The “How to Generate (Almost) Anything” project yields everything from dresses to graffiti
Have you ever considered what AI-inspired pizza might taste like? Or how perfume with AI-generated recipes might smell?
Pinar Yanardag, a postdoctoral associate at MIT’s Media Lab explores these questions and more through her “How to Generate (Almost) Anything” series, which she describes as an example of MIT hacking culture and designed to “playfully show how we can use technology to create cool stuff.” In this project, humans and AI collaborate on everything from designing dresses to cooking up recipes for pizza.
First, because AI has to learn from a database of similar objects, Yanardag and her team find datasets for what they want to create. This sometimes involves reading blogs to find what kind of resources artists working on that area use. After obtaining that data, they decide what kind of algorithm would fit well to generate something similar, train the algorithm for several days or weeks, and finally look at the results to see what the AI has dreamt up.
Yanardag and her team then contact potential collaborators. “Most of the time, people are super interested in doing a collaboration,” notes Yanardag.
After selecting a couple of promising recipes, they work to then bring the recipes to life.
Rather than strictly adhering to what AI generates, however, artists use it as inspiration. And while the ideas generated by AI may seem out of the ordinary, it’s that characteristic that can lead to innovation, allowing people to, as Yanardag believes, “create things that they wouldn't create otherwise.”
“When we collaborated with a dress designer, she really liked [the AI’s creations] because she found it really odd and really inspiring. She said she wouldn’t even think of [it] if it wasn’t for the AI.”
Something that surprised her, however, was how receptive people were to the AI’s unusual creations (which include a shrimp, jam and Italian sausage pizza and a futuristic-looking hat). People were very willing to try out the creations, even if they initially seemed strange, and some ended up really enjoying the results of the human-AI collaboration.
Yanardag has also found the process to be rewarding in more ways than one. Before this project, she had never tried making graffiti before, but now, she says, “I’m really interested in learning more about graffiti art and actually trying to create my own graffiti.”
But will AIs one day be able to do the creative work on their own?
“I think AI is going to be really good in creating things that probably don’t need human interpretation or feedback much.” Although AI systems are getting better and better, Yanardag notes, “I think people’s tastes and people’s opinions will change over time too. So maybe, [it’s the] feedback loop between AI and humans that will create the best outcome.”
“[Ultimately], I think AI and humans are dependent on each other. I don’t see a future where AI can completely take over.” In the future, however, just like people now collaborate with each other, perhaps people will also collaborate directly with AI.
Visit howtogeneratealmostanything.com for more information on Yanardag’s project.