Innovation Corner:
Top Three Lessons of Machine Learning
Syska Hennessy’s machine learning (ML) initiative started a little over a year ago, and in that time we’ve learned a lot about how AEC data has the potential to change how we approach our work. Below we present the top three lessons we learned through this initiative. But first, a definition:
What is ML?
Machine learning is a complicated topic. It has evolved from a long history of computing with unclear boundaries. Because of this progression, ML can defy easy definition. At Syska Hennessy, we define ML in as simple a way as possible. Here’s our definition:
Machine learning is the creation of automation using non-prescriptive methods from data.
We believe the goal of machine learning is always automation, ideally taking a task off a person’s plate and doing it for them. Machine learning that doesn’t make the user’s life easier isn’t worth it. Data is crucial for machine learning because the data is used to create the automation, replacing the need for humans to write complex procedures or code.
Machine learning is the creation of automation using non-prescriptive methods from data.
ML and Sources of Data
It’s really convenient and easy to say “ML = DATA,” but what impact does this connection have on an initiative? How should a data strategy change corporate behavior? What are the important attributes of data for machine learning?
At Syska Hennessy, we believe our data and records are our competitive advantage. If ML tools are created from data, then our competitive advantage grows when our own data is used to create those tools. Similarly, our competitors are at a disadvantage if they don’t have access to our data. Therefore, the source of data used in ML is an important criterion for evaluating any future ML tools. If a vendor offers an ML tool to Syska Hennessy and it was created using a data set common to everyone, it offers limited competitive advantage. If the vendor’s ML tool learns from our data, then we have a whole different ball game.
ML on the Solution-Problem Spectrum
In AEC, there have been many instances where we describe a technology as a “solution looking for a problem.” On the other end of the spectrum, there are problems that appear to have no solution other than adding more people, money, and time. When Syska Hennessy started our exploration of machine learning, we were closer to the solution side of the spectrum. Given our efforts since then, we’re now closer to the middle and much better at answering questions like, “How can ML help solve this problem?”
The most valuable asset at any AEC firm is its people, so they are the ones who should benefit from innovation and technology.
The Technology Problem Solutions
The Importance of People and ML
Syska Hennessy started our corporate innovation program with a distinct focus on people rather than on technology evaluation. The most valuable asset at any AEC firm is its people, so they are the ones who should benefit from innovation and technology. They can apply the resulting tools to their daily workflow and remove mundane tasks that eat into the more enjoyable parts of designing. People are critical to the success of ML in many ways, so we can measure progress by examining engagement with the technology. We’ve found that ML is being referenced on more and more internal innovation concepts, the quality of references to ML are improving, opportunities to learn and develop ML skills are oversubscribed by staff, and we see only growth in Syska Hennessy’s ML initiatives.
In summary, we still have a lot to learn about machine learning in the AEC industry, but we can also confidently say that we’ve learned more about machine learning in the last year than we have about any other emerging technology in AEC. We know that the source of data is paramount, that ML is most effective when it addresses a specific problem, and when it improves the workflow for our people. As we continue to roll out new ML-based techologies, we’ll report on what we’re discovering. Stay tuned; there’s more to come.