Amplifying human intelligence with artificial intelligence has the potential to help solve our thorniest problems, from disease to climate, but can AI create original art? That was the central question Jennifer Sukis, leader of the IBM Watson creative team, posed at Moogfest in Durham, Friday.

“The core of human evolution lies in creating something out of nothing,” Sukis said as she asked what she called the “core question.” Why do we create art?

Up until about 40,000 years ago, humans didn’t have time for art, she said. But then, as the earliest known humans, the Cro Magnons settled down, one of them watched the light play on the walls of their cave, picked up a burnt stick and began to draw on the walls. He could not have done that without the skills picked up from hunting, such as visual perception, using tools, language and understanding.

“Those are the same abilities we want to put in machines,” Sukis said, so that a computer system can perform tasks that usually require human intelligence. Speech recognition and understanding, visual perception, learning.

The potential of AI is enormous. Sukis quoted Kevin Kelly, the founder of Wired magazine, who said, “AI will flow across the grid like electricity. As it becomes more deeply integrated in our lives, it will become the new infrastructure powering a second industrial revolution.”

Right now, however, “We’re only in the earliest stages of seeing something exponentially more powerful. In the AI world today, we’re working to establish the foundations and ethical boundries of a system that may one day enable the next step in human evolution,” Sukis said.

Jennifer Sukis sat down for an onstage conversation following her lecture on AI and art at Moogfest. Copyright Capitol Cities. All rights reserved.

She then ran through a brief history of the development of AI:

  • The digital computers in 1946 that could do calculations in 30 seconds that took humans 20 hours
  •  The Turing test to determine if computers can think in 1950.
  • The birth of AI at the Dartmouth conference in 1956.
  • Machine learning and a checkers playing program in 1959.
  • Advances in technology in speech recognition, graphics, and faster processing from the 1960s-1980s.
  • IBM’s Big Blue defeats the world chess champion in 1997
  • IBM’s Watson defeats a human contestant on the TV program Jeopardy in 2011
  • Alpha Go, which defeates the world’s best human player at Go, the world’s most complicated board game, in 2017.

The Alpha Go neural network system trained 30 million times playing against itself to learn Go. In its match against the number one human player, Alpha Go played a move a human player would not, surprising its opponent. “It didn’t learn it from human moves, it made the decision on its own,” said Sukis.

AI collaborates on art and music

Sukis surveyed the use of AI in creating music, painting, photographs and even fake videos. Currently, she noted, AI is more of a collaborator than a creator in its own right. It augments rather than replaces human efforts. “We’re changing the relationship between ourselves and machines, making them creation partners,” she said.

Researchers and artists are teaching AI systems how to replicated and blend art styles in painting, for instance. “They apply the style of one image to that of another image. They can choose one artist or several and combine and blend them with varying degrees of artistic success. Sometimes, it starts to feel like a new style.” Google’s Deep dream, for instance, creates some psychedelic effects.

Jennifer Sukis, leads the IBM Watson creative team.

In Music, many experiments are underway. Sony’s DeepBach program creates music in the style of Bach’s patterns, and the result is fairly convincing. Anyone who wants to experiment can go to Amper online, create AI mixed tracks and download the copyright free results. In “Algoraves” happening in many cities, a DJ manipulates algorithms on a laptop to control the music, with his work projected on a large screen so users see the live coding as it happens.

Some style transfer tools allow artists to go even further, Google’s Magenta, for instance. Neural networks can even learn to understand the characteristics of sound and generate unique sound patterns.

A sinister aspect: fake news

Some aspects of what AI can do have a sinister aspect in a world of fake news. AI can be used to manipulate high resolution photo images, changing them from one situation to another (day to night, summer to winter) or to create a video of former President Obama giving a talk he never did in real life.

But in terms of AI as a creator, “We’re stepping into the unknown,” Sukis said. Now, it’s still rooted in human artwork. “We have yet to see an algorithm create something entirely on its own.”

Sukis answered questions following her talk on AI and art at Moogfest. Copyright Capitol Cities. All rights reserved.

Going forward, though, “Our whole notion of what we define as art is probably going to change.” That leads to host of questions. If it’s created by math with no meaning behind it, is it still art? Can it be copyrighted? Who owns it and makes money from it? Can it help us understand what we want to communicate more deeply? What does a relationship with a machine look like? Could our collaboration with AI create new kinds of art we never before imagined?

Eventually, she said, “Giving machines memory, language, understanding, reasoning, and learning may lead to AI making art of its own. I hope we’re mindful enough to remember that was once us with a burnt stick drawing on cave walls.”

More coverage: