Each artist once something started. Today we can apply this popular expression against the machines. You need to create creative artificial intelligence? Sometimes it seems that the difference between machines from man to machines can never catch up. But the AI already shows a growing tendency to be creative, whether composing a rock album in the genre heavy metal or the creation of an original portrait, strikingly reminiscent of the brush of Rembrandt.
The application of AI in the world of art may seem excessive: there will always be people who create beautiful work. Proponents of this approach, however, it is said that the real beauty of learning AI creative skills is not the final product, but rather in the potential of technology to expand its own learning machine, learn to solve problems outside the box, faster and better than people. For example, creative AI might one day make the decision to save the lives of passengers driverless car in case of failure of its sensors or offer an unconventional mix of chemical components, which will lead to the creation of the drug, able to cure previously incurable diseases.
AI with a creative flair will be essential for the development of highly automated systems that can respond adequately to human life, says mark Riedl, Professor, school of interactive computing Georgia Institute of Technology. “The fact that we daily do something creative, many problems can be solved creative,” he says. “If my son has a toy stuck under a chair, I’ll have to take the tool from the hanger and get it.”
Riedl notes that human creativity is also important for social interactions, even, for example, to tell a joke or recognize the pun. Computers cannot cope with such subtleties. For example, incomplete understanding of how people build metaphors, has led to the fact that the AI wrote a new Chapter of “Harry Pottery”, filling it with meaningless sentences, for example: “Half of the castle was like a big pile of magic.”
And still make the machines to accurately simulate the human style of Rembrandt or Rowling, it does not matter — a good start in creating a creative AI, said Ridley. In the end, the creators often begin with imitation skills, and processes established artists. The next step, both for people and for machines, will be using these skills as part of a strategy to create something original.
Modern programs the AI is not sufficiently developed to spontaneously compose hit songs or works of fine art. So the AI did this person needs to calibrate the program by feeding her a huge number of examples. German Mario Klingemann, for example, designed a neural network that is able to make strange frightening images based on existing photographs and other works. The neural network consists of a series interconnected processing nodes, resembling the neural structure of the brain. In the neural network, each electronic “neuron” takes an array of numbers, makes simple calculations based on this input and then sends the result to the next layer neurons, which in turn produces more complex calculations.
Approach Klingemann includes feeding a source material, drawings and photographs, generative adversarial networks (GAN), which combines two neural networks. One network generates images that are a specific theme or set of conditions; the other evaluates the images based on your knowledge of these conditions. Thanks to the feedback of the second network, the first network is gradually becoming better and makes the images more relevant to the given theme. “Now these networks are just tools that complement our own creativity,” says Klingeman. “We, the people, still need to recognize creativity or innovation”. His goal is to create art neural network, which can independently choose and even to publish their best works on a given topic.
Now GAN is used strictly for creating new content or images in a broader creative system, says Alex Champandard, founder of creative.ai, a startup in the development of AI tools for creative people. GAN can produce a lot of material, but still rely on people to define their terms.
According to Ian Goodfellow, research scientist at Google, which develops the concept of the GAN, content generation is a good start for the development of AI, capable of solving real-world problems. Goodfellow is working on machine learning models that enable computers to write dynamic narratives that go beyond the limited scenarios (like planning chess moves), where computers have long excelled.
Take the classic example of planning that people do all the time: when we go to the airport, we often apply our approximate map — pure in mind — the key points of the travel, traffic jams or transplant. GAN can plan such a trip, but will do it with all the details and will offer a lot of routes. We, in fact, the desired layer of the network that will skip all these options and intuitive to choose the best.
Another key component of human creative thinking is the ability to take knowledge from one context and apply in another. Gorgg Harrison taking sitar and plays it like a guitar. Shakespeare takes the story from Greek mythology and writes English play inspired by these stories. The Executive Director uses knowledge of military strategy or even chess for planning business transactions.
For this reason, experiments are being conducted that will help the AI algorithms that can mix and match material. For example, scientists at the University of California at Berkeley used a network of CycleGAN for converting the video with the horses in the video of the zebras. AI defines the basic shape of the horse in the first video and plays with the image over the video, instantly and invisibly substituting brown torso horse striped Zebra on the move. This work will help the AI self-driving car to adapt to unfamiliar conditions and to avoid accidents.
Artificial intelligence should not only teach the rules, but break them like an artist.