On March 25^th, 2021, Brett Young presented to AI4AECO and discussed his personal tech journey with artificial intelligence (AI) and machine learning (ML).  Brett is a former construction project manager, a BIM consultant, and a current software developer.  As the CTO of M2x.ai, he helps realize project delivery as a product configuration experience, creating automation for pre-manufactured timber panels, complex conference room audio-visual equipment, and MEPS in a “network-first” approach.  This presentation covered a series of proof-of-concept and client work in the area of AI and the lessons that were learned.

Brett related that his personal tech journey with AI began with auto-routing piping through LIDAR data. The algorithms used at the time fell into the scholastic definition of AI but were often met with confusion when described as AI.  This led to the insight that AI is loosely defined on the best day, and the AEC community has yet to find great ways to define how AI distinguishes itself from algorithm-powered automation.

AI can be defined in a way that’s similar to how data mining is distinguished from data analysis.  Data analysis involves creating a question, often using a structured query language (SQL), which then provides answers.  Data mining is different because it seeks correlation without knowing the question.  Brett illustrated this using an example of the discovery of a sales relationship between diapers and beer.  Similar to data mining, AI is the creation of algorithms without writing algorithms.

Brett then discussed how historical CAD files could be processed and refactored, which resulted in networks of parts that could be used as training sets for ML, and how Amazon Web Services and their AI tools can be used to gain an understanding of PDF files.  In both cases, algorithms were sufficient to perform the required task, and AI / ML was not needed.

In Brett’s opinion, the biggest opportunity is to apply AI and ML to tasks like RFIs and design checklists.  The project deliverable design process is loosely defined, which results in a need to search for omissions, latent defects, and construction process improvements as the construction team are engaged.  AI and ML can be used to make correlations between drawings and models with documents like RFI.  In this way, predictive analytics and inference engines can be created to improve drawing and model quality by identifying deficiencies rather than discovering them through the RFI process.

Other interesting aspects of Brett’s presentation:

• Brett meets interesting AI / ML – focused people on the internet.
• General and specialty contractors may be in the best position to capitalize on AI / ML because they have lots of high-detail data.
• Brett is kept awake at night by the concern about communicating complex topics, like AI in construction.
• MEP coordination and installation is an example of a multi-parameter optimization problem. All optimization requires a cost function, and the cost function definition is the art of applying ML and AI to the AEC industry.  As an industry, we know these cost functions in our heads, but we haven’t written them down or defined them concisely.