• Matt Edwards

Machine Learning in Construction, what it means for the industry.

Updated: Nov 28, 2018

I was recently invited to talk on a panel of experts on the topic of Machine Learning and what it means for the construction industry. As the 'construction guy' in our team I was surprised to be asked talk about such a 'techie' topic, but then I figured who better to explain machine learning to a room of builders than a non techie?

It was an interesting panel with folk from Procore (who are just kicking off their journey into Machine Learning) Smartvid (who have an excellent photo/video tagging product) and me, the co-founder of Nyfty.ai, an AI project assistant that we launched at the event (think Alexa for Procore).

Co-Founder and CEO Matt Edwards talking all things machine learning at Procore's Groundbreak 2018

Defining Machine Learning


How do you define Machine Learning in the context of your role and company?

At Nyfty.ai we work with Natural Language Understanding (NLU) which in construction has very particular challenges, i.e. in construction we have our own language and way of saying things that varies in each state and even on each project. So for us Machine Learning is about the system constantly improving with each user interaction - without us doing it manually. Each conversation, good or bad, contributes to the system's overall understanding of construction language.


How does Machine Learning differ from Artificial Intelligence?

AI (Artificial Intelligence) is broadly defined as the simulation of human intelligence, and can range from simply mimicking cognitive functions, like image recognition through to complex human like interactions like Hal9000 in my favourite film. AI was coined as a field of research in 1956, but the idea of 'thinking machines' has been around for centuries. ML (Machine Learning) on the other hand implies a type program that can loosely be described as 'improving without software developer intervention'.

Machine learning doesn't necessarily come in the form of a computer like HAL9000 (Open the pod bay door HAL), for instance my first interaction with machine learning was about 20 years ago when I worked in Building Automation for large scale construction projects. I setup a 'home optimiser' in my house that used machine learning to figure out what time my heating should start in the morning to ensure I was nice and toastie when I got in the shower at 7am. It was a basic setup using internal temperature, and whilst for the first few days it missed the target (either too early or too late) after a week or two it had 'optimised' its algorithm to perfection, i.e. the machine learned without any intervention by me.

Imagine if I had tried to achieve that without machine learning? I would have to measure the thermal dynamics of the house to an incredible degree of accuracy, understand gains and loses throughout the year, and write hundreds of lines of code to compensate for every variant. Practically speaking, I wouldn't have been able to do it.

So you can think of Machine Learning as an amazing way to do incredibly smart things, without having to actually be that smart.

Machine Learning is synonymous with Artificial Intelligence because simulating human intelligence, even at a low level, is so complex its generally accepted that machine learning has to be used, because to do so manually (writing every line of code for every scenario) would be monumentally difficult.


Why now? What’s been the catalyst for bringing Machine Learning to Construction?

In my mind the shift away from paper is the single most powerful catalyst. Yes, cloud computing is a factor, but machines can't learn without data, and paper is where data goes to die. The pace of digitisation in construction is dramatically increasing, with companies like Procore and their partners in the marketplace at the forefront of that shift. And just to be clear, when I say shift away from paper, I don't mean to PDF - as generally speaking the data is still inaccessible.

The shift away from paper (and PDF) means that a treasure trove of construction information is finally being unlocked, and that will lead to some amazing technologies and efficiencies.


Getting Started & Best Practices


What kind of construction problems are best solved with Machine Learning?

The single biggest problem the US construction industry faces right now is the skilled labour shortage, so any machine learning project that addresses that has to be taken seriously. And the great news is that machine learning and more broadly AI is fantastic at routine tasks in complex environments. That means you can use them to supercharge highly skilled staff by taking the routine tasks away and augmenting less skilled staff with the AI assisting.

We've seen really smart applications of AI/ML with companies like Smartvid and recently Holobuilder doing amazing work with photogrammetry (figuring stuff out in photos and videos), but I'm excited to see how far into the field we can push AI. At Nyfty.ai, our aim is to supercharge construction workers with our digital assistant, taking the routine tasks away, and making the complex simple - so skilled workers can do more of what they're paid the big bucks to do, and less skilled workers can be utilised to the fullest. We think that is key to solving the skills shortage - supercharging and augmenting.


How do you set yourself up for success when leveraging ML products/solutions?

Start with collecting data, and as much data as possible. If you're not already using systems like Procore in combination with their partner systems like APE Mobile, Smartvid, Sign On Site etc then you should start there. Once you're using those systems you can start leveraging the data you're collecting, its amazing how quickly you can get started.

If you're getting involved with a machine learning project, you should be prepared for some initial frustrations, its got to learn after all. The benefits of getting involved early however easily outweigh any frustrations, as the system will learn your way of doing things first.

Imagine a machine learning from a set of data that grows over time, just like a child's personality develops in the early years, so do the algorithms of a machine learning project, with the first learning interactions having more prominence than those that come later.

So the best time to get involved with machine learning is now.


How do you approach design for your products/solutions?

The benefit of using a new product has to be massive, thats a given, and I once thought that a product's design just had to be better than the status quo to affect change, but I've recently changed my mind.

I now think that the product's design has to be 10 times better than the status quo, and that means thinking radically different.

For example, its hard to persuade somebody to change if you simply replace a paper process with a digital analogue, even if the analogue is a much better user experience. But what if you could elimate the paper process altogether with just a conversation, thats a 10X improvement in user experience and massive benefit in terms of efficiency and accuracy.

Thats the kind of approach I think is needed in product design, particularly in an industry synonymous with low literacy and slow tech adoption.


The Future of Machine Learning in Construction


What are some of the challenges you've faced that are unique to construction when it comes to Machine Learning?

At Nyfty.ai our challenge is understanding the language, because in Construction we don't speak normal conversational english, we speak 'Construction'. And thats a problem for us as the tools and systems available are based on "Standard Models" which are created with generic, non industry conversations. So we've had to find a way to talk Construction which is tricky but not impossible. So that's been our challenge, and is why machine learning is so important to us. We can only go so far in teaching the machines how to speak construction, now it's over our users to interact with the system and let it learn.


How do you see Machine Learning transforming the industry in the coming years?

I see two areas where there will be a huge impact, first is supercharging skilled workers by taking away their mundane and routine tasks, making them able to handle more high skilled work. The second area is augmenting low skilled workers to take on higher skilled tasks as they receive instruction from the software. With the skills shortage the industry is currently grappling, I can't imagine a better or more urgent way to apply this technology.

Just to be very clear on this, I'm not implying replacing workers with robots or software. AI / ML products are very good at doing routine tasks, but they are very bad at making decisions. Humans have to 'stay in the loop', so when I say Supercharge and Augment workers, its not a synonym for replacing, quite the opposite in fact.



Nyfty.ai is available in the App Store / Play Store right now. Our 'Answers' plan is completely free, and for those who sign up to our 'Early Access Program' we'll also make our Premium plans free whilst we're building them.

You can join our early access program here: https://www.nyfty.ai/procore-early-access