The entire world is talking about artificial intelligence and what it means for our future as people and as workers. The $13-trillion global construction industry has undergone massive changes in the last decade, and now even more technological change will make a big impact on our industry’s future.
AI is an umbrella term. There are sub-fields of AI, like machine learning, computer vision, deep learning, and generative AI. Consider AI as a toolbox, and each subfield (machine learning, etc.) is a tool in that box. We are typically handed a new tool every few years, and right now the one everyone is talking about is called Generative AI. What’s the big deal?
When technologies impact blue-collar jobs, like new energy or automation technologies, it takes time to see the change, as these require building new infrastructure (like a new factory or a new power plant). It has happened many times since the start of the Industrial Revolution, and we have seen how it usually plays out. As a result, many people found ways to reskill themselves or move to different industries. This fact is mentioned when experts talk about tech’s impact on workforce.
We, as a society, have been lulled into believing it would still take decades for AI to impact jobs to the point where we view machines as competitors. This time, it’s a bit different for two reasons: Generative AI is impacting white-collar jobs (architects, engineers, project managers, etc.), and its infrastructure is already in place, so adoption is happening faster than expected. As a result, predictions about how it will impact the workforce might not necessarily be accurate.
According to a Goldman Sachs report, AI automation stands to impact 66% of all jobs and catalyze a productivity push that could boost global GDP by 7%. In the U.S., office and administrative support jobs have the highest proportion of tasks that could be automated, with 46%, followed by 44% for legal work and 37% for tasks within architecture and engineering. The life, physical, and social sciences sector follows closely with 36%, and business and financial operations round out the top five with 35%.
37% of task automation for architecture and engineering is going to significantly change the industry. Generative AI will primarily impact tasks performed at an office or remotely, and the impact on field work is predicted considerably minimal at this time.
I believe that will change with the next wave of AI innovations. When we talk about industry use cases and AI, the technology is mostly machine learning and computer vision allowing the machines to learn and see.
Our work environments are constantly changing in the field. That’s one of the reasons we don't see robots on work sites as much as we see them in factories. That’s changing with machine learning and computer vision, as robots are able to learn and adapt to the environment and operate autonomously. I have been working in the autonomous field of operations for the last five years, and I can report that autonomous ground and flying robots (drones) have improved significantly with the help of AI in the last two years.
Today, robots are either preprogrammed, use machine learning or use computer vision. In the latter case, you record an operation while manually controlling the machine, and then have it repeat the operation autonomously on its own. It can adapt to its environment to help avoid obstacles thanks to the capabilities of AI.
In the future, I see AI impacting field operations as several new technologies merge and work together.
Machine Learning and Computer Vision
Today, we can have a drone autonomously fly and identify rust on a steel structure, determine the severity, and then automatically record it to the cloud and create a work order to address the issue. The operation is autonomously completed through the use of machine learning and computer vision. This example is a working use case.
Generative AI
Using Generative AI, we at Oracle have created a prototype that can help automate a pre-bid schedule. In this example, we provided a Request For Proposal (RFP) document to a chatbot; it can summarize the document and then ask questions to further clarify items such as type of structure. With that additional information it will automatically generate a pre-bid schedule. By no means is this schedule final, as it should be reviewed and adjusted by a human scheduler, but it can save workers significant time shortening the process from hours to minutes.
There are many examples where generative AI can also write code. Now, it has multimodal capabilities to see, hear, and speak. In one example, the user provides an image of a dashboard, and with simple prompts they can request for it to write code to create that same dashboard. We’ve found the technology can repeat this process reasonably and accurately. It’s all done with large language model (LLM)-based chatbots and AI’s ability to see images.
LLM-Based Autonomous Agents
Another exciting breakthrough are LLM-based autonomous agents that are designed to accomplish tasks through self-directed planning and actions. You define a goal and it can creates task and executes them.
Let’s say your goal is to “build a two-story, 50-car, sustainable, electric vehicle parking garage.” AI would then create smaller tasks like searching for sustainable building methods, zoning codes, optimum spacing for parking, etc. It would list options such as sustainable materials and ask us to make choices. Based on your choices, it can also plan and execute them. I tried a prototype, and it was impressive to see how it performed. This technology is still in the research and development phase, but it has significant potential to improve how we run projects.
The Future? All the Way to ENR FutureTech
Now imagine all of these technologies working together seamlessly. You could set a goal, such as building a new parking garage, and it could create digital and physical tasks, and then execute them.
As an example, if the task entails surveying a site, AI could trigger an autonomous drone flight, scan the area, upload to the cloud, process and provide the information to the next task.
Additionally, if the task is to install an equipment, it could order the necessary equipment, process payment, track delivery, program the robot (LLM writing code), pick it up, and install it and provide the information to next task.
It sounds like science fiction, but as someone who has been working in tech and industries including construction for awhile, I can see how these technologies will come together one day. Through the use of LLM’s, machines can learn from text and predict the next word fairly well with some errors (these are called hallucinations for an AI). The next tipping point will be achieving self-supervised learning from video where we don’t have to annotate images or video to train the models. AI’s impact on our workforce and worksite will go well beyond our imagination, helping us with deal with labor shortages, work more efficiently, and enhance our safety.
Burcin Kaplanoglu is vice president, innovation at the Oracle Industry Lab in Deerfield, Ill. He is active in industry organizations and on LinkedIn where he provides educational content related to technology, innovation, robotics, AI and industry use cases.
