Mindful Algorithms: the new role of the designer in generative design

Note from the Editor, Tricia Wang: Next up in our Co-designing with machines edition is Che-Wei Wang, (@sayway), is a designer and architect who co-runs CW&T, a design study with Taylor Levy. In this post, he contemplates why engineers and architects will need to become more like ethnographers with generative design. He asks if it is possible to convert ethnographic data into quantitative data as algorithmic input. I’ve long admired Che-Wei’s ability to bring a poetic quality to the deeply mathematical nature of his work whether it’s in architecture or designing beloved products such as Pentype A and Pentype B. He’s currently an artist in residence Autodesk. Most recently, the Collective Design Fair featured CW&T‘s designs, followed by an article written about their work in Coolhunting.

The traditional design workflow is getting a turbo boost from algorithms (don’t worry. The robots aren’t taking over…yet). With new types of generative design processes like genetic algorithms, the designer’s role is changing from the traditional, top down approach of drawing ideas on paper, into a systems approach.  Designers traditionally sketch and develop ideas intuitively. With a genetic algorithm, instead of imagining a design solution, the designer develops a fitness criteria and coaxes the algorithm towards a final design.

As algorithms and data become crucial tools to a design workflow, ethnography will need to become part of the process. Engineers, designers, scientists will all need to become ethnographers. The best cab drivers know how to work with GPS navigation. You sometimes have to ‘trick’ the algorithm to get the best result.

I’m an artist and designer with a background in architecture.  I’m currently teaching a studio at Pratt Institute School of Architecture that’s attempting to integrate genetic algorithms into the design process in a meaningful way.  I started teaching this class primarily as a reaction to all the highly ornate generative design work that I’ve seen over the last decade. These algorithms are fetishized and have been used to generate highly articulated forms like swoopy skyscrapers with windows that vary in shape and size throughout the facade.  The question that always comes to my mind is…to what end?

Recent developments in software have started to shift generative design processes to incorporate environmental factors like sun radiation and structural forces to create more functional geometry. But, the question remains…What other forces and factors should be tied into the generative design process to create designs that respond to a site or a condition in a meaningful way?

Designers have been traditionally trained to conduct research, sketch ideas, refine ideas by moving between sketch, computer modeling, and prototypes. The designer in this traditional workflow does all the data processing in their head.  As algorithms increasingly become part of our workflow, the data will have a more direct effect on the outcome of designs.

How will algorithms change the design process? How do designers need to change their mindset to take advantage of algorithms? How does design need to change?… First I want to tell you what generative design is, and then give you an example, and tell you some thoughts I’m having about it and ethnographic data.

What is generative design? And the genetic algorithms?

Generative design is wide term encompassing any design process that involves algorithms in the design process. It’s often used to design complex shapes and optimized forms in relationship to forces, sun radiation, and various data that may influence the design.

I have a keen interest in genetic algorithms (GA) and its use in the design process. GA is an algorithm that evolves an optimized solution by starting with a random population of possible solutions, and generates new solutions based on the best solutions of each generation. It’s a widely applicable algorithm because all it needs is a random population, a fitness criteria to test against, and some mutation in each generation, so it continues to evolve over generations.

Here’s an example of genetic algorithms – my bike stem design

The current workflow of using a generative design software package like Dreamcatcher (Autodesk Research) consist of coming up with a design space with parameters, running an algorithm, and choosing from the iterations that the algorithm generates.  In essence, the algorithm generates a form out of thin air based on a fitness criteria that you’ve set.

Take for example this bike stem that I designed in Dreamcatcher. The problems that need to be solved around a bike stem design is relatively easy to define. It needs to withstand the forces at play and it wants to be optimized to be as lightweight as possible. In Dreamcatcher, you specify “ports” (geometry that the generative algorithm needs to connect to) and “obstacles” (areas that the generative algorithm should avoid).  Once you’ve declared your ports and obstacles, you can apply forces to your ports.  In the case of a bike stem, there are ISO guidelines for testing bike stems, so I used them to calculate the forces in Dreamcatcher. Aside from those main parameters, there are several others variables to tweak that greatly influence the outcome of the generative design. So it’s not quite as simple as setting up the parameters and clicking GO. It takes many trials of tweaking parameters to achieve the a satisfactory design. It might provide a solution that fulfills all the criteria you set, but it may not have the right proportions, or it might place geometry in places you never imagined. Over time, I’ve learned how the algorithm responds to the changes I make, so I can steer the design.

I chose a bike stem because it has a few inputs that lead to a tangible outcome. It has the potential to accept more inputs if you were to customize it for more specific/narrow riding conditions. It’s important to note that algorithms aren’t good for designing everything…yet.  And computers are really dumb. It only knows what you tell it. So the data that you input is crucial to the outcome of the design.  Genetic algorithms are good for solving problems with a well defined fitness criteria (what you want the design to do) and a large solution space (lots of possible designs that meet the minimum criteria)

If we were to go through a the traditional route of designing a bike stem, we would sketch a bunch of possible “new” designs.  Refine a few designs in CAD. Prototype a few to test for performance and repeat until we feel like we have something that we like.

A bike stem has easily quantifiable constraints and criteria.  But imagine a generative process applied to designing a building.  Instead of a few simple set of forces and a singular fitness criteria, a building has countless physical and metaphysical forces that need to be taken into account while serving and catering to multiple desires and needs. The current climate of output from generative design processes in architecture is a mix of hyper ornament, force driven topological optimization, and sun radiation optimization. I have yet to see any algorithm in architectural design use data derived from the cultural context of a building’s site.

Why I think genetic algorithms is such a big shift for design:

At this moment, we are at a crucial intersection between the increasing reliance on algorithms, wide availability of data, and computing power.  The three key ingredients are becoming mainstream so it’s important to think about how we incorporate the three in the design process and set precedents.

Here are two ways that I think generative design can benefit from more ethnographic data.

1. Creating better data inputs  – finding the data that matters the most

The challenge we face now is to figure out what data we use as inputs for the algorithms. It’s becoming easier to incorporate easily quantifiable and readily available data, but human factors are often hard to quantify and incorporate.

Projects often use data that’s available rather than data that’s relevant.  We’re still swimming in the novelty of big data. We use data, but we aren’t examining it. We take what’s available and apply it as if it’s the truth and as if they are the only things that matter. If a building is built to serve a community, how can the arrangement of the spaces and the geometry of the facade be designed in response to the needs and desires of a community?

2. Finding the best solution – helping the designer make the selection

In design, “best” is subjective, and it can’t just be based on easily quantifiable data. For example, to design the best bike stem, you have to understand an individual cyclist’s needs in terms of performance, aesthetics, safety, material use, manufacturing constraints, riding style, fashion, etc. It’s not as simple as applying forces after all.

There is always data that isn’t quantifiable – if you’re talking to people, not 100% of every interaction can be converted to data. As a human being who has visited a site, soaked in the environment and understood the problem with your 5 senses, you have more knowledge than the algorithm – so you have to sometimes make a decision to help the algorithm or the algorithm can help you. It’s up to the designer to select and configure the algorithm to arrive at a solution.

The UI for generative design is evolving quickly to take advantage of computing power. You’ll input data and the software will give you a bunch of solutions mapped on a graph, you’ll assist the algorithm by making a selection from the bunch. It might be easy to think the designer’s job is to simply make a selection, but the selection of data inputs has greater consequence to the outcome of the design.

What is role of designer in all this?

The designer will become more of a generalist – the algorithm is iterating and testing thousands of possible designs millions of times faster than a human.  But the algorithm is only as good as its inputs and fitness criteria. Future designers will be less like airplane pilots and more like air traffic controllers.

Designers have to be generalists to know enough about all the inputs that need to go into the design. They can’t be blind to the data and assume it’s right.  This is a hard problem to fight because current users of generative design tend to be experts of a specific domain so they are using familiar data rather than the right data – eg a structural engineer will take force data to come up with a structural solution for a building but fails to see how the structure influences how people move through the building, which in turn affects social spaces in the building.

Moving forward:

To be clear, I’m not advocating the replacement or eradication of authorship from the design process. There’s still quite a bit of art to a generative design process. An algorithm with the right set of inputs and fitness criteria can provide thousands of optimized solutions. I think this trend is similar to the way photography has evolved. We used to spend most of our time setting up the scene to take a few photos because film was relatively expensive.  Now we shoot thousands of photos to be edited down later. Generative design offers a new editor-like approach to design.

Important design decision are being made without the best data sets – without full context, often only on what’s available.  We need to think more ethnographically to build our own data sets that are more granular and meaningful to the context of the design. The big hurdle is to come up with a process that converts ethnographic research into quantifiable data so the data could be used in algorithms.

I think the problem we’re facing is the speed at which algorithms are being adopted into every process from finance and design. The speed of adoption is faster than we know how to use algorithms – so we are scrambling – we find data and we use it.  Those who use the data aren’t creating the data or scrutinizing the data enough. We’re searching for data before asking the right question.

If we can find a way to incorporate meaningful data into a generative design process, design might have a chance at making the world a better place.

This article is part the Co-designing with Machines Edition. Read other articles in this edition. 

Like what you’re reading? Ethnography Matters is a volunteer run site with no advertising. We’re happy to keep it that way, but we need your help. We don’t need your donations, we just want you to spread the word. Our reward for running this site is watching people spread our articles.  Join us on twitter, tweet about articles you like, share them with your colleagues, or become a contributor. Also join our Slack to have deeper discussions about the readings and/or to connect with others who use applied ethnography; you don’t have to have any training in ethnography to join, we welcome all human-centric people from designers to engineers.  

Translation

English中文(简体)FrançaisEspañol

by

You can help improve the automatic translation by checking off “Edit Translation” after you select a language. Thanks!