Generatively Designing for Humans, It’s not that simple

In my time away from architecture I experiment with sound design and music production. One way I’ve found to easily create a large set of audio is to assign certain parameters of a synthesizer (an electronic audio generator e.g. a knobby piano that makes digital sounds) to random LFO’s (low-frequency oscillators) and then loop it to continuously generate audio signals. This creates a sound that I wouldn’t have physically been able to make by adjusting the parameters manually. I can then select the best parts, chop them out and use them in my music. This essentially is an oversimplified form of ‘Generative design’ for music, using a computer to build a solution based on a certain set of parameters.

Image 1: What a simplified sound design approach to generative design looks like

 ‘Generative Design’ is a process where a computer algorithm generates a range of solutions based off a set of project goals defined by the user1. In architecture similar computational design methods have already been used for years to help inform architectural decisions towards sustainable and performance based design2. But more recently companies such as Autodesk and Sidewalk labs have been presenting ‘profound’ projects that can generatively design whole floorplans or urbans blocks based on quantitative data (e.g. façade sunlight exposure, m^2 green space on site) mixed in with data ‘somewhat quantified’ from human emotional and social observation (e.g. how much people like sunlight exposure).3,4

Autodesk, a main advocate pushing generative design tested its first layout based generative design project to design the MaRs office in Toronto. The project had a focus on optimising its office layout to best suit it’s employee’s.



Image 2: Process of the generative design Process used by ‘The Living’ to create the Autodesk office

The system used a data set of user preferences from a survey of 250 employees based on the following:

  1. Adjacency preference (Minimising the distance between collaborative teams and office amenities)
  2. Work Style (Best location for each team and check their noise and light preferences)
  3. Interconnectivity (Maximise activation of shared spaces)
  4. Productivity (Minimise visual and sound distractions)
  5. Natural Light
  6. Exterior Views


Image 3: Some designs generated by ‘The Living’s’ generative design algorithm

Using these goals the algorithm produced over 10,000 design options, filtered them out and presented the ‘best’ options to designers, who then can make their decision based on ‘subjective’ human qualities such as aesthetics. Lead designer from ‘The Living’ describes this process as ‘critical role’ to solve problems humans find complex and difficult to think through.[5] Reading this seems mildly ironical, a ‘critical’ problem solver designed for a data based design to evaluate complex problems from human variables, that are naturally non-quantitative. Why can’t the subjective nature of aesthetics be quantified but the subconscious psychology of why and how far someone likes to sit from a window can. Which human qualities should be used by the algorithm to decide the designs and how many should be used? Studies for University find that people experience up to 27 different emotions6, which are all experienced uniquely for every intricate human being. The design process can help generate a numerous amount of plans both good and bad but trying to let a computer pick out the best solution based off data that has been quantified from a range of visceral human emotions can oversimplify and can create the illusion that a computer has created a list of ‘good designs’7.This leads down another road of debate, is it be really a tool promote creative viable solutions or is it a marketing tool made to attract stakeholders to a method of design where they can see human qualities measured as percentages? The viability of similar methods extends to urban design solutions, such as Sidewalks Labs and emphasis on defined quantifiable quality-of-life measures to design neighbourhoods4.


Image 4: Sidewalk Labs Generative Design Interface

Other problems come to mind from such methods. Will putting an emphasis on a set of measurable goals cloud the judgment of designers or stakeholders and give them a justification to be distracted from areas of design that can’t be interpreted into numbers? If a system is creating 1000’s of design options how easy is it for people with invested interests to influence bias into the algorithm or even plant solutions into the system? I’m definitely posing more questions than answers but it’s a complex problem that is very early in it’s testing and real life application.

Ultimately generative design does have a place in architecture and can be used as an efficient tool to expand the design possibilities and conceptual boundaries of the designer, if it’s used with the right goals in mind. Remarkable things are being done especially in the manufacturing world, from 3D printed building elements to optimised plane components to reduce weight, both offering sustainable and cost efficient benefits8,9 . But a better future for architecture leaves computers to optimise solutions for hard data goals rather than attempting to create ‘critical’ design solutions based on attempted quantifications of human qualities.




[1] Caetano, Ines. Santos Luis. Leitao, Antonio. Computational Design in architecture. Defining Parametric, generative and algorithmic design. Frontiers of Architectural Research. (23 January 2020)

[2] Davis, Daniel. Threee top Firms that are Pursuing Design Research. Architect Magazine. February 2018.

[3] Souza, Edardo. How generative Design will impact Architecture. Archdaily (23 April 2020)

[4] Whitney, Violet. Ho, Brain. A First Step towards the future of neighbourhood design. Sidewalk Labs (11 December 2019)

[5] Howe, Mark. The Promise of Generative Design. World Architects (5 April 2017).

[6] Cowen, Alan. Keltner, Dacher. Self Report Captures 27 Distinct Categories of Emotion bridged by Continuous Gradients. Proceedings of the National Academy of Sciences, United States Vol. 114. (19 September 2019)

[7] Davis, Daniel. Generative design is doomed to fail. Daniel Davis Blog. (20 February 2020)

[8] Austin-Morgan, Tom. Airbus has moved from generatively designing components to the factories in which they are built. Eureka! Magazine (13 January 2020)

[9] Turney, Drew. Concrete Forms Get Stronger, Lighter and more Sustainable with generative Design. Redshift by Autodesk (2 April 2020)



Image 1: Personal Screen shot from inside Ableton, the DAW I used to create electronic music

Image 2: Souza, Edardo. How generative Design will impact Architecture. Archdaily (23 April 2020)

Image 3:

Whitney, Violet. Ho, Brain. A First Step towards the future of neighbourhood design. Sidewalk Labs (11 December 2019)

Image 4:

Howe, Mark. The Promise of Generative Design. World Architects (5 April 2017).


Frearson, Amy. Joel Simon’s Evolving Floor Plans Project Optimises Building layouts. Deezen (3 January 2019)