The Designer’s AI Apprentice: Starck & Autodesk Create “AI Chair”
Can an artistic genius and a whip-smart computer get along?
French designer Philippe Starck—whose work is a mainstay of museum exhibitions—decided to find out. About two years ago, Starck heard that 3D engineering software company Autodesk (NASDAQ: ADSK) was using artificial intelligence software to churn out images of possible products, guided by the design specifications of humans.
His call to the San Rafael, CA-based company led to a lengthy Starck-Autodesk collaboration, ending in the debut this month of the sleek “AI chair’’ at Milan Design Week in Italy. That’s it above, in three colors, fabricated by Kartell, the Italian partner that also manufactured Starck’s transparent plastic “Louis Ghost” chair—one of his most successful creations, with more than a million sold. (See below for a bigger photo of the AI chair.)
But it was the experience of shepherding a world-class designer through a partnership with a collection of algorithms that captivated Autodesk’s Mark Davis, who is now senior director of a new design futures group at the company.
Davis has been working on development of AI elements for the Autodesk software platform that is tailored for product design and manufacturing, which includes the company’s flagship computer-aided design tool, AutoCAD. The AI application that pumps out product model options is called “generative design,’’ and an early version of it was incorporated into Autodesk’s Fusion 360 subscription software collection in November.
Research to expand the generative design technology’s capabilities is continuing, bolstered by Autodesk’s collaborations with partners such as European aircraft manufacturer Airbus, and now, with Starck. “It’s getting more sophisticated,’’ Davis says of the AI software.
It could hardly help evolving some, when Starck came on the scene.
A long and sometimes testy dialogue between designer and machine
Compared with Autodesk’s work on the Airbus project, whose utilitarian goal was to design a stronger, lightweight cabin partition for an aircraft, Starck arrived expecting the computer to incorporate qualities like elegance and beauty into his plan for a functional chair.
Armed with a “napkin sketch of shapes he wanted to consider,’’ Starck assumed that the AI “would plop out what he could imagine,’’ after some further sketching and conversation, Davis recalls.
While Starck’s confidence was a compliment to Autodesk, the genius designer and the machine didn’t speak the same language, Davis says.
“He has a very poetic way that he expresses the beauty of something,’’ Davis says. “He knows it when he sees it. The software doesn’t understand that.’’
What followed for Autodesk was a process of explaining the limits of what the machine could do to help Starck realize his vision, and working toward the middle, Davis says. At the same time, the Autodesk team members were learning from Starck an entirely different way of approaching product design, Davis says. It turns traditional computer-aided design on its head.
The company was used to working with partners in the automotive and aerospace industries, which have exacting engineering standards that leave very little room for interpretation, Davis says. An engineer would already have a product configuration in her head, and the CAD program would draw up a schematic for it, he says.
While working with Starck, rather than thinking about preconceived solutions, the Autodesk team was thinking about problems, Davis says.
The collaboration’s mission was to design a chair with the minimum amount of material needed to safely support the human body. The partners had to adapt the design so that it could be manufactured using the process of mold injection—a specialty of Starck’s longtime manufacturing partner Kartell. The chair also had to be stackable. Not only did the seat need to be comfortable for a human body, but the shape of the bottom of one chair must conform roughly to the top of the next chair in the stack.
Starck focused his attention on the parts of the human back and arms that would need to be supported in the right places, Davis says. The AI software was equipped to do the mathematical grunt work, calculating angles, forces, loads, and the most effective use of material, he says.
Still, with all that combined cognitive power, the computer’s first rendition of the chair design evoked a “less than positive response” from Starck, Davis says. Here’s an image of one of the first versions generated by the AI software. It looks something like a cross between a blocky church pew and a high chair.
To arrive at the eventual AI chair, Starck had to engage in a repeated sort of “dialogue’’ with the computer—rejecting many new designs it offered, and explaining why.
“It took several thousand iterations to get there,’’ Davis says. Even Davis describes some of the computer’s earlier chair ideas as “organic, lumpy, and crude.’’
Rather than articulating his vision by writing a set of precisely defined standards for the machine, Starck trained the AI software by reacting to both good examples and bad ones, to gradually guide it in the desired directions. “More like this, less like that,” is how Davis describes Starck’s communications with the computer.
“He revealed to us that several times he almost gave up” because he thought “the computer wasn’t going to be able to get there,” Davis says.
It was always going to be a high bar to set for a machine, judging from the breadth of Starck’s view of the role of design in the world. His sweeping design theories—summarized on his website—include spiritual values, environmental principles, and societal goals, such as making elegant products affordable to large numbers of people.
But Starck was never just an artistic visionary, removed from the practical challenges of designing products for real-world use, according to a long biographical sketch on his website. The son of an aeronautical engineer and airplane builder, Starck opened his own industrial design studio early in his career, and over the years he has created products spanning furniture, toothbrushes, pre-fab homes, earphones, and external hard drives.
As the Autodesk project continued, the chair design was pared down through the process of “lightweighting,” or removing material not essential to the chair’s structural strength. The effort resulted in a final 13 “families” of design variations submitted to Starck, who picked the one below left to refine further into the eventual AI chair. You’ll be excused if you think it looks like a summer cabin seat crafted from tree branches, then coated with molten chocolate.
But the bone structure of that design, further simplified and smoothed, resulted in the final version of the “AI chair,’’ shown in the photo below right. You can still see the vestiges of that organic tree-branch look in the curving braces that support the chair’s arms and legs.
Codifying a designer’s style
Autodesk’s generative design software can be used for open-ended exploration of all product possibilities by a non-designer who simply types in limiting parameters, such as function, weight, materials, and price point. The computer can generate thousands of design models that the user might never have dreamed up.
But the Starck project raises some interesting possibilities that go in the opposite direction—to train the computer to think like a particular designer. As the Autodesk collaboration matured, Davis says in a company blog post, “the system became a much stronger collaborative partner, and began to anticipate Starck’s preferences and the way he likes to work.”
In a future version of Autodesk’s generative design software, it might more quickly adapt itself to a designer’s aesthetic sense and working methods, rather than requiring the designer to learn about the computer’s limitations, Davis says.
How far could that software adaptation process go in the commercial realm of product design? Could such an AI-assisted system someday review the output of a Renaissance master, capture the essentials of their design genius, and generate specs for a Leonardo da Vinci-style spice rack?
Davis says that’s not beyond the realm of possibility, though it’s likely to be far in the future. But he’s less interested in outcomes like the spice rack abomination than in the potential for product companies to articulate a strong, consistent “design style,” systematized by noting factors such as angles, materials, and ergonomics.
This could be “the mathematical representation of a brand,” Davis says.
Photos courtesy of Autodesk