Core77

[SK]

When I look at all those renderings of differently proportioned MP3 players, I just see "meaningless variations" so I guess you've succeeded.

Chris, you make a valid point. But they were generated in 4 to 5 seconds and hundreds of them. Wonder if you would agree that there are 1000's of such meaningless designs floating in the market, designed made and available for sale at much greater expense time and cost. The good thing about digital design now is that you are able to see what is being designed or generated and kill those that you don't like. If you have skill, nothing is stopping you from modifying it to your liking as your sure do with your other designs.

What you create, I can guarantee will not be liked by quite a few others. If they are IDers themselves, then we will possibility be down to a fractional percentage. Nothing to do with your design. But I am sure you would agree, that is the with the "I like or don't? " (full stop) - People. Now, if you ask those connoisseurs who do not like your design, non of them are likely to blame SolidWorks or ProE or any other CAD package you use.

I don't consider myself a competent product designer. I a clearly not. So, the designs displayed suffer in quality. This has been pointed out many times and I accept the short coming. But I am not sure why many intelligent designers are unable to distinguish the quality and capability of CAD tools from sample designs created by it ? Wonder if you can help explain ?

[Electroflux]

The big difference between those who use generative design and those who remain skeptical is that the first sector focuses on what it can do and the second in what it cannot - for now.

Perhaps some see limitations as problems, while others sees them as goals.

[nunoCR]

So, basically generative design is, the manipulation of the architectural constraints of an object (the "genes") in order to generate as many design variations as possible (offspring), and then do an artificial selection to arrive at the most fitting one towards the desired outcome (environment)?

I also think that these "genes" in objects have been labeled as temes or techno-memes, but I'm not sure if they are the same.

[cg]

When I look at all those renderings of differently proportioned MP3 players, I just see "meaningless variations" so I guess you've succeeded.

Chris, you make a valid point. But they were generated in 4 to 5 seconds and hundreds of them. Wonder if you would agree that there are 1000's of such meaningless designs floating in the market, designed made and available for sale at much greater expense time and cost. The good thing about digital design now is that you are able to see what is being designed or generated and kill those that you don't like.

The 1000's of meaningless designs were likely done without any kind of design process. They were done with a "shotgun" approach which is common in the East where manufacturing is cheap. The theory is that you flood the market first with as many inexpensive variations of a product as possible and see what sells. In the West, it's more common to have a "rifle" approach, which utilizes a lot of research, strategy and design process to ensure the product hit's it's mark, reducing risk.

Generative software is great for the meaningless "shotgun" approach, but designers strive for the meaning-full "rifle" approach.

For example, if an Industrial Designer were to design an MP3 player, a good design process would be to use "generative research" (aka co-creation) to help isolate preferences among those customers. In an hour, I would make a bunch of wood or foam shapes, and hand them to research participants. I'd ask them to talk about the merits of each shape and weight, talking about how they'd use the product in their life. I would then give them a bunch of cut-outs that represent controls, like screens and buttons. I'd let them choose among them, and place them on their preferred model wherever they'd want. I'd have them discuss why. I would then go back and improve the fidelity of the concepts, by sketching and model-making. I'd then do another round of research.

Here's why that process beats generative design:

1) You want physical models not renderings, and today's rapid prototyping software just isn't as efficient as a designer carving a bunch of concepts out of for a few hours.

2) You want to separate your research variables, and progressively-disclose choices to the user. In your MP3 example, you've created a bunch of renderings, but you'd really need to create thousands more to cover all the variations. That's just not practical for the research participant. Rather than give them 10x10 choices, you want to give them 2 choices of 10: pick your shape, pick your controls.

3) You want your subject to co-create with you. Give them a bunch of controls and have them choose what they are and where they go, and tell you why. This is cheaper, faster and gives you more meaning.

[EngineerErrant]

For example, if an Industrial Designer were to design an MP3 player, a good design process would be to use "generative research" (aka co-creation) to help isolate preferences among those customers. In an hour, I would make a bunch of wood or foam shapes, and hand them to research participants...

And where, pray tell, would you obtain these shapes?

-Josh

[cdaisy]

I want it. Is there still a module for Pro/E?. (sorry to be lazy and not look it up but I'm a busy guy!)

I can't believe some people here are missing the point. This is a happy accident machine that can give you hundreds of subtle variations in no time. Obviously you won't use it 100% of the time, but if you can't think of a few good uses for it you should quit the design field.

Then again there are IDers that dismiss the power of CAD and refer to renders as "just pretty pictures", so I guess I'm not surprised.

[cdaisy]

1) You want physical models not renderings, and today's rapid prototyping software just isn't as efficient as a designer carving a bunch of concepts out of for a few hours.

I don't understand this statement. You're saying its more efficient to carve objects by hand out of foam instead of modeling it and sending it to a desktop printer or the like? Cheaper maybe, but more efficient? You must be one hell of a whittler!

[Electroflux]

BTW, the excel macros from the opengenerativedesign site have a few issues. If there's interest I can post the one I'm using, but if you're familiar much with scripting it shouldn't be hard to fix up.

Though I think I'd rather move this into a SW macro. Does Pro/E have a scripting/macro utility?

[NURB]

For example, if an Industrial Designer were to design an MP3 player, a good design process would be to use "generative research" (aka co-creation) to help isolate preferences among those customers. In an hour, I would make a bunch of wood or foam shapes, and hand them to research participants...

And where, pray tell, would you obtain these shapes?

-Josh

What? Where do you think?

I don't think CG was taking issue with the output (hand carved, rapid proto, etc.), his issue was the disconnected nature of flushing out 1000 variants of one design without any thought to it. His process will get you from concept to completion with a much better outcome. The shotgun approach rarely works well.

[cdaisy]

I think it depends on what you are using it for. The image of the spoons is a good example. A spoon is a spoon. Do you really need a ton of research to design a spoon? I would say not really, but you do want a shape that is both functional and pleasing to the eye. So if I can get 1000 variations quickly and simply pick the one I like the most, I would say that's a pretty efficient use of my time.

Isn't that the argument people make for sketching being so important? Showing variations of the same object quickly?

[NURB]

Of course, but as someone else said you'd need to teach the computer your aesthetic in order to get the same result. It would have to learn from it's visual mistakes, and build on a "happy accident" that makes the for more pleasing. That's something I doubt you will get by simply modifying a few parameters and pushing Go.

[Electroflux]

Which is why you'd somehow need to build on what's going on here.

As I said before, the architects aren't just pressing a randomize button either. If they were, few would see the value of this either.

They have a question they want answered, they may know roughly what the answer should be, but the human-time to get it is quite large. The computer can just create 1000 variations, run physics simulations, and delete the 900 that fall apart.

But teaching it aesthetics is a bit different than teaching it to make a strong roof.

[EngineerErrant]

I don't think CG was taking issue with the output (hand carved, rapid proto, etc.), his issue was the disconnected nature of flushing out 1000 variants of one design without any thought to it. His process will get you from concept to completion with a much better outcome. The shotgun approach rarely works well.

The point I was making was that having generative design software to create things like truly random shapes is exactly the sort of tool he'd need for the design process he's talking about.

I think the shotgun/rifle dichotomy being set up is meaningless. There's an element of randomness requisite in creativity, and generative design is great for that. It's not like the final product is being spit out by the program and sent straight into production with zero analysis with regard to usability or aesthetic or structure or whatever else. (On the flipside, it's not like there isn't a randomly-generated-shapes aspect when we sit down in front of a piece of paper and draw shoes. Whatever we've got going on the radio gets assimilated and processed.)

Re: algorithms & aesthetics, exceptionally simple evolutionary algorithms can do amazing stuff. If you think that we couldn't generate aesthetics with user input, you've clearly never used Pandora.

-Josh

[Electroflux]

exceptionally simple evolutionary algorithms can do amazing stuff.

Neat. It takes a while but eventually it figures out a "bicycle" is the best solution with a "tower" shape just behind, until the tower wobbles over. Also reminds me a bit of this: http://sodaplay.com

[SK]

The 1000's of meaningless designs were likely done without any kind of design process. They were done with a "shotgun" approach which is common in the East where manufacturing is cheap. The theory is that you flood the market first with as many inexpensive variations of a product as possible and see what sells. In the West, it's more common to have a "rifle" approach, which utilizes a lot of research, strategy and design process to ensure the product hit's it's mark, reducing risk.

Generative software is great for the meaningless "shotgun" approach, but designers strive for the meaning-full "rifle" approach.

Now, this is a serious point of difference. In the early days, flint lock guns were as dangerous to the user as the enemy. Shot guns do scatter and target a larger area. But it all depends what you are shooting at. If you are going for ducks then shot guns will do, as ducks fly in a group (products too are similar and belong to groups). So they are many satisfactory solutions (its ok if your bring one of those flying fellows down). With a rifle, if you get the target wrong it is a miss. Not so with a shot gun. Its wrong to assume that you know exactly what will succeed, so a rifle approach is a gambling approach. But gamblers always believe they will win - else they won't gamble. But companies have less and less use for gamblers. This is a reality that few would argue with.

The description of your own design process is useful, but it is the best ? You don't seem to be using any of the capabilities that computers can provide you.

Here's why that process beats generative design:

1) You want physical models not renderings, and today's rapid prototyping software just isn't as efficient as a designer carving a bunch of concepts out of for a few hours.

Often you need a combination. Desktop printers are going to be sub 1K in a few years. So if you want to prototype a lot of designs then you will find generative to be most useful.

2) You want to separate your research variables, and progressively-disclose choices to the user. In your MP3 example, you've created a bunch of renderings, but you'd really need to create thousands more to cover all the variations. That's just not practical for the research participant. Rather than give them 10x10 choices, you want to give them 2 choices of 10: pick your shape, pick your controls.

The range of designs that you can generate are in the billions. So you will never cover the entire range of possibilities - even if you wanted to. It is up to you, to narrow it down to a few that you like to show your clients.

3) You want your subject to co-create with you. Give them a bunch of controls and have them choose what they are and where they go, and tell you why. This is cheaper, faster and gives you more meaning.

Agreed. Generative design is perfect for this. Because genetic model can also be driven by consumers (replacing random inputs). Pls take a look at the workshop on allowing consumers to co-create.http://www.genometri.com/DIY/
another example in mass customization http://no-retro.com/home/2009/11/01/fro ... rs-part-2/

I hope I have convinced you of the merits of generative design.