Slashdot Mirror
Ask Slashdot: Could An AI Conceivably Create Futureproof Product Designs?
dryriver writes: Whether you are into consumer electronics, cars, furniture or other manufactured things, one aspect of them doesn't change -- the physical design or "look" of the product tends to age badly in our perception as newer products are released. When you first buy the product it looks "sexy and new"; 5 years down the road, it just looks kind of "old" or "less sophisticated" compared to the newer, sleeker products. To the question: Could you get an artificial intelligence powered by a neural network to train on hundreds of product designs created over the last 20 years -- possibly by laser-scanning products in 3D or providing 3D CAD files -- and learn with great sophistication how product design or "product looks" evolve as time passes? Could that AI then be coaxed into making fairly educated guesses about how a particular product might look if it were designed in the future, in say 2030? In other words, could a suitably trained AI give a laptop, car, or designer chair to be manufactured in 2020 the "design look of the 2030s" ten years early by extrapolating forward from the training dataset of past product designs?
It doesn't matter what designs a person or an AI comes up with, because people will always want something new. Just like every generation goes on to develop new music or new movies even though it's a lot of the same chord progressions or plot lines, it's nevertheless something new. It doesn't matter if you extrapolate further ahead and give us a design ten years before its time. We'll still be tired of it five years down the road regardless of when the countdown starts.
This question seems to hinge on the (almost certainly false) notion that design, and the perception of design, evolves in some trajectory that's independent of its context.
Things that look old do so in no small part because we can compare them with things that are new. Things that look cutting edge do so in part by rejecting design elements that look familiar.
Even if we assumed a scary good, probably better than human level, bot put to the task of inferring what "2020 thinks 2040 will look like" its output would, immediately, be part of 2020-era design, albeit probably a visually distinct flavor; and what 2040 actually look like would include reactions to, away from, against, with nostalgia for, etc. that "2040" design from 20 years ago.
When humans try this we get zeerust. It's not clear why a bot would do better; or that even an arbitrarily talented bot could beat the fact that the future it predicts will automatically become part of the past that the actual future evolves from(recursion is fun and unproblematic, right?).
There's also the problem, outside of some purely decorative objects or ones that aggressively try to defy the constraints of material culture(either trying to look more futuristic than the tech really is, like sci-fi TV props; or are deliberately throwbacks, like SCA longbows and stuff), that things look the way they do in no small part because of the constraints of technology that no amount of industrial designer resistance can get around.
Reading the summary about how designs look dated after a few years, to me it is like reading about a theological argument within a religion I never heard of. The claims made are completely alien to me, have no reflection in my own experience. I don't think products a few years old look dated. I don't think newer products look sleeker or more sophisticated. Often I think they look stupid and annoying. Can AI make that stop?
Starships were meant to fly, Hands up and touch the sky - Nicky Minaj