To Use or Not to Use

by Stacy Rinella | December 9, 2024 7:00 am

[1]

Artificial intelligence, or AI, is moving fast—faster, in fact, than any technology in recent memory. It promises to make work faster, more efficient, and more profitable for contractors while reducing some of the tedious paperwork and information searches that are a part of every project. However, along with that promise come many unknowns and potential risks.

For this reason, associations such as the Sheet Metal and Air Conditioning Contractors’ National Association (SMACNA) are taking a leadership role in figuring out how best to use or not use AI in various situations.

It has been said that the role of associations is to connect industry with the audiences they serve, including the market, government, and others, as well as facilitate, collaborate, and educate each other when it comes to solving societal problems. No one can doubt that AI has the potential to be one of the most challenging advancements the building industry has faced in decades, so it makes sense that SMACNA has convened a group and engaged outside experts to study the issues raised by AI. This group is charged with proposing policies, recommending training, and other guidance for the HVAC industry. This guidance largely highlights where AI may provide quick wins, such as HVAC layout, quantity takeoff for estimating, or running through shop configurations to find the most efficient workflow. A focus is always placed on known limitations within the technology, but how, even in system design and layout, the tool can give the user a good starting point.

[2]

 

A new wave of AI

AI, in one form or another, has been around for decades, but until recently, most AI products were not very useful outside of technical circles. Then, in November 2022, everything changed with the introduction of ChatGPT and “Generative AI,” a new type of AI that is easy for everyday consumers and professionals to use. ChatGPT and its more recent competitors, like Anthropic’s Claude and Google’s Gemini, allow users to ask everyday English questions and get sensible, often useful responses.

Users can ask it to draft an email, create a marketing plan, or even devise a weight loss plan. The results are genuinely helpful, even if not always 100 percent on the mark. For many, it has been almost a shock that software can now produce work once thought only humans could create. This capability is where all the excitement has emerged.

 

Limits emerge

AI is not the first technology that seemed like it would change everything and later was taken down a peg. In the late 1990s, there was huge enthusiasm around the World Wide Web, but by 2000, its limitations became evident with the dot-com crash. This pattern occurs frequently, as initial excitement often tempers once a technology is put to real-world use.

While AI is not in danger of going anywhere, 2024 has been a year of questioning whether, how, or when AI will provide the most tangible business value and what that value might be. Beyond writing an email, are there other uses of the technology or improvements to the technology that will drive more value for more professionals? While the answer is almost certainly yes, it is also true that enthusiasm has toned down, and serious use cases are being developed and explored. For example, AI is expected to improve administrative and information retrieval activities that support and surround putting work in place. Indeed, AI has proven very good at document management, such as contract risk review, specifications management, and even drafting RFIs. Other areas where AI has shown promise include computer vision, where tasks related to safety, quality assurance (QA)/quality control (QC), and perhaps even live tracking of building progress may begin to be automated.

One issue that has slowed the development of these use cases is the tendency of AI, especially chatbot-based “Generative AI,” or GenAI, to produce outputs that are not completely true. Known as “hallucinations,” these false statements usually come wrapped in perfect English and expert-level logic, so they can be hard to spot. While better products and better models are reducing these issues, they nevertheless present a real risk to professional users.

 

Cautious optimism

The risks and possibilities drove SMACNA’s Construction Technology Committee to convene an AI Leadership Project. Only by properly studying, understanding, and defining the risks of AI use can SMACNA and its members trust the technology in those areas where it is not as risky or where the use cases allow for oversight and risk management. Another outcome of the convening was an engagement with an AI specialist firm, The Link.ai, co-author of this article, along with SMACNA.

For example, almost every video conferencing service has an AI “summary” assistant. These tools listen to the conversation, transcribe it, and summarize it. The problem is that even if the AI were perfect at summarizing, these tools have no context and no awareness of the past interactions of the video call participants or the projects they are discussing. Add to this the very real issue of hallucinations, and the potential for highly misleading summaries is unavoidable.

And yet…the summaries produced to save time often catch things that participants might miss in their notes and can generally improve the quality of meetings.

An association like SMACNA can surely develop a more nuanced policy rather than banning AI assistants. And that is precisely what this task force is doing.

 

Current use examples

Both within SMACNA, and in early surveys of industry, some early uses are becoming common. These uses have one thing in common: AI is not yet being used for “mission critical” tasks, where the risk of an uncaught error to the project, or the organization is significant. Members are finding that there is significant value in using AI to jumpstart processes, summarize them, and otherwise move communication along more efficiently. Specific examples include:

• Meeting Agendas—It can be much faster to feed in a prior meeting’s agenda, have AI update it, and then give it a quick edit.
• Presentation Outlines—Provide just a sentence or two about the presentation, and AI can provide a high-quality outline that speeds up the process and ensures nothing gets missed.
• Job Descriptions—These are critical, and almost everyone gives them less attention than they need. Having AI do the first draft can ensure a more complete, thought-through job description. AI can also critique an existing JD.
• Software help—Modern software has so much functionality that just being able to ask for something in English instead of a series of canned questions is huge.

 

Fast approaching future

Perhaps the most important motivation and outcome of the SMACNA AI Leadership Project will be elevating the discussion from individual conversations to one that includes the entire industry. This can be done in several ways, including through training, expert webinars, policy papers, and curated talks at events.

As AI continues to develop at breakneck speed, the industry must continue to discuss, test, and intentionally decide how to leverage this new technology to benefit workers, customers, and the industry. Only by dedicating time and effort to examine these issues thoroughly can society avoid some of the more negative scenarios often highlighted by scaremongers. Only through collaboration can people ensure that AI enhances work and improves quality of life.

 

Travis Voss is SMACNA’s director of innovative technology and fabrication. He is also a co-host of The Construction Dorks podcast and serves his local community as a volunteer firefighter.

 

Hugh Seaton is CEO of thelink.ai, a construction software company dedicated to making construction documents consumable for the field through innovative use of large language models (LLMs) and deep knowledge of specifications. Before The Link, Seaton was a general manager at the Construction Specifications Institute (CSI).

Source URL: https://www.metalarchitecture.com/articles/to-use-or-not-to-use/

---