Archive for the ‘Genetic Design’ Category
Future breakthroughs in CAD is more likely to come from the design tools for creating microscopic wet wear than from the crude tools known as CAD designed to combine metal and concrete into architectural artifice. Because the latter is bound to be dumber. We wish it to remain that way servile to our limited imagination.While the design tools for cells, organs and organisms are being designed to operate beyond human imagination and in the mode of discovery – opening up unimaginable possibilities.
Interesting both Bio and Non-Bio CAD started off life similarly; representing geometry and design data. Ah.. then they got cleverer as they attempted not only design, but to make design better – one objective at at time – usually through optimization. Both CAD systems now do single criteria optimization reasonable well, because before you optimize you need to define a “design problems”. Once you make design into a “problem” there is no problem there after. All you need is to call in the engineers – they all now carry with them a bag of tools for optimization. But even then, they can only do “single criteria optimization” in other words solve only one problem at a time.
The trouble starts when you have more than one criteria and that is a real problem; because all real design problems are multi criteria problems. Here the “problem sovlers” have a real problem. Though some may retreat into a theoretical multi criteria problem solving mode, most of them know that they have no idea how to solve it. And it is here that Bio CAD is beginning to overtake, because those who fashion it know that problems and solutions don’t appear at once – they are grown through a complex developmental process.
Next generation Bio CAD
From a review of genetic computational tools
The first generation of tools dealt predominantly with singular objectives such as codon usage optimization and unique restriction site incorporation. Recent years have seen the emergence of sequence design tools that aim to evolve sequences toward combinations of objectives. The design of optimal protein-coding sequences adhering to multiple objectives is computationally hard, and most tools rely on heuristics to sample the vast sequence design space.
The article revives a wide range of CAD tools and there are some interesting similarities with architectural CAD. Interestingly most of the CAD tools are open sourced, community based and operate online. Some of them like clotho are platforms (like grasshopper) allowing users to create component CAD systems that they can create, share and connect.
There seems to be a clear evolution here. Bio CAD seems to be moving from rule (or grammar) based creation tools into tools that can handle some of the complexity of biological design processes which are still being discovered by scientist. Hopefully from these efforts we will learn the wonderful ways that nature exploits to explore vast design spaces that allowed it to fashion complex organisms like our selves capable of understanding our own making.
Cells contain massive amounts of information. If we stretch the total DNA in our bodies it will be about 16 to 32 billion kilometers. Now, that is a lot of code.
We cannot pack in more information than that. If we were to include the exact location and dimensions and geometric details of the circulatory system for example, as we would do in a CAD file, it would require more than trillion kilometers of code. Hence, nature constructs such designs with code. This is beautifully explained by Prof.Robert Sapolsky’s in his Stanford lecture.
Why use Algorithms ?
Algorithmic code is good for creating very complex geometries with small amounts of data. It works very well with the way nature constructs using cellular components. The fractal (or self similar) nature that you see in trees and leaf veins and arteries is due to this. But the code here is embedded in the cell itself and cells organize themselves to create complex forms based on of relatively simple code.
Despite the significant interest we have in nature as a source of design inspiration, we do not adopt her design methods, except for genetic algorithm, developed by John Holland in 1975. This however is an optimization method and not a design method. Her design solutions however, have begun to inspire the design of new products through the emerging field of bio-mimicry. But these are based her vast repertoire of design solutions and not based her design methods.
Even though nature’s design processes are now known – they remain purely as a source of inspiration. Why then are her methods of no practical use to designers? I am beginning to suspect some fundamental problems. Here are some of them :
Natures has no intentions
Design seems to be by definition a human driven process. Humans have intentions. Nature does not. God on the other hand may have intentions, but if he cared about nature he would not have created us. The book the “Selfish Gene” (published 30 years ago) illustrates the claim that it is not life forms but bits of code that compete for self replication – which could then be seen to be the only intention if there was any. So nature is into code play. Disturbingly, nature’s design processes seem autonomous and direction less, and worse it is driven entirely by the selfish propagational interest of bits of code – that hitch hike on living forms.
This model of autonomous and mutually dependent conglomerate of code competing with each other to propagate seems to be an impractical and uninspiring model for designers to adopt. More depressing is the fact that the resulting biomass that we so admire is only a packaging for the all important the bits of code to propagate itself. Once the packaging is gone past its usefulness and makes errors in replicating the code, it is dully discarded (suffering death) while the code moves on to younger packages that can do a better job at replication and propagation. The deviousness of this strategy is nauseating. Our body bags are nothing but code replication devises. Whats even more annoying is the fact that the intentions of these bits of code seem to be independent of the bodies that carry it. A good part of design history remains in great awe of this package and it is hard to think that the very brain with which we understand this i,s just a small part of this package. I am not sure if we will every come to terms with this.
I vaguely remember reading his book which you may remember for its blinking led lights in its cover, which was fairly weird then for a book on architecture. It was a pleasure to hear him speak in the design conference in Nov. In such conferences there are often the established and known – their views are known and most often, they have nothing new to say; not that they said much earlier.
Then you have the cutting edge folks – whose presentations sound like teenagers discussing sex, “I did that this and that, and then…..” listened intensively by an equality excitable audience ready to applaud the finale of resulting in orgasmic geometric forms. Generative design, has sadly become the means through which such geometric entertainment is now effortlessly created, leaving little room for restfulness or reflection, or any form of serious thinking for that matter. I wonder sometimes if the refusal to be easily aroused, is an “age thing”, being no longer a teenager and having to deal with them instead.
As thoughtless forms take over the screen and as I hear freshly spun design philosophies blurted out with the accompaniment of architectonic lullabies, it provided for me – the perfect time for a conference catnap, only to be woken up by Prof.Frazer. His lecture was delivered with the thumping energy of a British steam engine. You can see him live in an AA lecture. The things he had to say were of interest to me and perhaps I thought, to the readers of this blog. So I approached him after his lecture and kindly, he agreed to be interviewed.
It is refreshing to see promising research and useful methods emerging from lesser known quarters. Despite decades of academic research the “layout problem” as it is called, is till today solved by intelligent guess work.
Christian Derix, director of Aedas R&D Computation Design Research (CDR) group seem to be close to cracking a rather long standing problem. A problem that obsessed early design researchers from the end of world war 2. Architects returning from the war seem to have been keen to shake out the “irrational: image of their professions. Their engineering colleagues got back to peaceful production and were focusing their efforts on improving production and making it efficient. 50~30% of production costs are attributed to what is called transport cost, or the cost of moving material from one place to the other.
The cloud brings together the possibility of massive computational resources and connectivity in an unprecedented scale across a wide range of business, educational and entertainment activities.
Are the Architects ready for the cloud?
The answer is ” No”. But, will they get there? ” Yes”. Most likely, in the same wrong way they adopted CAD – to replicate the drawing board with CRT screens – without consideration to the true potential of computers. This was a big jump for many architects. It happened only because they were assured that the Cathode Ray Tube (CRT) was better than the drawing board. Companies such as Amazon, Microsoft, Apple and Autodesk are all now busy building the rail roads in the kingdom in heaven – in which platforms of great promise will dominate the next era of human dependence on computation. So, everyone will get there for sure.
But what will design be in the cloud ?
I believe that the cloud will initially be used in the same way that computers were used to replace existing PC based practices. PC bound CAD systems will soon be operating on cloud platforms. Speed and connectivity bonuses are good enough to lure most CAD dependent designers. But once they are all there, it is likely to transform the practice of design in way that it was transformed by the PC/CAD revolution. But then, without them realizing it, the clocks will be turned back on them. Design processes will go back a few billion years – to where design began.
Design will be – as in nature
Nature in itself is a massive computational environment that has evolved over billions of years. Its key virtues of building complexity based on shared code and ability to explore possibilities through random exploration using highly evolved strategies and methods will come to dominate the art of design – orchestrated by human designers, the way humans have harvested the potential of natures design capability to turn grass in to wheat and rice and wolf into dog; primarily by manipulating a highly evolved refined and structured design processes.
Design as it is now
Before we consider the lofty heights that clouds can take us to, let’s review where we are with CAD now. The turbocharged drafting machines now connected to data-bases powered graphically by games technologies have got us quite far. A diverse set of capabilities and professional work practices – are now slowly coming together; but mostly at the back-end of the design process. But here it is too late, as all the important designs are already made and opportunities to make significant improvements are limited. It is known that more than 80% of decisions and commitments are made in the early stages of the design process (shaded in green) where now computers play a very limited role.
The codification and comodification of CAD
Most CAD packages now handle the drudgery of 3D manipulation fairly well. The dark regions shown – is dominated by code that reduce design labor ( most CAD companies have similar capabilities in this area). The push now, is into early stage design, where significant improvements can be made. Software like Grasshopper and platforms like Vasari are now extending the reach of CAD into early stage design. Further up stream is generative design.
What the clouds mean for generative design?
It is like asking what gasoline means for your car? Generative design can drink it all – all the computational capabilities that the cloud can provide. It will soon be possible to roll apparently dumb, random and computationally intensive approaches that nature has chosen in its great wisdom. Hopefully, it will be based on an open and shared genetic infrastructure – so that knowledge generated will not be lost but be shared and built upon.
The fundamental change will be the ability to consider multiple possibilities in virtual environments. In sharp contrast to the singular and somewhat perturbed linear approach mastered by designers on account of their limited mental processing capabilities. The design processes now used by designers are based on the limits of the processing capability of the human mind and its ability to consider only a large but limited number of possibilities. Kasparov is no longer the champion of chess.
The maturing of many CAD technologies has already greatly reduced the human labor in taking early stage concepts to reality that is close to real, making it possible to consider multiple possibilities of great maturity – instead of dumping them at the end of a doodling process as part of an ancient design ritual.
Customisation is often an afterthought necessitated after the product is launched, bringing with it the pains of late adjustments.
This is now changing. Due to market saturation products and services are being designed for customisation. Generative Design has a defining role to play in this.
CAD was first developed to replace paper based design. A piece of paper can hold only one design at a time. But nuts, bolts and many mechanical parts shared similar geometry and were only differentiated by dimensions. So engineers got efficient. They created table driven configurations. Then, marketeers found going back and forth between engineers accountants and production managers a drag. Online configurations were stitched together based on existing work process to help customers wants to what companies can offer. But underneath it all, an even greater, more powerful phenomenon was simmering.
In its ideal form, it would eliminate altogether the human involvement in mitigating between what consumers want and what companies can produce. The growing collection of web-based configuration technologies can now ensure that Joe, the customer, can create something that is both useful for him and viable for the company to produce and support – all by himself. But it is possible mainly due to a series of hidden rules that prevents Joe from doing the wrong thing; so that Joe does not configure a laptop that he cannot carry home. Most customisation solutions today are based on rules.
Why rules are bad
It is known that even simple combinatorial design problems can lead to not millions but trillions of possibilities. Currently, this nightmare of choice is narrowed down by writing rules. Often, hundreds or thousands of rules will be required to produce a decent set of viable designs.
Rules are there to avoid confusion. They need to be simple and straight forward. Hence they are crude. They are good when the context of its application is simple. If it is complex (as in design) you need to make exceptions (or clauses). Therefore, rules tend to multiply rules. Worse still, difficult to judge the effect of one rule on another or on the design itself. Rules written in one domain will affect another. Worse still, only experts know how to write rules. Even if you mange to write them – you are left with another big problem. You need to generate the designs out of these rules. But despite their limitations, rule based configurators are very much in use – because that is currently the only way to trim solution into a viable range.
Setting limits is easier than setting rules
What is now achieved though patchwork methodologies may be achieved much more elegantly – if the problem of configuration is resolved at conception – directly on the representation of the product. Genetic modelling can do that. It will allow designers to conceive designs that are fundamentally configurable. Genetic modelling will allow them to represent 100o’s of design possibilities based on a single CAD model. Instead of waiting for customer requirement (that now force companies to create variations) it is now possible to create ‘variable models’ as part of the design development process.
The walled garden
The holy grail of consumer creation is for companies to provide the greatest choice to consumers, allowing them to design/configure products according to their own individual requirements. The danger here is that consumers may come up with designs that are dysfunctional, un-manufacturable or beyond their means. Hence, companies provide not only choices but methods by which consumers can make intelligent choices. A combination of genetic modelling, filtering and performance measurements working jointly can crack this problem elegantly – keeping Joe within a walled garden of constrains.