Pindar Van Arman Is Teaching Robot Artists To Paint Like Humans (HBO)

Pindar Van Arman Is Teaching Robot Artists To Paint Like Humans (HBO)


— The Singularity is the theoretical day that artificial intelligence
advances beyond human understanding and begins to change civilization
in ways we can’t conceive of. Many fear the Singularity. — It needs… wanna try the gray? — Not Pindar Van Arman. He’s doing his small part to make it
happen by teaching robots to make art. He’s writing code that nobody’s seen before, and thinks his machines are
on the precipice of creative autonomy. — Here is a De Kooning and this one, the style’s just as interesting. — Wow. That looks cool. — And here is “Guernica.” A little darker, but… — Yeah, that looks like human suffering on my face. — Pindar is a computer security contractor
in northern Virginia. But for the last 12 years, he’s been making robots artists on the side. These portraits are made by
sending the computer two images, and running an algorithm called style transfer. It takes content and style from one
and applies it to the other. — And here is you as Marilyn Monroe. — So, it doesn’t quite capture what
Andy Warhol is doing though, right? Because instead of just making my hair blond, it gave me, like, Marilyn hair. So how would it have to progress to
make it more in the style of Warhol? — Yeah, it doesn’t really capture context right now. So style transfer does amazing things. And a lot of data scientists are trying
to make this even better, including myself. We’re trying to find a way to add the context to it. We just don’t know how yet. — The hardware is pretty basic; the robots can be built for just a few hundred dollars. — There’s the A.I. work-in. — The special sauce is his software program,
a decade in the making, called Cloud Painter. Like a human artist, Cloud Painter remembers its past work
and can try to improve it, and its style evolves over time. And the robots can see what they are doing, because they use cameras to watch
their work in progress and make adjustments. — I want them to always watch
what they’re painting, because that’s when they make
the really good decisions. — Do the robots ever create something
that falls in the uncanny valley, where it’s kind of creepy? — I’ve been at exhibitions where
a fellow artist has come and is like, “I don’t know whether to be really impressed
with this or really disgusted with it.” And so, it’s in the uncanny valley in that
I am trying to replicate human creativity. And it does make some people uneasy. — Pindar can tell the robot to paint
whoever walks in front of the robot’s camera. Or he can program it to paint something
based on what’s trending in Google News. But that still essentially means Pindar’s the artist, and his robot is his medium. — Can a robot be creative? — Yeah, it definitely can be creative, but it’s like a question of, like… what level of creativity. I think my robots, when they make portraits, they make all of the same creative decisions I make. — It’s actually making decisions about what to do? — Yeah. I’ve some paintings that
I don’t make a single decision— all I do is turn the robot on and say, “Paint the next person walks in front of the camera.” So there is a way that a robot can use an
evolutionary approach to develop an algorithm. It can randomly select all these algorithms, paint a bunch of brushstrokes randomly,
on 10 different canvases, throw ‘em up for sale on eBay. Whichever one sells the highest,
it tries to paint a little more like that. — Do you think your robot has started
making more beautiful paintings over time? — If beauty is measured by how interesting it is, definitely much more beautiful. But aesthetically, the aesthetics are
going in an unsure direction, because if I just wanted to make a beautiful painting, I could use a printer and print it out and be perfect. And so, I’m going away from that. I’m trying to get more interpretation,
more serendipity. — Ken Goldberg is an artist
and professor of engineering who runs a robotics lab at UC Berkeley. We showed him how Pindar’s robots work: — Robots will do unexpected things all this time. So the fact that it does something a little
bit surprising doesn’t mean that it’s creative. It’s saying it’s the nature of most machines—
that they’ll do things you don’t expect. As soon as you inject any kind of
randomness into a program, you get behavior that you may not predict. And this is essential to a lot of the deep learning or A.I. methods that are out there now. Being able to show where this uncertainty
is arising is very interesting conceptually, but I want to make a distinction between that and saying, “Well, the robot is now being creative.” — I think when people see a great work of art, one reason they relate to it is because you think, “Oh, this great artist had the same emotion as me.” So how could a robot create that? — Yeah, I dunno. Obviously, a robot won’t be able to make
emotional art work until it is emotional. But that doesn’t mean that when
we look at a piece of artwork, we can’t get emotions from it. So every so often my robot paint
a picture that looks like a sad person and that might make me feel sad. The robot was obviously not sad when it made it, but it didn’t stop me from
having that emotional reaction.

47 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *