Great Ideas Travel Slowly : Review of Generative Design 2015

Great ideas travel slowly, and for a time noiselessly

- James A. Garfield

Generative Design used to travel with a lot of noise, with academics and CAD companies heralding its arrival. But not any more. The noise seems to have died down. It is traveling noiselessly now, perhaps into a more serious stage with very positive developments in its favor.

Scripting takes center stage

Autodesk succeed in the architectural CAD market by presenting architects with computerized drawing boards – that hid the code, assuring them that nothing has changed. Five decades later, the code is presented for architects to play with in the form of DYNAMO following the successful example of Grasshopper with a dynamic and dedicated user community. The growing popularly of both these platforms is making the architectural community come to terms with reality – of the coded nature of design. This is an essential development for the acceptance and propagation of generative design.

Integrated Data

There is sufficient noise making now many decades later on the virtues of considering designs as information model instead of computer drawn lines – generating significant efficiencies in the design process, making coordination amongst the various related disciples easier. The dominance of this approach will allow generative representations of design to carry much better quality of information assessable across different disciplines.

Light weight representations

The advent and popularity of apps is creating an entirely new type of light weight CAD representations ( I like to call them design avatars) . These representations are connected by API to heavy duty CAD representations of design. This is allowing consumers to create architectural components on tablets opening up architecture to end user creation. This again is a very positive development in opening up the selection part of generative design processes to end user input.

While these are some of the positive developments, there are also some popular pipe dreams

Cloud dreams

In the imagination of some CAD companies, design can be improved by creating endless random variations and analyzing them with turbo charged cloud hosted analytical engines. This certainly fits nicely with their cloud hosting based business model. It works well with the current practice of running that 1 millionth iteration in search of that 0.2% improvement that most engineers are familiar with. The underlying assumption here is that the complexities of architectural design can be resolved through more storage and more analytical power - without changes to design representations and design work processes.

Current challenges

Nobody really knows how to represent early stage design in variable form while most CADist are able to represent late stage design in parametric form. It is also not clear how such early stage design representations can be matured into fully fledged BIM models.

There is still very little consensus on definitions surrounding computational and generative design. The architectural computational academic community continues to fail in in resolving these terminological issues. There are also no signs of significant contribution to computational or generative design generative design being made by them with significant number of MSc and PhD students recognizing this deficiency.

In is also unlikely that CAD companies servicing the needs of established companies using old world design techniques will come up with anything other than the integration of existing solutions in more useful and marketable forms - requiring more cloud hosting and more computational horsepower.

An HTML future

I speculate here, that the next generation design technologies will evolve within web browsers (thanks to the greatly enhanced representational capacities brought about by HTML5) making best use of connectivity, shared smart representations (genetic models), relying less on storing dumb data in large volumes and analyzing marginal changes with massive computational power.