Some say that AEC is slow to innovate, but it only takes one talk with a company like BuildingSP to see how tired that idea is.
CEO Brett Young describes BuildingSP as a software development firm that “uses technology and computation to improve the building information modeling (BIM) process, which improves project outcomes through better planning.” That’s underselling what they do, because the company is already forging a path toward a truly advanced artificial intelligence for AEC, and all the efficiency gains that entails.
Software that Learns and “Narrow AI”
BuildingSP started on the path toward AI with their first release, GenMEP. This add-on for Revit asks users to define the start and end point of an MEP system, and then routes it automatically through the model.
Young says the application uses heuristic algorithms. If you think of classic, hard-computer-science algorithms as a set of processing instructions that remain the same every time the program is executed, a heuristic algorithm differs in that it accounts for the preferences of the person using the program.
“For example,” Young explains, “if you’re looking at a conduit route through a building, the rules for how it’s supposed to be routed are definitely important. You also have preferences for how you want it to be routed, or how you would optimize it yourself. Heuristic algorithms try and pick up both preferences and rules as part of how they work.”
BuildingSP’s latest product, ClashMEP, is even more sophisticated. As you model your conduit in Revit, the ClashMEP plugin recognizes clashes dynamically. Next, it uses machine learning to sort those clashes by priority, determining which ones are unimportant, which are more important, and which are important enough to warrant your immediate action.
Young told me to think of Netflix’s movie recommendation algorithm, which produces a list of movies you might be interested in watching: ClashMEP does the same thing with clashes. And, much like Netflix’s algorithms, Young says that ClashMEP learns from its users. “The user input is part of a feedback loop for machine learning,” he says. “Like the Netflix recommender, it gets better over time as you use it.”
If it sounds like BuildingSP has already reached the goal of building artificial intelligence for AEC, it has—to an extent. Young told me that these programs constitute a “narrow AI, or vertically focused AI.” In other words, they are AI developed for a single, specialized task.
I asked Young, What’s stopping his company from developing more sophisticated AI?
How to Get AI in AEC
“If you think about it,” Young says, “AI and machine learning are most commonly used in the internet space, where a lot of data is generated for training. In construction, that’s flipped. In construction, we don’t have enough automation yet to produce reliable data to inform AI and machine learning.”
BuildingSP is attacking this problem by focusing on automation first, and then using those automated processes to generate the data they need to train AI. This approach is reflected clearly in ClashMEP, which performs automatic clash detection, and uses all the clash data it generates to train the system that determines clash priority.
Young believes this automation-first approach is key to developing much more advanced computer cognition in the future. “We absolutely intend to create more sophisticated AI for AEC,” he says. “Our strength, as a firm, is an ability to package tech so that it solves a problem. Advanced AI is a part of where we’re going.”
Closing up our interview, I asked Young what that truly sophisticated AI for AEC might look like.
“It will look like advanced automation,” he said. “It will feel like you have a large studio or engineering office full of people working on your project at insane speeds. For example, think of Mad Men and the row upon row of people at typewriters, sending out the correspondence to create ads across the country. Today, we use very sophisticated AI to do this, instantaneously, across the country. It will be that paradigm — shifted to AEC.”