Autodesk University 2018 | Is generative design GOOD for space planning?

watch the video here
Autodesk holds a conference every year in Las Vegas that brings together 10,000 professionals in the construction, manufacturing, architecture, engineering and media industries to learn about innovations in the industry and to share experiences.

This was my first year at the Autodesk University conference. One of my favorite sessions was the Tuesday morning talk ” Is generative design GOOD for space planning?… emphasis on the GOOD. I was excited for this talk because I do a good amount of space planning and am eager to take advantage of the benefits that generative design provides.

The session was an Oxford-style debate that posed the question of generative design in space planning. Two teams were put head to head in a For or Against competition.

Those fighting against the motion:

  • Lead Digital Designer at Proving Ground, Kirsten Schulte
  • Architect and Founding Director of the Smart Cities Institute, Mark Burry

Those fighting for the motion:

  • Computational Designer, Alyssa Haas
  • Computer chip engineer and parametric modeling specialist, Deepal Aatresh

Although I am definitely for the motion, there were a few powerful arguments from the against team.

The Against Arguments

  1. Use the current technology to solve the problems you have right now. The technology isn’t GOOD enough to create designed space plans that we can sell to clients out of the box; so, we should focus our energy on using generative design for what its good at today such as business intelligence, area calculations, data visualization, etc…
  2. Generative design may be useful to design but it is not GOOD for it; rather, it is dangerous for design because it could end up replacing architects. As an architect, we think beyond the plan and are required to bring so much more to the picture. Architects greatest sensibilities are around placemaking and the computer cannot, and should not, be doing this task. “Places don’t have boundaries but everything we do computationally has to have boundaries.”
  3. The curse of standardization is threatening to architects in some ways. If generative design tools are created for developers and clients then the architect may lose business. Suburbian housing is a good example of this.

The For Arguments

  1. Do what we can with the tangible tools that we have now. As long as architects are in control of the final output then it is ok to let the computer take on some of the work and help. There are a lot of rigid rules that can be computationally considered while space planning ex. code compliance, the program, adjacencies, light quality, building typology, and site. If we allow the computer to do some of the work then we can spend more time focusing on design.
  2. As architects, we can’t ignore these technologies and need to figure out where we belong in it so that it doesn’t fall into the wrong hands. We need to make sure designers with good intentions are the ones who have the control. Maybe architects could take suburbia back by efficiently using computational design. Additionally, a lot of building types cannot afford an architect so generative design could be a way of democratizing good design.
  3. “This is not a new problem, it has been solved in other industries…Why is this even a question?” Construction is at the bottom of the productivity chart and it isn’t getting any better. Considering this, we have no choice but to create new and use existing tools for generative design in space planning.

Other good points along the way

  1. The important keys when discussing technology adoption is people and processes. If your team doesn’t adopt and use the tools properly then you get no benefit.
  2. “There are a lot of bad repeatable buildings that are built right now that computers had absolutely nothing to do with, that people spent quite a lot of time designing.”
  3. Hospital architects used to be considered more as systems architects because the hospital is very rigid. Typologies like this can be good for generative space plans.
  4. The tool maker should be the tool user. We need a new generation of architects who work with code in order to make the generative design tools.
  5. “Algorithms can supercharge any craft”
  6. “The tools don’t make all the decisions, the humans still do. The tools get you from point A to point B really fast.”
  7. For success with machine learning, you need good data to feed it.
  8. Products are not bad. There are good products and bad products. Customizable good products are what’s missing in architecture
  9. The business answer to adopting automative processes haven’t been figured out in this industry.

Conclusion: Yes, generative design is GOOD for space planning but soon it will be GREAT…

The generative design tools we already have can do a lot to supplement a good designer but there is still a way to go before they can design space plans on their own.

One of the limitations of having a computer generate designed space plans is that we are not providing it with the right type of parameters and information. For now, we can give the computer numbers and maths and it can calculate sqft and provide basic area maps, along with some other cool things; but, it isn’t designing a sellable space plan…yet. This is where machine learning, big data and AI step in.

With developments in AI, neural networks and big data we are closer than ever to being able to teach the computer how to generate options more similar to how we humans do. Today the computer generates random iterations of a variable set but we need to teach the computer how design. To do this we need to find the core values we hold in design and practice those values with an AI.

For example, in space planning adjacencies are very important. We are given some of these adjacencies by the program or the client but that only takes us so far. We can feed these adjacencies to the computer which will limit some of the options but there are still a ton more that the computer will generate and that we have to sift through. Many rules of space planning are learned with experience and human intellect. As designers, we know that the restrooms shouldn’t be placed along the glass curtain wall but the computer doesn’t. As humans, we know that the restroom is a private place so placing it in such a visible and prime location is not acceptable. We know this, not necessarily, because we were told; rather, we feel this way because it is what we have seen over and over again. By this same principle, we can teach the computer that restrooms shouldn’t be placed on the curtain wall by feeding it a lot of good space plans. And when I say a lot, I mean a big data amount i.e. 10,000, not 100. It will recognize and repeat the patterns in a similar way to how we do. The more we do it, the better it will get. It will then start to recognize patterns that we couldn’t see and eventually become a very powerful tool that knows the sensitivities of creating a good space plan.

The beauty of this in a creative industry is that each AI will be different the same way that humans are different. The AI will be a reflection of the person or dataset that is influencing it. Creativity is relative and we will still see this in an artificially augmented world.

The technologies on the rise will broaden our horizons by an order of magnitude. Although I agree that they are still a while away on a grand scale, in all four of the speakers we are seeing the grassroots projects that will pioneer the industry forward.

watch the video here