EXPLORING ARTIFICIAL INTELLIGENCE
By Gregory Ceton, CSI, CDT
  

   
Artificial Intelligence (AI) is everywhere. This article is just one more in a virtual flood of postings, TED talks, videos, and forum discussions about this topic in recent years, and that’s just what’s on the web. Odds are it has also crept into or  been responsible for  your favorite TV shows, not to mention its central presence in movies such as The Matrix, Ex Machina, or yes, A.I. Artificial Intelligence.

Part of the problem with talking about AI, as shown by the examples above, is there are still differences in what may be meant by AI or any of its components. AI often implies autonomy, which carries with it self-awareness and ability to act without any direction, but that isn’t where we’re at right now. Even without autonomy, AI is still a powerful and flexible tool to analyze data and make decisions.

ARTIFICIAL INTELLIGENCE: A DEFINITION
For the purposes of these articles, I define artificial intelligence as “software agents that can perceive their environment and take actions that maximize their chances of success at some goal.” Artificial intelligence also mimics cognitive functions that humans associate with human minds, such as learning and problem-solving. A good example of artificial intelligence that fits this definition is DeepMind, developed by Google, and its AlphaGo program, which has triumphed over many of the world’s leading Go players,  starting with Lee Sedol in 2016.

Another distinguishing factor when discussing AI systems is their conceptual breadth. Most fictional autonomous AI are examples of general AI in that they can act as a human would in any situation. Current implementation of AI is entirely “weak” or narrow AI. Narrow AI is tuned to specific datasets and problems. The AlphaGo program is a good example of a narrow AI. These narrow systems have AI cognitive abilities (learning, decision-making, maximizing success), but can only apply these abilities to specific problems. General AI is still theoretical, but a subject of ongoing research globally.

MACHINE LEARNING
Central to artificial intelligence is its ability to learn and be trained, in contrast with most software that executes set responses based on its programming. Through learning, the algorithms that power AI become modified as new information and relationships are acquired.

Learning is usually done by presenting the AI with datasets, often generated by human crowd input (for example, “tag your Google photos”) or collected by sensors or other digital equipment, and then allowing the AI algorithms to use those datasets in combination with rules established by the programmers. As relationships between data points are found through continuous review, the intelligence of the AI grows and its performance at its established task (for example, “identify photos with cats”) becomes optimized.

Another form of learning that has taken precedence in recent years is known as deep learning. Deep learning exceeds the capacity of the task-specific machine learning described above, by focusing on the relationships in a brain that are formed during learning. Deep learning systems are based on artificial neural networks (ANN) that mimic brain structure and neural function. The algorithms that make up the ANNs subdivide the task of learning by each specializing on a specific feature area of the presented dataset to learn. Connections between concepts are established by ANN “neuron” algorithms, giving this method of learning greater speed and flexibility than task-specific machine learning with its focus on problem and goal. Instead, the individual neuron algorithms and their small specializations can be recalled, and relationships reestablished as new information is gained by the AI as a whole.

So, to recap, in this article we have established starting points for:

A definition of artificial intelligence.
The importance of machine learning to artificial intelligence.
Methods of machine learning.






















An illustration of the types of work most humans engage in. Taken from a presentation given by Dan Chuparkoff at the South by Southwest (SXSW) festival.


Importance of machine learning to AI
The chief difference between directed software solutions and AI is the latter’s ability to improve its responses over time by “learning.”

Methods of machine learning
As last month’s article explained, machines usually learn through interaction with datasets and examples provided by humans to train them. The article also explored deep learning that mimics the way the brain works by breaking problems and responses into smaller pieces and combining those learned behaviors in different ways.

They’re coming for our jobs!
As you may expect, machine learning, algorithms, and specialized AI have already started to affect the construction industry, largely in the area of design.

Your job may not be eliminated, but it will change. Certain aspects of your work, such as activities requiring little or no complex thinking, can simply be done better and faster by some form of AI. Removing these job functions may also be welcomed by many professionals.

The graphic below is from a presentation given by  Dan Chuparkoff at the South by Southwest (SXSW) festival a few years back. It illustrates broad levels of work types most humans engage in. Tasks closer to the bottom are suitable for automated solutions, while the ones on top of the hierarchy may take more specialized knowledge, or in the case of “ingenuity,” combinations of knowledge that are tougher to accomplish with AI.

Most jobs contain tasks that are on different levels of the hierarchy, as a quick examination of your daily workload will reveal. For example, specifiers engage in significant amounts of pattern recognition and problem solving but there is no shortage of memorization and mimicry or even some innovation and ingenuity.

If some of your job functions are listed on the hierarchy, it doesn’t mean you are going to be replaced by automated processes, but it may mean that some work, especially repetitive tasks requiring memorization or pattern recognition in data, may be eased by new software. Many in the design community have benefited from some level of automation in their jobs for years. Model checkers like  Solibri and energy analysis tools built into BIM software rely on algorithms to eliminate a lot of tedious work that designers could otherwise be expected to perform. Design of complex facilities like hospitals has also been eased by use of  data capture and algorithmic design to help make the layout of spaces and paths better for healthcare workers and patients.

Looking forward, AI has the capacity to replace tedious work, improve quality and constructability of designs earlier in the project schedule, and provide an expanded palette of both solutions and inspiration for designers. Adaptation to new tools will be critical to succeed in a changing design marketplace, so keep an eye on topics like  algorithmic and  generative design, and the outcomes of machine learning on architecture.
 
Robots everywhere
Use of AI has even more promise when applied to the construction phase of the project and onsite. For example, as mentioned in  a recent The Construction Specifier article, many firms are facing a shortage of trained workers. AI may offer a solution to this and other construction issues, by offering robots and robotic tools that help the human workforce increase its speed, strength, efficiency, and overall productivity.

California drywall subcontractor  Martin Brothers recently posted a video that shows one employee using a Microsoft HoloLens  augmented reality (AR) headset with a construction model loaded into it to effectively frame a drywall partition quickly and correctly without the help of other workers or using traditional tools and documents. Possible outcomes of this kind of technology include accurate as-builts and a reduced construction schedule.

Japan’s declining birthrate has resulted in labor shortages in the country’s construction sector. In response to this shortage, Japanese construction firms  have turned to various forms of robotics. In addition to onsite robotic arms and other forms of worker-like robots,  Japanese workers can use drones and tablet software on larger projects to drive construction equipment remotely, thereby replacing the need for multiple operators.

Whether an increase in automated solutions, use of information tools like AR, or a combination of the two, the presence of robots and automated tools on the worksite is a permanent fixture. Just like designers, construction firms and employees have to modify their expectations when considering possible solutions for building issues. They may also need to modify their bids given the potential reduction in project timelines due to more productive workers.
 
Conclusion
The result of AI and its appearance in the workplace is not likely to be the fully automated job wasteland that many fear. Instead it will likely conclude as most other technological advancements historically have—with a change in work habits and a renewed focus on the skilled work firms are paying for. This kind of change means you will probably need to acquire new skills with replacement by a humming box in the corner being the possible cost of not keeping pace.

To conclude, here’s a summary of what was discussed in the article:

  • Increased use of AI and related methods of algorithmic analysis have already made design more effective.
  • More solutions and information about design and constructability will become available as AI tools become cheaper, better-integrated, and more commonplace.
  • Increased use of robots onsite will help to relieve the burden shortage of skilled workers presents and could increase worker productivity and worksite safety.

To Be Continued....