ChatGPT is like a con artist, disguising itself as knowledgeable but making up answers if it doesn’t have them, as was shown in tests. It presents itself well, but it’s deceptive. You won’t know when it is giving you the right answer, or an answer that just sounds right. This may be fine for essay questions on college homework, as many students have discovered, but it doesn’t work for design and engineering.
We’re at 3DEXPERIENCE World 2023, and questions from the ChatGPT buzz fill the air. AI-infatuated journalists are firing questions at Dassault Systèmes every chance they get. It’s not fair to ask CAD companies to whip up AI-assisted CAD on short notice, but the questions keep coming.
At the end of the second day, with still no answer to the ChatGPT questions, there are two back-to-back sessions that promise a glimmer of hope that something like ChatGPT for CAD may be in the offing. The first is “What Will the Next 5 Years Bring?” presented by Gian Paolo Bassi, SOLIDWORKS' previous CEO. The second is “The Next Big Thing,” by current SOLIDWORKS CEO, Manish Kumar.
But midway into each session, with no ChatCAD announcement, I have lost hope. I should have known. If there was going to be a blockbuster announcement, it would have been on the main stage the first day of the conference. But maybe they’d throw us a bone. A hint of ChatCAD in development, perhaps?
But with time and patience dwindling and not even the slightest hint of the next big thing coming, I was forced to ask for it.
“When can we have something like ChatGPT?” I finally blurt out.
Why can we not have an AI-infused designer, one that takes a design requirement and creates the geometry? I hate to create geometry. Why can’t you make CAD do that?
A brief pause from Manish Kumar, then: “If I told you, I’d have to kill you.”
That was a joke, I'm pretty sure. Kumar is too smart to issue a death threat in a room full of witnesses. But with this joke, he might have revealed more than he intended. It was certainly enough to make me jump to conclusions: that there was a Code Red issued at Dassault Systèmes, just like Sundar Pichai of Google did when he had to answer to ChatGPT. That Dassault Systèmes was hatching something in its labs. Didn’t a Dassault Systèmes executive say earlier that the company spends over $1 billion in R&D. Dassault Systèmes had to be making a ChatCAD.
Or, like Google, did Dassault Systèmes have the idea of an AI-assisted flagship product, immediately followed by a fear of its consequences. It would be so revolutionary; it would upset the status quo and blow up the existing business model. Let’s take our foot off the gas and put on the brakes.
How Hard Could This Be?
Conceptually, the idea of AI-assisted design is an easy one. Here’s how it would work, using a simple example: an electric lawn mower.
A company that makes all gas-powered lawn mowers wants to modernize their product line with a new line of electric, battery-operated lawn mowers. ChatCAD would take the existing CAD models and modify them to have electric motors and lithium-ion batteries. Such a conversion is hardly new. Batteries and motors for this kind of redesign exist; they may even be off-the-shelf items. The rest is a geometry problem, something 3D CAD is built for. Employ AI to accommodate the motors and batteries and dispense with the gas tank and noisy two-stroke engine. Make it all fit, and give me a few possible configurations to evaluate.
Honestly, why bog down your engineers for hundreds of hours by making them wrestle with their CAD system, or fiddle with fillets and faces? It’s pure drudge work. Moving features around a solid model is not worthy of engineering talent. Better to use that engineering talent for new ideas, with the time freed from CAD drudgery available for creative thinking.
Lawn mowers may seem like a niche market, but transportation is not. Already more is spent on electric bikes than is spent on regular bikes. Electrification of transportation and conversion to renewable energies is happening. And electrification is a solved problem; we have plenty of examples before us. These examples should serve as data sets for machine learning and the training of neural networks. Every case that starts with design requirements and ends up with a usable solution, can be used to guide similar conversions.
Every car company is on a path to electrify its fleet, switching out the IC engines for electric motors and batteries for gas tanks.
Kumar does not say, “We’re going to do that,” out loud; I am reading his thought bubble.
Isn’t Generative Design AI?
When I asked, “When will CAD be able to provide a design solution?” in the previous session, I was told, “We already do that.” But being able to change all holes of one diameter to another is a trivial matter. And trying to pass off generative design as AI is a stretch, given that generative design produces useless designs 99 percent of the time.
We have given generative design based on topology optimization every opportunity and lots to time to sort itself out and provide us with useful design solutions. CAD vendors continue to position generative design as the free-thinking design spirit capable of producing light-weighting miracles. But engineers have accepted so few generative design shapes for production that when they do, it is newsworthy.
I sign off from 3DEXPERIENCE World 2023 for the day, hoping next year some company will indeed have a ChatCAD to announce.