I’ve been away from the forums for the past couple weeks so I’m getting into this a little late. I was a bit taken aback by how cold the initial response was, but more by how little staunch advocates (sk) of generative design have moved towards a position that is workable for most IDers. GD needs to enhance and be integrated into the design process, not try to take over (that may not be the intent, but that’s how it’s come across). So here’s my $0.02 (sorry for the length):
While “Generative Design” sounds cooler, I think “Random Generative Creation” or “Curious Tool” or some other name seems more accurate, because it’s not really “design”. It’s not replacing the decision making, problem solving process. And it’s not going to replace research, marketing or other inputs. So generating lots of random shapes is somewhat pointless, because basic decisions such as large or small, square or round, etc. should be made through other means besides picking from a sea of variations. Plus, as has been mentioned, it’s not generating variations in the primary generating-variations phase of the project (2D sketching).
That being said, I do think it could have some good uses, which may or may not be true GD. Actually, probably most aren’t, but maybe that’s my point, that just random variations isn’t useful enough:
Work with engineering/structural needs - Sometimes we need to design parts that are structurally important, and it’s good to know possibilities for ideal structures. But instead of giving it a blank slate, give it a few designs you think will work and have it generate variations on them that will actually work or be stronger or more material efficient. Then tweak and run some regular FEAs. Sort of like the Morphogenisis stuff from SolidThinking, but let me suggest solutions then give me structurally tuned variations.
Better, more understandable & tweakable parametric design – sorta like the sliders interface that had been proposed by Cdaisy. For example, maybe I know an edge should be curved, but I have to figure out exactly what curve. So show me circular, elliptical, parabolic, etc. with a 2-3 dimensional variations for each, plus let me throw in a hand drawn spline or two to help me resolve this. Then I could interactively throw out what immediately isn’t working, tweak what is, and maybe do some more random dimension variations for the ellipse. Or maybe I can link a few different edges and en masse switch them all from circular to elliptical to see what that’s like. Or randomly tweak dimensions or the mix of curve types to help me experiment with the refinements. The randomness helps solve detail problems, not the full design. And the user always has final control, and can always add their own designs to the random mix (maybe be able to draw the major lines of a form, and have it work off that). And not 1000s or 100s of choices, but a more manageable number (5-20?) that can be whittled down then re-expanded if necessary.
Equation driven designs & Patterns - kind of an extension of above, and seems to be the only consensus on this thread. But this seems only really useful once it has been decided by traditional means that a certain general shape is desired.
Sparking ideas, happy accidents - I’d never use this as my only means of diverging, but maybe 3/4 through the process I could see running this and see if there was some interesting/useful direction I missed. More of a check than anything. BUT, unfortunately, this generally happens pre-CAD, so having to go into CAD just to do this is a deterrent for many projects & designers.
But with those benefits do come some dangers, I think, which would need to be addressed (by the software itself and the way it’s used):
Missing ideas that the computer didn’t generate. Instead of all those handle generations, what about a knob? Or a hole in the handle? Different levels of detail? etc. Right now it doesn’t seem like the generators are going to create anything “out of the box,” but maybe will in the future (random word → google image search → random selection → combine with sketch. Sounds more like a toy, but could find something interesting, so useful if that takes 5min of human work).
Becoming too reliant upon it for idea generation. Then instead of having designs looking and feeling like they were inspired by nature or whatever you previously used as inspiration, you get bland and cold results that aren’t very interesting or human.
Not tweaking the results. I don’t know why any serious designer wouldn’t, but it seems like many negative reactions assume that the designs would go straight from generator to manufacture.
Going to CAD too quickly. With good reason, most or all of the GD tools for ID & architecture work with CAD. But for ID, most of the diverging and concept generation is complete at that stage. It seems like 1000s of times it’s been reaffirmed that sketching is important, so skipping that to get into generative design immediately seems like a non-starter.