Generative vs Parametric Modeling

Replies to This Discussion

Permalink Reply by Sylvain Usai on September 6, 2013 at 3:28pm

There is a great line in the book "generative design", basically saying that generative design is more about phenomena recognition more than formal research. I don't have the exact sentence in english anymore but here it is in french : 

« au fondement du design génératif ne se posent pas des questions formelles, mais la reconnaissance de phénomènes[1] ».

---

[1] Julia Laub, Hartmut Bohnacker, Benedikt Groß , Claudius Lazzeroni.- Design Génératif : Concevoir Programmer Visualiser  – Éditions PYRAMID, 2e semestre 2010, Paris. p.5

                titre original « Generative gestaltung », Hermann Schmidt Mainz, 2009.

This mean you design a process rather than just one final object. In this way, grasshopper is generative.

But as you mentionned, it's often strongly linked with emergence, and therefore, with simulation. For some academic writing, I was (i'm still on it)  trying to define the difference and complementarity between parametric (as animation) and agent like system (as simulation or emergence).

In my view, Animation refers to the variation of a reference element to fit in a certain context (non standard connexion node in a beam network for example), this use is well exemplified by DigitalProject's "PowerCopy" or any grasshopper geometry defined on a reference surface that follows the reference variation. So parametric via animation would be the "distribution of difference"

Simulation refers to circle packing, dynamic relaxation, agent based modelling, etc. The resolution of a global problem at a local scale. Emerging design via simulation would be the "distibution of complexity"

With these definition, grasshopper enters in the animation (and so in the parametric) definition and not so much in the simulation (and generative design as an emergence tool) but kangaroo would be more about simulation than animation. 

my interest in the topic came from these lines by Roland Snook : 

« Parametric models are structured hierarchically, however, having direct cascading, causal relationships –an obvious impediment to this description of generative design [designing process rather than artifact]. The parameters within these models -the ubiquitous sliders in software programs epitomize- confine the model to a known set of limits. So while parametric models enable a distribution of difference, this is not the difference that emerges from intensive processes, but rather a directly described, top-down, smooth gradient operating within a predefined range. Here, all possibility is already given within the starting condition[1] ».

---

[1] Roland Snook.- Volatile Formation, in ‘Reclaim Resi[lience]stance // ……R²’ Log°25, summer 2012, edité par AnyCorporation and MIT Press, pp.56-62

Permalink Reply by Dimitrios Papadopoulos on September 6, 2013 at 3:50pm

But if we accept that "you design a process and not an object" what better description of algorithmic than that?

Excluding the "Parametric models" from Snook's text, his description is very descriptive of the process we are talking about. I like his link to "intensive" vs "extensive" thinking, (with all it's historic/physics background), and the clarification that "all possibility is already given within the starting condition".

I like your "animation" approach. I would expand/compliment the "fit in certain context" with something like "adapt to unpredictable changes of the forces that shape it"...or something to indicate that the need to adaptability is not only context driven. Unless you expand "context" as the whole of the factors that affect the result.

Permalink Reply by Sylvain Usai on September 6, 2013 at 4:22pm

- But if we accept that "you design a process and not an object" what better description of algorithmic than that?

Right! And that's why I was first really puzzled by Snook's or Alisa Andrasek's writings. When you take a look at their work and writing, they seem to reject entirely the use of parametric and produces fancy shape you can only fabricate with a 3d printer at small scale ! Hopefully, auto-organisation systems are also use in "real world" projects (see Fabian Scheurer's article) but as a starting point for a parametric work and I am not even sure he makes a distinction between parametric and generative...

I realize I really miss a complete definition of "generative design" as the "design process rather than artifact" seems too wide.

To add some bullets, "In practice, when we look at computation, there have been two schools of thought. Both have emerged from a fundamental difference of control – parametric and generative – and they reinforce the top-down/bottom-up discussion in the most primitive form. With my students we’ve been speculating in the domain of the generative." (Theodore Spyropoulos in 'John Henry Holland and Theodore Spyropoulos in conversation' published in Constructing adaptive ecologies : notes on a computationnal urbanism). ... So there is for some people a difference : parametric as top down and generative as bottom up ? (I think I'd dig more into Spyropoulos, he should define generative and parametric but I may have forget it)

- "adapt to unpredictable changes of the forces that shape it"

if by "it", you mean the context and not the instantiated object. By context, I was meaning "local condition" : if you draw a line between to point, the context is the two points... "reference input" may be more relevant (?)

"the need to adaptability is not only context driven" what would it driven by then? I think I agree with you but I am not really sure to understand...

Permalink Reply by Sylvain Usai on September 6, 2013 at 4:38pm

(I write as a new reply not to be confusing with editing if your are writing at the same time)
As I understand Snook's critic, parametric is arbitrary defined (you set the slider) while generative is... less arbitrary (you still have to define the rules)

Then, where does genetic algorithm belong? Is it a top down approach as the user define the fitness ?

Permalink Reply by Dimitrios Papadopoulos on September 6, 2013 at 4:42pm

Spyropoulos is always a very valid source. 

"the need to adaptability is not only context driven". I wasn t disagreeing with the context change. More I was trying to point out -as I always found very interesting- the variety of reasons ofwhy the context changes. Could be structural reasons, could be location/site change, client changes his/her mind, designer changes his/her mind etc. etc.

If interested to historically trace some critical projects that paved the way, among others, a very important historic reference, would be Takis Zenetos' "City Planning and Electronics"

http://issuu.com/angelosfloros/docs/architectonika_themata_003_1969

http://issuu.com/angelosfloros/docs/architectonika_themata_004_1970

http://issuu.com/angelosfloros/docs/architectonika_themata_007_1973

Permalink Reply by Dimitrios Papadopoulos on September 6, 2013 at 5:05pm

Snook's description is pretty accurate in it's own context when speaking about the dominant notion of "parametric design" by saying that all the results "pre-exist" when you set the initial rules. in the sense that you will not really be surprised from any result. Even if the results are infinite. In fact in practice you know the results already (or at least the "kind" of the result, let s borrow a term and say "phenotype"?, and you want to build a dynamic model in which the local values will adapt to an overall schema. In that sense I would risk to say it s a deterministic model. 

But maybe using the word "rules" is a bit confusing as usually it s used for initial rules someone sets for a system to evolve, bottom up. Very thin differences, but usually in practice in GH you would say/think you define "degrees of freedom", "constrains", "relationships", "parameters" and "constants" etc.

A wide definition of top-down and bottom-up design/approach can be found in Wikipedia or around the net, and I believe there is no room for misconception. In genetic algorithms the user defines the fitness for sure as the user defines everything anyways as there is always a user short of speak, external to the process. But theoretically the system evolves afterwards without a central control.

• ‹ Previous
• 1
• 2
• Next ›
• Page