounding artificial intelligence forward. The impact of it on American industries is unclear, and even as the dust settles it still may not be obvious what ChatGPT — an artificial intelligence-based chatbot that can quickly generate text based off of prompts — can do for construction. 

Here, Construction Dive talks with Will Senner, vice president of preconstruction at Skanska USA Building, to get a better sense for the industry’s data evolution, AI’s role in construction and how contractors can be prepared to use it.

CONSTRUCTION DIVE: How could an AI tool like ChatGPT apply to construction?

WILL SENNER: The reality is that as part of our data journey, both at Skanska and within the industry, we have a process to go through around data literacy and building an understanding of data tools and how those can be integrated into the work that we do. Construction is an industry that has relied on people's experience and intuition. So, we're doing a lot of work in sort of a foundational capacity in thinking about how we develop our people's ability to think differently about how they integrate data into the work they do every day and make more data-informed decisions.

We're not creating a future where AI is going to take everyone's job. So just building that level of understanding and awareness, I think, is really important to getting people to a point where we can have meaningful conversations about what tools like ChatGPT can actually do for the business.

What does that process look like?

About a year ago, we went through a pretty structured and rigorous process to think about our pipeline of data use cases, AI- and non-AI related. We went through interviews with like 20 different persona groups throughout the company — different representatives from our project teams, our operational leaders, HR, pre-construction — to get their perspective on how they can use data and AI to improve their work. 

Through those workshops, we generated a hundred different use case opportunities. Then we worked with a consultant to come up with a framework for how we evaluate those use cases in terms of technical feasibility and speed-to-value. 

What's been really interesting is, I think we're more focused on use cases around structured data because it was more easily accessible. And then, ChatGPT sort of opened up this Pandora's box of natural language processing, which has forced us to go back and reevaluate some of our prioritization because there are a few use cases that will really benefit from enhanced natural language processing and capability. 

So, there's ChatGPT and other personal assistant types of AI. Microsoft is integrating similar technology into Word, PowerPoint and Excel. So, you have those kinds of tools that are going to be embedded in our native applications.

Then there are more customized tools that we're looking to develop, which would be more proprietary solutions. So using the same kind of natural language technology, but training it not on the data on the internet, but training it on Skanska-specific data.

Is natural language processing AI like a search engine for one’s own data?

I think that's not a bad way to think about it. It's probably a little bit of an oversimplification, because of the sheer power of some of these tools. But it does work as a super-enhanced search engine. I mean, another great example is thinking about safety planning, right? We have access to thousands and thousands of daily hazard analysis forms that each crew is filling out. What are the risks today? How are they mitigating it? 

Right now, those are things that are often done completely from scratch everyday. It's done manually, it's done electronically, but it's filled out every day by hand. And you think about the ability to layer a tool like ChatGPT, that natural language processing, over the top of all of that historical data and be able to then give team suggestions. "Oh, I see you're working on this. Have you thought about how this might be a potential risk?"

The ability to pull in those other insights both from historical project data, but also your current project data, unlocks a lot of really interesting opportunities to just improve existing processes and make our teams more efficient.