Applications of Artificial Intelligence, Deep Learning, Machine Learning, and the EPC Industry
Advances in big data collection and processing, distributed computing technologies, and statistical algorithms have mutually accelerated over the past few years. Companies of all sizes are taking advantage of their collective power to implement numerous forms of artificial intelligence, including machine learning and deep learning, to benefit their customers and business. More so than ever before, companies that are not traditional tech companies, like those in the EPC industry, are finding value in AI and data science. This added value is proving to be not just for Silicon Valley heavyweights. Bechtel is leveraging applications of artificial intelligence to evaluate project data to better understand schedule correlations, as well as improve its prediction capabilities.
As artificial intelligence (AI), data science, and big data become the buzzwords du jour, companies increasingly find themselves searching for data scientists, architects, and analysts without a plan. Understanding what AI is, what data science can do, and what it requires for success is a necessary first step.
- AI refers to technologies with the ability to perceive their surroundings, formulate strategies, and make decisions in pursuit of a goal.
- The precise threshold for what counts as AI is subject to interpretation. While people tend to use the term AI to describe machines completing tasks once performed primarily by humans, the reference point has shifted as the domains of machines and humans increasingly overlap. In what is known as the ‘AI effect,’ humans discount the ‘real intelligence’ associated with a task once a machine has successfully performed it.
- With respect to determining ‘real intelligence’, AI can be theoretically divided: general or strong AI and specialized or weak AI.
- General AI describes intelligence theoretically equivalent to the intelligence of human beings. Problems that require General AI are called AI-complete or AI-hard; the solver must learn, plan, make decisions in uncertainty, and even fundamentally re-program itself in order to arrive at a satisfactory result. Such problems cannot be solved with a specific algorithm or combination of specific algorithms. Presently, no AI-complete problems can be solved without a human in the loop and General AI does not yet exist.
- All of our existing AI technologies constitute—at best—weak AI. That is, AI designed to focus on a specific problem. Weak AI is not self-aware and it does not have the ability to apply intelligence to any problem.
- Siri and Alexa are examples of sophisticated-weak AI. What they do, they do well; but they operate within a defined range and struggle to deal with inputs outside of their limits. Despite not being generally intelligent, weak AI is still powerful; it can automatically regulate a city’s power and predict the stock market, as well as knock out an electric grid or spark an economic disaster (e.g., the May 2010 “flash crash” for which much of the blame fell on high-frequency trading algorithms).
Applications of Artificial Intelligence Augment Human Expertise
One may object that every project is unique and thus AI, which relies on the presence of patterns and relationships in data does not apply. Every project is unique, and we can leverage AI to improve project delivery. We do not propose to remove humans and their necessary expertise from the process—general AI, after all, doesn’t exist. But AI-based solutions can range from automation to augmentation. It is with the latter, ‘augmented intelligence’, in which the most dramatic impacts can be seen. AI can process billions of data points and synthesize the most significant possibilities.
For example, AI-backed decision support of construction sequencing will aid in both long-term planning and immediate decision-making by helping make apparent the (potentially unintended) impacts of different scenarios. In each case, we want to leverage big data and artificial intelligence to support humans, so Bechtel’s experts are focused on the highest value tasks.
- The primary driver of AI is machine learning.
- Machine learning enables technologies to learn on their own. More specifically, machine learning applications use algorithms to teach machines how to learn from data, as opposed to what to learn. These algorithms identify patterns in observed data, build models to capture those patterns, and use them to predict new outcomes.
- Deep learning techniques, a part of the machine learning family, aims to learn meaningful representations of and relationships within the underlying data as opposed to simply accomplish a specific task. We may then ask questions of and make predictions from the deep learning representations. Roughly speaking, deep learning attempts to mirror the information processing and communication patterns of the human nervous system and brain. The results of deep learning moves us closer to general AI, but still have a ways to go to reach human intelligence, particularly with respect to uncertainty and previously unseen information.
In all, AI can be used to understand and predict risk, optimize planning, detect anomalies, and respond to unexpected events—to name only a few.
AI for EPC
Success in this pursuit can disrupt our massive industry. Construction-related spending accounts for about 13 percent of the world’s gross domestic product (GDP). In 2013, global investments in energy, infrastructure, mining, and real-estate-related projects was about $6 trillion. By 2030, that could be almost $13 trillion.
And the biggest problem facing the EPC industry? Productivity.
Where nearly every other industry is progressing, productivity in construction has advanced only one percent over the past 20 years. To put that into perspective, productivity in manufacturing has nearly doubled in that time. In a market that contributes to so much of the world’s GDP, even small improvements to labor productivity would have tremendous impact. With advancements in data, computing, and algorithmic learning, we have the opportunity for immense progress.
Bechtel is particularly well-positioned to lead the charge, and we’re doing so through our Big Data and Analytics Center of Excellence. With 120 years of data increasingly at its fingertips through digitization efforts, we can ask where hidden inefficiencies might lie and what might drive them. Presently, for example, we are developing a machine-learning tool to identify the most efficient construction packaging sequences for our most complex projects. Through artificial intelligence we are able to find connections and learn new solutions we would not be able to find if every possibility had to be tested in the wild. We can do so in the presence of dynamic factors and shifting constraints. Once faced with unexpected weather, material or labor shortages, our project teams will be able to ask: “What is the best approach from this point forward?” Machine learning will allow us to optimize in real time as events impact schedule and execution.
AI will continue to improve for the EPC industry as the technology advances and consistent baselining can be captured. In follow up pieces, we will dive into examples of how Bechtel is driving innovation by using AI and machine learning to tackle the industry’s most pressing challenges, and how our organization is transforming in response.