Skip to content or navigation

Curators of Spatial Computation—What Autodesk Project Fractal Says About Future Architects

In this special feature, Architosh discusses Autodesk’s Project Fractal presentation at AIA Philadelphia, placing it within the context of the growing AEC industry interest and development in computational design software. Algorithmic design tools are far from a passing fad; this article discusses why and how some of the work of the Generative Design Group at Autodesk is pushing the envelope in interesting directions.

Computational design continues to advance in the world of architecture unabated. In the past year the number of algorithmic design softwares options has nearly doubled and many more architects—and not just the very young—are taking notice. It shouldn’t be surprising then that at this year’s AIA National Convention in Philly, several software rivals vied for attention in this particular area, bringing this somewhat intimidating software specialization to the mainstream audience.

So while computational design (sometimes called generative design) isn’t turning into the passing fad some older architects may have been thinking it would—and it definitely is not—what isn’t quite clear is where it all fits in and, even more importantly, where it may be taking the field of architecture.  To help ponder those kinds of questions, Autodesk hosted an invite only event for its Project Fractal work in Philly during AIA. In a small hotel conference room, with teas, cakes, coffee and water to assuage a potentially footsore crowd, key Autodesk customers and invited press got a chance to see early work progress of where Project Fractal is going.

Project Fractal—Of Generators and Evaluators

Project Fractal is another computational design tool that grew out of the new AEC Generative Design group at Autodesk. Anthony Hauck, Autodesk Director of Product Strategy for that group, was our presenter. Like other generative or computational design tools, Project Fractal utilizes algorithms and the power of compute to power through design problems. One way to help explain what we mean by this process of “powering through design problems” is to imagine scenes from that spy film you last saw where a criminal utilized a computer to crack a code to gain access to something. What we know from those scenes is that the computer crunches away at millions of options per second to find the password combination that lets them in.

01 - An image of Autodesk Project Fractal showing the slider control mechanism for brushing along the axes of variables in the particular algorithm at work. In this case, studying facade layout data.

01 – An image of Autodesk Project Fractal showing the slider control mechanism for brushing along the axes of variables in the particular algorithm at work. In this case, studying facade layout data. Note the Mac OS X Chrome Browser. Project Fractal works through the web.

In many ways that is what computational design tools are trying to do, except that computational design isn’t necessarily searching for just one solution. After all, in architecture there isn’t necessarily one perfect or superior solution—in a floor plan arrangement, the design and layout of a facade, or especially in a building’s overall three dimensional form. Yet, there is very much the case in architecture that there are superior options over inferior ones, as measured against numerous numerical or quantitative criteria.

The Design Turing Test

Project Fractal can attack various problems germaine to mainstream architectural design workflows, such as floor plan fitting (planning) to a given site with its various constraints. In one memorable sample, Hauck showed Fractal running through several algorithms to generates programmatic floor plan options. As he noted, some of those options were totally unreasonable and flawed at some vital level no matter how small the error, while other passing solutions may seem nearly right but are missing something intangible to a computer. In other words, a human architect would say, “okay, this passes all the core programmatic requirements beautifully—but, it missed a key opportunity with the design.”

02 - Floor plan generation using algorithms in Project Fractal.

03 – Floor plan generation using algorithms in Project Fractal.

03 - Sliders enable the designer to scrub through the viable numerical range data across the axes of the particular algorithm.

04 – Sliders enable the designer to scrub through the viable numerical range data across the axes of the particular algorithm.

How do we teach computers to recognize “opportunities” that are unimagined at the beginning? That is how real human architects work when they are combining intuition and logic to solve various architectural issues. Hauck said the real ultimate aim of such technology like Project Fractal is to pass a type of design Turing Test—in other words, it looks like a human architect did it.

Generators and Evaluators

Hauck, and many of the audience members in a follow-up Q&A session, noted that computational design isn’t going to replace human architects, even if small and tightly defined architectural problems can be solved by algorithms and pass a Turing Test. Instead, human architects begin to look at tools like Project Fractal as “digital design assistants,” attacking problems that are too time-consuming for even large teams of architects, like testing out 100 floor plan options or modeling 50 different varied facade options against solar criteria measured against daylight criteria.

02 - A similar facade study as image 01 above. (Image: Kyle Martin, Shepley Bulfinch Architects, Boston, All rights reserved.)

05 – A similar facade study as image 01 above. (Image: Kyle Martin, Shepley Bulfinch Architects, Boston, All rights reserved.)

As such, Fractal features two components which the company refers to as “generators” and “evaluators” and, yes, both of these are powered by algorithms. But what was different in Project Fractal was the method by which the user interfaces with each. As Hauck noted, “part of the issue with software that simply seeks to generate many options for designers is that designers are left stuck with the task of evaluating too many options.” To address this, Fractal features evaluator graphs that function like widget controls in the form of simple sliders.

What Fractal does is generate the cross product of the max, min, and median of all the parameters one has set as variables in a Dynamo script, inside Fractal. So 3 parameters = 3^3 = 27 options; 3^5 variables = 245 options; and 3^13 options = 1.59 million options. Where Fractal is today…is this system that displays a very large “design options space” and somewhat reasonably allows the user to filter, or sort, through the options via slider controls that could be described, as Hauck says, “as brushing the axes of the variables in a parallel coordinates display.”

While the power of algorithms, combined with near infinite compute, portend to many possibilities, Hauck noted that in the end “Autodesk may provide the generator tools but the architect will be the editor and curator of the final forms.” So where does this leave us today?

The Rise of Algorithms = The Rise of Questions

As author Warren Berger notes in his book, A More Beautiful Question: The Power of Inquiry to Spark Breakthroughs, we live in an age where questions are rising in value while answers are declining. It sounds totally counter-intuitive but experts note that due to the Internet and Google, “explicit knowledge” has been turned into a commodity. This has challenged the wisdom that an educational system that still revolves around teaching students to memorize facts has value. One such critic of the value of explicit knowledge is Sugata Mitra who, in a TED conference, tossed out the provocative question, “is knowing obsolete?”


Mitra, like Sir Ken Robinson, another TED talk visionary who speaks eloquently about learning and the power of creativity, is impassioned in the belief that human learning needs to change. Moreover, that children can learn from each other by tapping the power of their own intense curiosity and natural questioning. There are many experts now commenting that because mankind can’t possibly compete with the power of computers to hold, find, order, or filter vast amounts of explicit knowledge, it is best that humans develop the capacity to do what computers still cannot—question.

Against this cultural and technological backdrop, technical design fields are advancing themselves through the convergence of data, science, and computational design. Algorithmic and visual programming software tools are likely just the first stage of this new direction, as architects learn to turn a page in their own history and begin to harness the power of near infinite computing resources (in the cloud) and algorithms powering visual scripts that are tailored to attack specific types of design problems.

What this means in real practical terms, and how the architect literally works through her daily routines, is a significant new question. But what we learn from the Project Fractal demonstration is that architects serve the job of human guidance, editing the computational process—a process that involves visual script authorship much like how musicians work with chord arrangements to create a new song. Instead of a song the arrangement of algorithmic-powered scripts is aimed at both defining and attacking a specific type of problem in an architect’s workflow. What Fractal seems to be encouraging, or at least pointing to, is an increased importance in “framing problems” through the power of algorithmic thinking. But this framing process, when aided with the sense of infinite compute, has the power of stirring curiosity and wonder. In other words—questioning.

As the architect harnesses an architectural questioning machine, the pursuit of questions like, “does shortening the path distances clinicians walk in their daily routines enough to improve patient care?” and “if so, what is the ‘link’ between patient care and distances clinicians walk?” and “how can we test this in built environments?”

The architect may become not just an editor and curator of both “generator” and “evaluator” algorithms but also of his own curiosity and query process in the pursuit of optimized design.


cover image: Image: Kyle Martin, Shepley Bulfinch Architects, Boston, All rights reserved.