The Center for Data Innovation spoke with Pinar Yanardag and Emily Salvador, co-founders of GLITCH, a clothing line based in Cambridge, Massachusetts that uses AI to generate designs. Yanardag and Salvador discussed the advantages of AI-powered design, and how human-AI collaboration can enhance creativity and achieve productive outcomes.
Eline Chivot: What inspired you to launch GLITCH, and what is the purpose behind your business?
GLITCH Team: The field of AI (and computer science in general) is highly male-dominated, and it can be difficult to find emerging technology products specifically designed for women by women. At MIT, we started exploring how AI tools can create products for women, and furthermore, how these tools can empower designers and novices alike. So far, we’ve created AI-generated perfume, jewelry, shoes, chocolate truffles, and fashion designs!
We started our fashion journey with “Little Black Dress—AI” (LBD-AI) as we were initially inspired by the 1920s “Little Black Dress” concept. The little black dress is considered an essential item for any woman’s wardrobe. We wondered how AI could reimagine it. The result was a combination of unexpected and enchanting—the design featured a normal sleeve accompanied by a bell sleeve, asymmetric details and more!
We believe that there is a unique connection between the LBD concept and AI. LBD is considered as an essential item for any woman’s wardrobe. And AI is now considered as an essential tool for any computer scientist. Given that AI is a very male-dominated field, we believe it is essential for female computer scientists to learn AI. Connecting these two concepts, we are hoping to raise awareness and increase participation of women in AI by promoting LBD-AI and GLITCH.
We shared our LBD-AI work with women in the field of science, technology, engineering, and mathematics (STEM), and the feedback was largely positive, with many women suggesting they’d wear an AI-generated dress as a conversation piece. Fashion is an extension of one’s identity, and these women were particularly excited to wear clothing that reflects their passion for technology! After seeing the feedback from many women in STEM, we decided to start a fashion brand with a number of AI-generated designs we have been working on. Currently, we have three collections with nine dresses in total, and we are donating 50 percent of our profits to the Anita Borg Organization, a leading women in STEM non-profit.
Chivot: How does your algorithm produce your iconic little black dress? What type of AI do you use, and what kind of data do you need to collect to train and teach the system?
GLITCH Team: We train a type of AI (a generative adversarial network or GAN) on tens of thousands of fashion designs. The AI is able to go through the images of these fashion items (dresses, blazers, and shoes) and learn important features and connections between them. Then, it is able to imagine (or dream-up) new designs that don’t exist in the collection we provided to it. That’s the core algorithm behind GLITCH, where we use AI to generate new designs, and we curate the most interesting ones! Then human designers interpret AI-generated designs by adding their own perspective and creativity, and the final product goes to GLITCH.
We are allowing our customers to generate new designs, and vote for the best ones. We are currently working on curating our next collection with this crowdsourced data.
Chivot: How do you evaluate and validate the algorithm’s original creations to know that they are “a good fit” in general, and for a customer?
GLITCH Team: We use GANs to generate fashion designs, and we are using machine learning to curate the most diverse and original designs out of a huge set of candidates. Then, human designers (currently, us!) go over the top candidates, and curate collections. So far, we have three collections at GLITCH, and we are currently working on curating a crowdsourced collection with the help of our customers. So far we have over 10,000 votes on our dress-generation tool.
Chivot: How can AI technologies like yours empower designers? Which stages of the process still require human intervention and, conversely, what can the AI produce or create that a human designer wouldn’t be able to?
GLITCH Team: In the midst of this “AI awakening” where machines are becoming increasingly good at many “human” jobs, people are worried that AI will ultimately lead to mass unemployment. We believe that on the contrary, humans and machines can work together to enhance each other’s complementary strengths and skills to achieve the most creative and productive innovations. Especially, democratized AI-systems have the potential to unleash hidden creativity and empower people from all backgrounds and skill levels. In particular, AI makes design accessible for the public by eliminating the need for technical skill and practice. Imagine tools that allow you to create unique fashion designs in just a few seconds!
We believe that humans and AI systems have complementary skills. AI systems are optimized to analyze vast amounts of data, extracting features, and making unexpected connections between them. Humans provide unique insights through emotions, empathy, creativity, and critical thinking. We think of our AI system as another member of the team, and we believe that we can achieve amazing results through human-AI collaboration. We experienced this phenomenon first hand: The AI came up with a unique dress design such as a dress with one bell sleeve and one normal sleeve, pointy shoulders, asymmetric neck, and skirt details, which inspired us to create LBD-AI.
Chivot: In the future, how do you think AI will affect retail and fashion—for instance, the way clothes are designed, the way people shop, and the roles of designers and customers?
GLITCH Team: An unprecedented evolution is happening with the recent advancements in AI, and it enables unpublished creative horizons in fashion and beauty industries. We believe that AI will become a powerful assistant in the designing and manufacturing processes through human-AI collaboration.
In addition to the designing process, AI is also going to be a key player in manufacturing processes and supporting sustainability. Imagine an AI system that creates dress patterns to optimize for minimum fabric waste, or an AI system that streamlines the entire pattern making process in a matter of seconds.
We are currently working on a new AI tool with the aim to help people upcycle their existing fashion items. Imagine telling this AI tool that you have a pair of jeans and a pink t-shirt to upcycle, and it generates unique fashion designs that you can make on your own by using these items!
We are already experiencing how AI is revolutionizing the way clothes are designed: AI allows consumers to be a part of the designing process. As mentioned, we are allowing our customers to generate new designs, and vote for the best ones. Our larger goal is to research cross-cultural preferences on fashion and discover country-based trends.
We are also working on an interactive tool which would allow people to utilize AI and style transfer to design truly unique items of clothing.
