Is generative design doomed to fail?

Published on November 1, 2020 by Erik Pols.


Yesterday I ran into this article by Daniel Davis who takes a fierce stance against the current trend in architecture where architects and property developers are exploring what generative design means for our field.


Obviously my first reaction was one of disagreement. At Vellum we are making huge progress to automate the building design process using generative design, and we’re proud of the recognition of our clients so far. Vellum is able to generate buildings with a few clicks. Aren’t we proving Daniel wrong?


However, when I read the article more carefully, I actually found myself agreeing with most of the arguments Daniel is making. The article is worth reading, but for convenience let me summarise the main points here.


Generative design systems are limited in their options by nature. Daniel argues that generative design systems simplify the problem by limiting the amount of options that can be adjusted, which often results in a very uniform set of generated designs.


Quantity doesn’t substitute quality Generative design systems can generate hundreds of options, but just as you can generate hundreds of paintings in mere hours, this will not mean any of them will be necessarily good.


Comparing options is hard work What defines the best option is often not a matter of picking the highest or lowest number. What defines a good building depends on hundreds of factors, and therefore comparing options is as well. You could even go as far to say that the actual numbers don’t matter so much in architecture (what you can measure isn’t what matters).


The ideas behind generative design are old already, however it is only recently that we’ve seen companies embrace initiatives in generative design. Many of those do exactly what Daniel warns against: Press-of-the button, no-human-necessary generation of hundreds of options. Or at least that is the promise.


At Vellum we don’t believe in this type of generative design. Actually, many of the customers we worked with pushed back against this idea as well. Does that mean generative design is useless? No, that also is quite far from the truth.


Diversity is not always required. Sometimes you don’t need diversity in your design options. In the early stage of project development for example, the main question is: Can this investment work? Many of the more intangible criteria of what makes a great building are not so relevant yet.


Generative design should focus on the boring parts. Many parts in the architecture process can be called repetitive and thus not creative. Designing safety routes, construction parameters, going back and forth between excel sheets and sketchup - these are not the exciting parts. And they happen to follow strict rules as well. Why not automate those? This allows everyone to do what they do best - architects to be creative, and computers to be fast. Don’t use generative design to generate hundreds of options, but rather allow experts to generate their own options - a lot faster.


Generative design provides a benchmark. Although generative design will probably not generate complete buildings in the near future, it does provide a benchmark of what is possible. Many of the people we talk use defaults for example for GFA/NFA ratios, maximum possible buildable areas or even energy requirements. Generative design can show that higher or lower ratios are possible, which is a great way to manage the conversation.


Generative design as an analysis tool. Generative design generates buildings based on a set of outcomes. These outcomes can be anything from financial feasibility to energy usage or shadow and sun analysis. The prospect of having immediate access to these outcomes without having to consult external advisers is very useful.


As with any hype, generative design initially over promised and underdelivered. But just like the fact that we are probably not going to have our own fully automated cars for a while, that doesn’t mean our cars are not getting smarter. Innovation in the end is nothing if it does not solve real problems, and that is the exciting challenge we are currently in.