Recently I wrote about Generative Adversarial Networks, a revolutionary class of AI algorithms used in unsupervised Machine Learning, and their importat relation with the research in Computational Creativity.
This type of framework for the design of Artificial Intelligences was invented by Ian Goodfellow in 2014. The system consist of one network that generates new data after learning from a training set, and another that tries to discriminate between real and fake data. By working together, these networks can produce very realistic synthetic data. According to Yoshua Bengio, one of the world’s leading experts on Machine Learning, this approach offers a powerful way for computers to learn from unlabeled data and will help make software that learns without explicit instruction.
We agree with web-developer Afaan Bilal that the idea of something that can learn and adapt to changing circumstances and produce information and data relevant in solving real world problems has always been one of the top-most researched areas in the field of Computer Science. Consequently the growing importance of Deep Learning architectures such as Artificial Neural Networks, used in Generative Adversarial Networks for example, a kind of Computational Connectionist Systems inspired by the biological neural networks that constitute animal brains. This systems learn to do tasks by considering examples without task-specific programming. As neuroscience has shown, all information that our brain processes and stores is done by the way of connections and networks between different neurons and that’s the concept on which this Artificial Neural Networks (ANN) are based upon.
As we have seen Connectionism is an approach in the fields of cognitive and computer sciences that hopes to explain mental phenomena using ANN. But while this technology was initially designed to function like biological neural networks, we agree with professor Margaret Boden that the activity in our brains is far more complex than might be suggested by simply studying artificial neurons. According to Dr. Gaetano Licata the high complexity of the human brain makes it impossible to consider neural networks as good models for human mind, but that doesn’t mean they are not good devices for computation in parallel. In contrast Blake A Richards, a professor in University of Toronto, notes that in the next decade discoveries by scientists will put us on the path of new AI which will help us understand our experimental data in neuroscience.
According to programmer and Deep Learning enthusiast Jonathan Hui Generative Adversary Networks (GAN) are about creating. Like drawing a portrait or composing a symphony the main focus for GAN is to generate data from scratch. As stated before GAN are designed with of two Artificial Neural Networks, the generator that feeds with data and the discriminator which acts as a critic. Just because a generator alone will just create random noise, conceptually the discriminator provides guidance to the generator on what images to create. By training both networks into a fierce competition to improve themselves, eventually the generator creates images that the discriminator cannot tell the difference and the GAN model eventually converges and produces natural looking images.
As maintained by researchers Glenn W. Smith and Frederic Fol Leymarie we can now begin to think of the machine, not as the artist’s subject matter or medium, but as creator or co-creator. With the current development in technology, and with GAN especially, we can without a doubt begin to speak comfortably of the machine as artist and consider that an aesthetic sensitivity on the part of the machine might help lead to a friendlier and more sensitive machine intelligence in general. According to Blaise Agüera y Arcas -founder of the Artists and Machine Intelligence program at Google– the transformation of artistic practice and theory that attended the 19th century photographic revolution is equal to the current revolution in machine intelligence, which promises to democratize the means of reproduction and production.
Many will remember here an essay of cultural criticism called “The Work of Art in the Age of Mechanical Reproduction” (1935) by Walter Benjamin, which proposes that the aura of a work of art is devalued by mechanical reproduction. Some may argue -in a clear romantic stance- that modern means of artistic production and reproduction like photography and video destroyed the aesthetic, cultural, and political authority of art. Contrary to this belive we should accept that the diffusion and popular use of different technological innovations, like photography or machine learning, will help us explore new domains of art and creativity.
Nowadays artists are beginning to change their relationship with Artificial Intelligence to be more of an ideation partner, rather than simply a tool for making new tools. Mario Klingemann for example builds art-generating software by feeding photos, video, and line drawings into code borrowed from machine learning research. In his opinion he is like a photographer who goes out into the world of neural networks and frames good spots. As Jennifer Sukis Design Principal for AI & Machine Learning at IBM writes: “We need art to imagine what AI can become, and understand it’s impact on who we are becoming”.
Ahmed Elgammal, professor of Computer Science at Rutgers University and director of the Art and Artificial Intelligence Laboratory, wonders what will happen if we teach machines to generate novel images… Would it generate something that is aesthetically appealing to humans? Would that be considered “art”? Their creative system is inspired by the psychological theory of art evolution proposed by Colin Martindale who hypothesized that at any point in time creative artists try to increase the arousal potential of their art to push against habituation. Elgammal told world-famous auction house Christies, in relation to the artworks created by their AI, that “if you consider the whole process, then what you have is something more like conceptual art than traditional painting”. There is no doubt that in this case Computational Aesthetics helps us improve understanding of human aesthetic perception. Because when intelligent machines start generating their own designs and art pieces, free from our aestethic constraints, how are we as human beings going to be able to understand their own original outcomes? Are we ready to adapt and fall in love with non-antropocentric forms of creativity?
Philosopher and art critic Arthur Danto thought that theory is what takes an artwork up into the world of art and that the role of artistic theories is to make the artworld possible. Danto quoted Leonardo da Vinci by saying ogni dipintore dipinge se (“every painter paints himself”), but today we may want to say ogni automăta dipinge se (“every automata paints himself”)…
*(Pictures: technologyreview.com, becominghuman.ai & "Permutations" by AICAN.io — Ahmed Elgammal).