We're now seeing the first practical applications of of Artificial Intelligence (A.I.) in construction. As construction sites become increasingly digitally connected and as datasets grow, the question is not if AI will impact construction, but how an industry with relatively low levels of innovation can take advantage of a rapidly improving AI toolkit.
What is AI anyways? A Very Brief Introduction
Every day we hear news about new applications of A.I., machine learning and deep learning, all related terms, often used interchangeably. They describe a series of technologies which are creating both excitement - and confusion. A.I. has often been used as an overhyped marketing buzzword, we hear about robots taking our jobs, and the dystopic risks of intelligent machines have even been touted as potential threats to civilization by high-profile technologists. But A.I. is also foundational to step-change improvements in how computers can interact with the world and solve complex challenges. So lets start by talking about what A.I. is - and what it isn't.
When we say A.I., we're using a blanket term to talk about algorithms that can be focused on problems which, up until recently, could only be solved by human intelligence - like speech and image recognition, language translation or how to win a chess game. Machine learning refers to a specific form of A.I. which can be trained to recognize patterns and make better decisions with more inputs - it can learn without being explicitly programmed to do so. Amazon's book recommendation engine is a well known example that looks at all of your past purchases and recognizes what you're likely to want to read next. Deep learning is a sub-category of machine learning using multiple levels of abstraction to organize inputs into accurate categories, for instance recognizing that an edge in an image is part of an eye, which belongs to a face, before deciding whom the face belongs to.
The combination of 3 important technology trends have resulted in a drastic performance improvement - and an great expansion of the applications for AI:
1. The rapidly falling price of parallel computing which lets artificial neural networks (the mechanism behind machine learning) perform multiple operations in parallel. Graphics processing units (GPUs) are examples of neural networks. They can see a pixel of an image in relation to all the pixels around it, and can thus make classifications based on complex patterns. When running in the cloud, GPUs can recognize your friend's face in an image uploaded to Facebook and make millions of 'tag suggestions' a day.
2. Better algorithms, such as a tweak to hierarchical pattern recognition that enabled what we now call deep learning. When this algorithm was applied to Google's online translation service in 2016, its error rate fell by 60% overnight.
3. Big data and a widely available set of data sources. Think of a toddler - the more chairs he sees, the better he'll be at identifiying a chair next time she sees one. A.I. needs exposure to a very large quantity of data to accurately identify patterns, classify inputs and to solve the problem its being trained on.
We're not (yet) talking about super-intelligent builder-robots which execute designs generated by algorithms, but even some early explorations of this are underway and could be indicative of future engineering trends. But let's save that discussion for another day.
A.I. for Construction
Nowadays, a construction site manager with a mobile device can quickly capture photos, videos and updates to the latest project conditions (models, drawings, schedules). The latest media and data is uploaded to cloud-platforms, where it can be analyzed at a project or cross-project level. These data-sets will continue to grow exponentially as the cost of drones, laser scanners, wearable cameras and other data collection devices falls, as mobile-apps for construction add functionality, and as technology becomes adopted by more and more project team members.
Smartvid.io lets users upload photos, voice and video recordings to their platform, where their A.I. engine VINNIE then looks at and listens to the media to place smart tags on images and audio. Thanks to this automatic tagging, contractors can later accurately sift through gigabytes of media taken on a typical project to find exactly what they're looking for. In some cases builders can find documentation that protects them against a claim, saving of millions of dollars. It's not just precise, it's quick too. In 2016 Engineering News Record (ENR) hosted their annual photo competition, but this time they partnered with smartvid.io to make sure none of the pictures depicted hazardous safety situations on-site. Alongside human safety experts, VINNIE scanned through the more than 10,000 photos in under 10 minutes - the same task took the human judges over 4.5 hours. What's more, VINNIE correctly identified 10% more photos depicting people than its carbon-based competitors, flagging 32 pictures featuring people not wearing safety helmets and 106 without high-visibility clothing. The safety experts agreed that an AI like this could help them quickly sort and identify on-site safety hazards which require a more careful review.
The availability of software that can automatically sort, filter and extract pertinent features from images strengthens the case for increased onsite photo and video monitoring. Drones, already popular in construction as a way to perform inspections of inaccessible areas, may soon "see" sites as humans haven't been able to, not just from a different angle, but identifying patterns that lead to higher safety risks or impending schedule and cost overruns. And as "smart helmets" such as DAQRI go from prototypes to field-ready devices in the coming years, A.I. image and video recognition tools will assist site inspectors to identify similar hazardous behaviours or conditions in real time.
Patterns don't just manifest themselves visually. Most construction management software applications are built on cloud-based platforms that allow users to connect real-time project data to an increasing number of external tools such as document management systems, scheduling, business intelligence and project reporting apps. Contractors will soon run machine learning algorithms on their ever-increasing cross-project datasets to identify patterns that humans simply can't - like the effect of model changes by a mechanical designers on schedule delays and cost overruns. On a project level, project managers may want to use AI to search through keywords in digitally documented reports or inspections, to flag potential high-risk observations for urgent resolution.
A.I. is starting to see, hear and understand our construction sites, and offer us better insights into how we should be managing them. Contractors can look to other industries to imagine what the next years might hold for the adoption of these tools in construction. The tech giants are already in an all-out A.I. Arms race. Google's CEO recently stated that the world is becoming A.I. first. As algorithms become more fine tuned, companies who successfully implement machine learning to assist project management may significantly reduce risks in key areas like site safety and schedule reliability. They will also inevitably gain key competitive advantages in labour efficiency - A.I., as any new technology, will replace some human workers whether we like it or not. But A.I. won't replace human judgement (yet). Instead it's beginning to assist project managers by uncovering insights that help them make better decisions.
If you want to dig deeper into the topic of A.I., machine learning and deep learning, the following are good places to start:
What the heck is… Machine Learning (5 mins read)
The Great A.I. Awakening (45 min read)