Making the most of AI: treat it like a new colleague

One way or another, the fact is that AI has the potential to be one of the most profound changes in human history. Or as the philosopher John Culkin puts it: “We shape our tools and then our tools shape us.”

However, people who work in digital construction know that tools alone will not change humanity. For this to effectively take place, we need to understand the tools we have in our toolbox and master how we use them.

Training the AI

To start effectively implementing AI in our day-to-day activities, we need a change in mentality. I like to imagine that I need to treat the AI with respect. I try not to make the AI work for me, but to make it work with me. We need to mentor the AI to enhance our capacity to finish tasks with the level of quality we want, and by the deadline we need to meet.

It is as if we have a new colleague at work, who needs to be introduced to some of our tasks. Initially, this might take up some of our time, but in the long run, showing a colleague how the job is done will pay off, as we will be able to share the workload. The following are some tips on how to work with this very special ‘colleague’.

Defining clear objectives

Your new colleague has just arrived at the office and introduced themselves. You wouldn’t follow that by asking your colleague to create an essay on the Battle of Hastings, would you? Most sensible people would introduce themselves, saying what they do in the company and then taking the new colleague through some of their initial tasks. AI performs better when we adopt the same process.

Until general purpose AI is developed, we need to describe the context in which we are going to perform a task as well as we can to our current AI tools. If the task is construction-industry related, we need to specify this. So too, if the task is specific to a discipline such as civil engineering or structural engineering. Which role is the AI is taking on? What is considered a good result and what is considered a bad result? When do you consider the objective to be met?

In fact, this can become an arduous task. Defining an unambiguous objective and the context surrounding it demands a lot of thought. There is a lot of value in defining the parameters in which the AI will operate. This is why you might see terms such as “prompt engineering” or “AI personas” becoming more and more popular.

Use what you already know how to do

At initial stages of implementation, AI will work at its best when we ask it to carry out tasks that we already know how to complete. This helps you and other people who might be interacting with the AI to build trust incrementally on the system. Again, we can carry on with the new colleague analogy here. After some time working together, the colleague has gained your trust, but for other people to trust them, you will need to show that they can first give reliable outputs. Once you have managed this, you can venture onto more complex processes – ones that might have less predictable outcomes.

To get reliable results, you will need to evaluate and validate whatever the AI has come up with. You can do this accurately when you ask the AI to perform tasks you are familiar with, as you already know to distinguish a good answer from a bad answer. This validation process can also be used to train and fine-tune the AI.

Using AI to complete tasks that are well-documented in your organisation (e.g. the process is mapped and you know how much time will be spent at each step) can also be a good starting point, as existing methods can be used as a benchmark for measuring AI’s capabilities and improvements over time, as well as whether an AI system is something that is worth it putting your time and effort into.

Use in non-nuanced tasks

While AI models are fast, accurate and don’t get fatigued or distracted, they lack common sense, and are not that effective at making ethical and moral judgments. The good news is that humans, generally speaking, have almost the opposite attributes!

AI excels at automating repetitive, rule-based tasks with clear boundaries. For example, it can handle data processing, basic data entry or routine calculations. However, AI models often struggle to understand truly the context and nuances of human language and situations. AI lacks the common-sense reasoning abilities that we have. Understanding and interpreting the unspoken or implied aspects of nuanced tasks can be challenging for it.

However, AI systems are less prone to errors in non-nuanced tasks because they are machines and don’t get tired or lose focus. This leads to higher accuracy and consistency in tasks like data validation or quality control. By assigning non-nuanced tasks to AI, organisations can optimise resource allocation. Human employees can focus on tasks that require critical thinking, creativity, empathy and nuanced decision-making – areas where AI often falls short.

Augmented not artificial

All these points might appear obvious when listed like this, but it can be helpful to go through them when we are trying to use AI to complete a task. Perhaps due to the hype around AI and the good first impression that we have when we interact with tools like ChatGPT, people tend to have high expectations about what can be achieved with these tools.

However, the fact is that in its current format and with its current capabilities, AI is a tool that still requires some user skills to be employed effectively. That is why perhaps we should be less inclined to think of it as an artificial intelligence, and more inclined to consider AI an extension of what we can do and achieve: an augmented intelligence.

Don’t miss out on BIM and digital construction news: sign up to receive the BIMplus newsletter.