Humans often make mistakes when defining engineering drawings. While we may want AI to fix this problem, writes our columnist SJ, doesn’t learning from humans mean that AI will just dish out the same mistakes?
As the old saying goes, ‘Those who can, do. Those who can’t, teach.’ Whether you agree with the adage or not, it certainly tracks when it comes to a recent conference appearance I made. Except this time, it was more of a case of, ‘Those who can’t be presenters (because they can’t get approval through their company’s comms department), moderate panel discussions instead.’
At this particular advanced manufacturing conference, yours truly was left doing precisely that. And, when it came to skimming topics for the panel to cover, the summary read as if my LinkedIn feed had just regurgitated six months’ worth of buzzwords all over the page. Somewhat inevitably, AI was front and centre.
In Gen Z terms, this preamble was ‘giving me the ick’. And if it was giving me the ick, there was a high likelihood that my panellists felt it as well.
Each seemed nervous, and when I spoke to them ahead of the session, their eyes widened like a deer in the headlights each time I brought up AI-assisted workflows.
I saw the panic spread across each of their faces. One man, brave enough to speak, whispered in a tense voice: “But, SJ, we’re not really using AI-assisted workflows! I mean, we’d like to, but we just haven’t figured it out yet.”
I beamed a 1,000-watt smile back at him, before responding: “Exactly!”
Realms of possibility
That panel discussion taught me so much about all of the topics we covered, from cloud-based collaboration to the adoption of topology optimisation in the ancient behemoth that is the military-industrial complex.
Finally, I arrived at the dreaded question: “How are AI-assisted CAD workflows reshaping engineering roles, particularly when it comes to tasks like feature creation, optimisation and automated constraint management?” The room paused and an uncomfortable silence fell.
I’ll spare you the details, but as my panellists began to hem and haw around the topic, a hand shot up from the audience. An audience member with a question about AI drawing generation. Fortunately, my quietest panellist shyly picked up the microphone and responded firmly: “We’re not there yet.”
In my humble opinion, we aren’t at the point of AI drawing generation workflows yet, because we still haven’t truly figured out our current engineering workflows
But why aren’t we there yet? I think that’s the question we should be asking. And I specifically mean that it’s a question we should be asking each other, not a bunch of AI chatbots.
In my humble opinion, we aren’t at the point of AI drawing generation workflows yet because we still haven’t truly figured out our current engineering workflows. How do we currently communicate design intent in a drawing? And is that methodology synthesised enough that we could build heuristics for an AI to learn from?
Many engineers – myself included – also struggle with certain aspects of GD&T. I know this is true, because suppliers are still raging about every drawing that appears on their doorstep. ‘The drawing is king’ is a saying that I grew up hearing over and over on the manufacturing shop floor – the source of absolute truth when it comes to fabricating parts.
I’m a big fan of holding a large paper drawing with all my dimensions laid out for me, the notes cascading from the top left corner of the page. However, in recent years, I’ve noticed we’re moving away from physical paper drawings in favour of model based definitions (MBDs).
So, if your source of truth is constantly changing and being updated, which source do you feed your AI to help it learn?
I imagine that AI will one day advise me on GD&T tolerances, but in its current state, where it’s still learning from our databases filled with our human biases and mistakes, wouldn’t an AI drafter just offer copy-and-paste engineering? And how might that impact supply chains?
Trust and verify
I believe that AI has a place in engineering and a very bright future in design workflows. However, at this stage in its infancy, it is critically important that we should ‘trust but verify’.
Put simply, I can’t trust that AI is going to give me a perfect part or a perfect drawing, but I can trust that it will do the best that it can, based on who it is learning from.
I also think that when we discuss how AI will change engineering roles, a large part of our future responsibilities may include verifying AI model creations, changes, updates and so on.
So what I take away from that is that my job in 10 years’ time may be entirely focused on double-checking drawings created by an AI, rather than having my drawings double-checked by another engineer. Or, conversely, maybe I’ll be creating the drawings myself and AI will be used to verify my work.
Maybe the future engineering role isn’t that AI does all the work, but that it’s simply a tool we use to verify our own work?
Either way, it’s giving serious vibes of ‘Those who can’t, use AI.’
This article first appeared in DEVELOP3D Magazine
DEVELOP3D is a publication dedicated to product design + development, from concept to manufacture and the technologies behind it all.
To receive the physical publication or digital issue free, as well as exclusive news and offers, subscribe to DEVELOP3D Magazine here
