Welcome to Engineers Week at Encorus Group. As we celebrate the contributions of engineers across the nation, we also look at future possibilities within the field. In honor of this year’s theme of Engineers Week, “Welcome to the Future”, we look into the future of engineering through the eyes of Geoff Chadwick, PE, Control Engineering Group Lead at Encorus. In this blog, Geoff discusses the trends that could shape the future of our industry.
The Future of Engineering
When I was fresh out of college, if you’d asked me what I thought the biggest change from 2000 to 2025 in engineering would be, I’d have said improvements in design software – like CAD and analysis tools. As computers were increasing in performance by staggering amounts and the internet was becoming ubiquitous, it was easy to assume software would be the biggest change in engineering.
As we finally approach 2025 and I reflect back, I can say pretty confidently I would have been wrong – improvements in software for communication and collaboration have been far more important.
In 2000, email was the fastest way to communicate across a country (and forget about instantaneous translation software!). Now email is the “snail mail” of the 2020s, thanks to software like Microsoft Teams, Cisco Jabber, Slack, etc., where you can communicate around the world – sharing your thoughts, your desktop, software, and your ideas instantly. Collaboration has simply skyrocketed, and that has added value at every level of every engineering project.
So now let’s look forward to 2025 to 2050 – where do we think engineering is headed in the next twenty-five years?
The easy answer is “AI.” We’d probably be correct, but as “AI” is going to impact every industry worldwide, let’s speculate on how it’ll change engineering specifically. To kick things off, engineering “AI” falls under two main categories:
The first are “language models” where “AI” uses basically all of the text it has consumed to generate the most likely letter or word in response – for those that remember Microsoft’s “Clippy” – that was a rough draft of these systems. If you type in “What is the speed of light?” to ChatGPT it gives you back “300,000,000 m/s or 186,000 mi/s”. It doesn’t know what light is, what speed is or even what a meter is. It just knows that it has consumed a thousand physics textbooks and whenever the book has “the speed of light”, the same sentence always follows, and that’s the one you probably want, so it regurgitates it back to you.
The second is “data analytics”. If I asked you to compare a year of daily weather forecasts to weather reports for your local station, you could probably find some patterns – but AI consumes billions of data points in seconds – think every weather station in the world for every day in the past fifty years. Just like “language models”, though, current analytics need prompts for what to look for, “criteria” to filter the data. We’re starting to see AI analysis tools which can find patterns with no human prompts and finding correlation (not causation!) between data sets. Just like the “language models” though, it doesn’t know what “wind speed” is, nor does it know what “temperature” is.
Both of these tools do one thing really well and one thing really poorly – they process large amounts of data at incredible speeds. They iterate text responses, data simulations, and answers to questions, even though they don’t know what any of the words mean, and they’re getting good at it. The problem with it all is that they “hallucinate” and can make up data, criteria, and information on the spot. As AI systems are generally “black box” systems, they don’t show their work and you’ll never know where it all went wrong.
For engineers? We’re often required by code to show our work – and that key fact is what will set the tone for AI and its limits in engineering over the next many years.
Let’s use an example of a new billion-dollar chemical plant. Assume it takes a team of two hundred engineers five years to complete the design. Once done, almost every design goes through “value engineering”, where they look for opportunities to reduce cost. The problem is re-analyzing the entire design to reduce the footprint, amount of material needed, or time for construction is nearly impossible. Usually a few percent is considered “good enough” because the cost of optimization is greater than the savings.
Enter future AI tools. AI tools will be given a few thousand requirements, a hundred design rules and every detail of the design. It’ll run ten thousand simulations to reduce construction materials, footprint, and labor. The real trick is that as these tools don’t show their work, we won’t ask them to design a new chemical plant. We’ll ask them to “optimize” the plant and come up with common errors and overuse of resources across the thousands of simulations to make a list of suggestions to the engineering team, leading their review process to eliminate mistakes and waste.
As these AI tools are (and will ultimately be) black boxes for decades to come, we will still need to show our work. Engineers will need to understand code, constructability, and design requirements better than they ever have before, as well as how to enter those requirements into an AI prompt. AI tools are going to streamline and optimize engineering, letting engineers focus more on the design and human elements of facilities, while the AI optimizes cost and materials concerns, improving the return on investment of capital projects and plant improvements across the board.
Written by: Geoff Chadwick, PE
