fuel us to be more creative.” Being ex-
posed to extremely different influences
can be inspiring, and is often essential
to producing breakthroughs, he adds.
“What better way, save for visiting alien
cultures, is there to experience truly dif-
ferent ways of thinking and artistic ex-
pression than letting AI produce art? I
predict it will unlock vast new cultural
and scientific landscapes for humans to
explore.”
The art produced by networks has
greatly informed our scientific under-
standing of how deep neural networks
see the world, says Clune, and they do
so very differently than humans do.
He references the Turing Test, and
how computers must not only produce
art, but “produce art at a level of quality and originality such that it is indistinguishable from art produced by a human” to pass the test. Clune decided to
test that theory by submitting art the lab’s
deep neural networks (DNNs) produced
to the University of Wyoming’s 40th
Annual Juried Student Exhibition, which
accepted 35.5% of the submissions.
Clune and his team developed an
AI “artist” agent and a “critic” or judge
agent. The purpose was for the artist
to try and produce art the critic/judge
would like, he explains. To do that, the
team created an innovation engine
algorithm with the artist and judge
agents inside. The critic was a DNN
trained to recognize 1,000 categories
of images, Clune says, “and it’s the AI
artist’s job to produce images that the
critic agrees count as each type of im-
age,” like a motorcycle.
“We have found that it’s important
for artificial intelligence, while it’s
learning, to have to constantly switch
goals and have many goals,” he explains. That is why they challenged the
AI artist to make 1,000 different types
of images, instead of just one. The team
went through the images the artist
agent produced and submitted a handful for the competition they thought appeared to show interesting artistic interpretations of concepts, such as a beacon
or a prison cell.
There were no rules forbidding such
entries, he notes, “largely because I’m
sure the competition organizers didn’t
realize it was possible for AI to produce
high-quality art. In fact, if the rules for
art competitions begin to be rewritten
to limit submissions to humans, we will
know that AI has become such a great
artist that many humans are afraid to
“We have found that
it’s important for
artificial intelligence,
while it’s learning, to
have to constantly
switch goals and
have many goals.”
compete against it.”
The outcome? Not only were the
images accepted, but they were also
among the 21.3% of submissions to receive an award. The work was then displayed at the university’s art museum.
Clune adds that the judges were evaluating all submitted art on its own merits
to see which was good enough to be accepted. “I imagine the judges never considered the fact that the art might have
been made by AI.”
Music to the Ears
Perry Cook, professor emeritus of com-
puter science at Princeton University, be-
lieves art has been combined with tech-
nology longer than many people realize.
“The march of music has been about
technology. Music has, in some cases,
brought about new technologies, and
certain music has taken advantage of the
technology of the day,” says Cook, who is
also a research coordinator and IP Strat-
egist for SMule, a social network based
on creating and sharing music, and co-
founder and executive vice president of
Kadenze, an online arts and technology
education startup. In a blog post (http://
bit.ly/2feVCyd), Cook cited numerous
artists who have created artwork using
machine learning, which he says opens
up all kinds of new possibilities.
Although a common perception is
that computer music came along late
in the evolution of electronic music, ac-
cording to Cook, people were looking at
using computers to store and analyze
music in the 1950s. “Forward to today,
the computer is just another technol-
ogy, in my opinion. It’s somewhat com-
parable to putting a valve on a bugle so
you could play more notes, [which] al-
lowed music to get louder and softer.”
The arts are an especially good way
to introduce people who normally
would not have thought they should
or would want to learn concepts about
computer science, he notes, specifically artists, children, and underrepresented minority groups “who think
they’re kind of locked out or aren’t
interested and don’t want to be engineers or computer scientists. Getting
to them through the arts is a way to
open up [computer science] to a whole
new community.”
Making CS “Culturally Responsive”
That is exactly what Gaskins is doing at
Images produced with innovation engines were not only accepted to a selective art
competition and displayed at the University of Wyoming Art Museum, but they also were
among the 21% of submissions that won an award.