Three centuries ago, steam engines sparked the industrial revolution. Today, another technological tipping point is approaching, with the potential to drive even greater transformation.
Until recently, information systems were limited by the algorithms and data selected by their users. No matter how skilled, human choices still impose constraints. Just as the steam engine provided freedom from the limitations of muscle power,1 artificial intelligence (AI) holds the promise of surpassing many of the limitations of brain power.
AI has already arrived, so the time has come for contractors to decide when and how to incorporate it into their businesses. However, as with every advancement, there are rewards, risks, and trade-offs.
This article provides a practical look at AI’s role and future potential in construction, as its influence on the industry rapidly expands.
Understanding How AI Learns
Before evaluating the benefits and risks of using AI in your construction business, it’s important to understand what AI and machine learning can do, and what it should not be expected to do.
Patterns vs. Logic
What is fundamentally different about AI, and why it is disrupting the IT landscape?
Legacy computer programs rely on hard-coded logic that applies a fixed set of rules (i.e., algorithms) to a set of data. The rules themselves do not change. This creates a state-machine, in which a given set of inputs return a set of outputs in a very predictable way. The intelligence or logic in these programs is built in via rules established by the programmer.
AI systems are typically built to identify patterns in a set of data. Once a pattern is discerned, a prediction can be made about likely next steps. For example, if almost every time one typed “CFM” the next character selected was “A,” then the next time “CFM” was typed, the system might suggest “A” for the next character.
Unlike legacy programs, this suggestion can change based on ongoing use of the AI system. If “CFM” is followed with a “B” often enough, then the suggestion will eventually change, not because fixed rules dictate it, but because the rules of the AI system are adaptive.
Getting Smarter Every Day
Many AI systems learn from experience, with users training them based on their choices of inputs and outputs. This is the essence of machine learning, in which the logic of the AI system comes from the logic in its data set.
Underlying this simple statement about machine learning is a sophisticated computational model, but exploring it fully is outside our scope here.2
In general, AI systems get smarter in a few primary ways.
Supervised Learning
In supervised learning, a desired result exists against which the outputs of the AI system can be compared until that desired result is achieved.
For example, to train an AI system to be able to recognize the image of a dog, one would feed it a large set of training images labeled “dog” or “not dog,” and then compare accuracy of the predictions to the training data.
Over time, correct responses are given more weight until the system learns what visual parameters signal “dog” with a high degree of accuracy.3 Many people first encounter AI through recommendation systems, like autocomplete or product suggestions, that are trained through vast user data.
Unsupervised Learning
In unsupervised learning, the AI system is set loose to identify patterns in a set of data, employing techniques such as clustering to build data relationships.4 Familiar examples are the applications that take the virtual meeting transcriptions and summarize the most important issues and action items.5
Reinforcement Learning
In reinforcement learning, the system is rewarded for producing positive results, much like training the family pet with treats for good behavior. A classic example is in game play in which the system learns through trial and error for how best to win a game, once it knows what winning means.6
For systems that operate in changing conditions (such as the construction industry), reinforcement learning is a powerful tool.
These three primary methods of learning include numerous subcategories and nuances, and many AI systems incorporate a hybrid approach. For example, Netflix recommends titles based on techniques that include supervised learning and collaborative filtering.7
Understanding how AI systems are trained helps lead to a better understanding of what type of results to expect. AI systems that use machine learning are pattern recognition engines.
They can:
- Look for existing patterns in new data (e.g., “find the dog”)
- Look for new patterns in existing data (e.g., “identify the meeting action items”)
The beneficial applications of AI for construction operations are many and growing. Before exploring them, it’s important to understand the risks that accompany these benefits.
AI: Real Governance
It is important to bear in mind that machine learning systems fundamentally rely on patterns and probabilities. They are only as logical and accurate as the data used to train them. Even when they’re accurate, there are times when relying on legacy software applications or human intuition remains the better choice.
When deciding if and how to implement AI tools, consider whether the task or process falls into one of the following three categories: high-risk decision-making, sensitive or private data, or decisions requiring empathy. If so, it is likely best to steer clear of an AI solution.
High-Risk Decision-Making
Relying exclusively on AI for critical decision-making is rarely a good practice, as these decisions are often nuanced and leave little room for error.
AI systems may offer helpful insights but can also contain biases rooted in their training data. Regardless of how AI is deployed, a human must always remain accountable for the business decisions taken, especially those which are mission critical.8
Sensitive or Private Data
The use of personal information in AI models potentially exposes the company to data breach risks and data security laws such as the General Data Protection Regulation (GDPR).9
It also has the potential of introducing bias to an AI model that could negatively impact its performance.
Decisions Requiring Empathy or Human Context
AI systems do not come equipped with a moral compass. Not only for the sake of accountability but also for ethical guidance, a person should remain as the final arbiter of AI-informed business decisions.