With Google, it always comes back to cats.

A demonstration on how computers recognize cat pictures kicked off Google Brain’s much-anticipated “Magenta” conversation at Moogfest on Sunday afternoon. Along with IBM’s Watson, the talk was a must-see headliner of the technology and music festival’s major theme of Art & Artificial Intelligence.

Magenta is a side project of Google’s deep learning research division Google Brain, an attempt to give artists and creatives the power of machine learning and data that Google is applying to other fields. Machine learning (which is the more accurate term for what the public generally calls “artificial intelligence”) gives software the ability to automatically change how it operates based off data. Or put another way, to learn.

It also helps to find cat pics. Douglas Eck and Dr. Adam Roberts (both former computer scientist academics turned Googlers) opened the event by explaining how traditional “rule-based” programming doesn’t work with machine learning, because there would never be a way to write a rule for every possible cat image. The trick, Eck explained, is to teach computers to teach themselves to find cats.

I think the phrase is: Give a computer a rule to find a cat pic, it will find you one cat pic. But teach a computer how to teach itself how to find cat pics, and it will find you cat pics for a lifetime. Or something like that.

Whether Google’s affinity for explaining tremendously complex concepts with cats is a clever disguise for the terrifying power of artificial intelligence or it simply understands its vital role to deliver cat photos across the globe in milliseconds, Magenta provides a window into the unique place Google operates. It’s a utility-sized entity but also a company that sometimes just likes to be “one of the gals or guys” at a tech festival.

Eck touched on Google’s size almost immediately, the $540 billion elephant in the room he referred to as the “Hand of Google”, making it clear that he’s not simply pushing Google’s products and that most of the panel is of the non-Google variety. Adam Florin is creator of freeform music generating software Patter; Dr. Tobias Overath is a PhD at Duke University researching how the brain processes sound; and Gil Weinberg (along with his collaborative robot Shimon) is director of the Georgia Tech Center for Music Technology. But Eck also admitted tacitly that Google’s enormous size is what allows projects like Magenta to happen in the first place: ideas without real goals in mind.

“Is it worth it?” was a major question Eck posed to the crowd early on, and he wasn’t kidding. He admitted that even if Magenta was successful in its goal, it still might not have any use. That’s a sobering but liberating feeling, he said before introducing the panelists.

Nobody is sure how Magenta and machine learning might benefit the creative fields, but that’s why we’re all gathered in the Durham Arts Council building in the first place: to talk about artificial art, and the fact that Google has paid some of the best minds in the field to invent it, whether there will be a use for it or not.

Eck opened by arguing that there’s always one way to know if computers can make good art: if people like it. If no one likes Magenta’s art, then it failed.

Things got technical fast. Eck asked everyone in the crowd to raise their hand if they could program a computer from scratch, and maybe 75% raised their hands. Matrices and input-output graphs slid across the screen as Eck outlined how TensorFlow (a feature development of Google Brain) connects graphs of data across mathematical operations, how Magenta uses equations to guess which notes would (or should) come next in a song, and how it’s actually easier to teach computers to solve lots of problems than it is to teach computers to solve just one problem.

Eck wanted “real feedback” from the crowd and he got it from an extremely engaged and packed house that paid plenty for tickets to attend. Crowd members were asked to guess language models used in certain demonstrations and the rules that an algorithm might employ to find, of course, cats. There was a sense at times that the speakers had to prove themselves to their audience, as in an academic setting of their peers. (A world where most panelists come from.) It was a fascinating mix of academia and hipster energy that simply can’t happen in a lot of places.

Inevitably, the conversation evolved into whether computers making art was something even worth doing. The dialogue turned quickly from a simple Luddite Humans vs. Robots discussion (Eck joked that “paint in a tube was once the end of painting”) to whether robots should be making art at all and the implications that has for art made by humans. Do we need artificial art? Can art be correct or wrong?

If Moogfest can be commended for one thing, it’s certainly conducive to such a high-level conversation. It’s an impressive sight to see some of the most complex problems in the world broken down in such a laid-back, welcoming and fun atmosphere. If that’s the allure and goal of the young festival (hosted in New York City, Asheville, and now Durham) named for the Moog synthesizer and the founder of modern electronic music, then it was certainly a success, and downtown Durham couldn’t be a better place for it.

How the Magenta talk fit into the wider festival as a whole is hard to nail down—Moogfest is such a fragmented collection of oddities that it’s hard to even consider it in such terms. Maybe that’s the point. Maybe Moogfest is a lot like Magenta, still evolving and figuring out exactly what it’s here for.