The year of AI exploration

AI NEEDS PEOPLE
If you are someone who is worried about machines taking over for humans, don’t expect to see AI causing layoffs in construction any time soon. The data is too messy for people to lose their jobs, and humans are needed. It’s definitely going to be a scenario of AI augmenting what people do, not replacing them. (PHOTO: PHONLAMAIPHOTO / ISTOCK / GETTY IMAGES PLUS /GETTY IMAGES)

Last year was a year of first steps in the investigation and use of Artificial Intelligence (AI) in construction. There have been many marketing releases and statements about AI, but it’s very much in the early stages when it comes to construction companies incorporating AI into products or business practices.

What we did see in 2023 was many companies starting to announce their strategies, plans and beliefs about the use of AI. And we saw some early adopters using AI in their planning departments to evaluate historical data, including using the technology to explore what went wrong and how long certain initiatives took.

With AI still in the “new and evolving” stage for construction, it is difficult to predict the specifics of how software companies will provide AI for the industry, but the intent is certainly there. It’s been somewhat of a boom for the rise of AI startups in the construction space, as well as related domains such as site capture, planning, tendering and contract review.

We also saw many small teams form companies to tackle these fields, and we have seen more and more people using ChatGPT for search and quick research but, in reality, we saw more marketing than traction. We’ll need to see real-world data sets and use cases to move AI forward at a faster pace.

Big-name vendors are, of course, working on AI products, but seeing the number of AI startups formed to service the construction industry has been fascinating to watch. This growth has been catalyzed by the number of construction companies projected to use AI to improve their business practices. Most construction companies aren’t developing AI products in-house, so they will look outside for AI vendors.

And the impact of language models, such as OpenAI’s ChatGPT, Google Bard and Microsoft Bing should not be understated.

The move from closed providers and models, for example, ChatGPT and Google Bard, to Meta’s pioneering of LLAMA/ LLAMA2, has driven a lot of innovation in the space. With closed models, you don’t know how it works or how it was trained. Further, there are limited ways to extend their abilities.

LANGUAGE MODELS AND DATA QUALITY

As the technology has advanced, we saw that RAG, or resource augmented generation, was essential for enhancing language models such as ChatGPT with document data. If you want to ask a question about financials, you can use a language model to quickly pull out the answer with a semantic search to your question instead of scouring through multiple records.

Data quality is vital to corporate success, as researchers have found that language models can be significantly more accurate and cost-effective when the training data is curated correctly. We’ve heard a lot about having big data, which is good, but ultimately it comes down to building proper data pipelines to have robust data quality with minimal errors.

And cross training may improve the ability of AI, as the use of different data sources allows models to become multi-modal to be able to “read” and “see.”

Currently within the construction sector, AI mainly lives in the IT departments of contracting companies. They are the custodians of the data infrastructure and, therefore, the best place to try to find applications of AI that will correlation with the data they have. This is good and bad as it provides oversight and control but also limits user capability – at least for now.

The construction site has seen limited adoption, mainly for information capture and with field devices to drive productivity information. We also saw some progressive companies use a bit of AI for planning and analyzing historical data to help with accounting, procurement, reporting, fraud detection, and transactions, however this is in the early stages.

There are some promising developments on the horizon though, which may open the door to more advanced use in this sector.

To start, combining RAG with language models and tools with semantic search and external data sources will make it much faster to conduct document reviews, contract reviews, tendering, etc.

TALK TO MY AGENT

Agents, which are a computer program, system, app or bot that can interpret the environment to take an intended action, have a conversational interface so that you can feed in text to predict subsequent text. With agents, we can augment language models with tools, such as a calculator, to answer questions and enable users and non-users of programs to ask questions in natural language to get answers. We won’t see this in full force this year, but we should see the start of this trend.

Also expect to see the democratization of machine learning to advance this year. We should see programs such as Excel and Power BI incorporating AI to enable users to take historical data, tools, and BI apps to perform forecasts for future projects. This will allow them to use large amounts of well-structured data to improve business decisions.

TRENDS WITH BENEFITS

There are two primary reasons why construction professionals might want to pay attention to these trends.

The first is to improve productivity. Construction has tight resource limitations with very thin margins and people in construction are often overworked. This can include spending hours looking for crucial information in existing data and documentation. RAG and agents will allow people to be more productive and save time when seeking information within documentation.

The second is that democratization will move AI from IT departments to users. This is critical as IT is experienced in managing the data infrastructure but not in using and interpreting the data.

Democratization gives the data and AI to the actual users and true subject-matter experts, empowering them and enhancing data-driven decision-making during daily tasks. We should also expect to see planners, site supervisors, and various other players in the industry begin to use new tools in conjunction with AI to augment their AI experience.

This will all require guardrails to ensure that confidential data is protected, stays within the boundaries of corporate governance, and meets government regulations and guidelines. There will be friction and learning along the way, as well as training and upskilling of people.

As AI matures, we will see planners using AI-driven tools to analyze historical data and rich, current data. We should also see some site supervisors and workers adding more field devices like cameras and sensors to help capture more of that loop for AI usage and improvements in safety and progress tracking.

This past year, we saw the groundwork for AI being laid throughout construction and entering the industry’s consciousness. AI in the building world will take off in both usage and further development, but with that, let us remember that we are seeing quite a bit of “AI hype” right now when we need to see less fanfare and more problem solving.

Complicating matters is that some AI vendors tend to keep their development projects under wraps, so it can all be a bit confusing as the AI market is being developed and sorted out. Many startups that focus on AI in construction are in the works, and we should see many of these new ideas stick in the office and on job sites.

Many construction companies have included AI as part of their business strategy, and we’ll see the transition occur from a strategic level goal to trials and experimentation, or what I call “AI Exploration.”

Daniel Hewson is the data capability manager at Elecosoft, a company that helps its customers implement technical innovations. He oversees development of overall data and AI strategy to focus on how AI can be leveraged to improve project planning and to identify inherent project risk.