Can machines become artists? As artificial intelligence (AI) grows ever more sophisticated and accessible, it has found a place in art. Now, we are witnesses to the advent of a new genre of artwork: algorithmic art.
- Algorithmic art refers to “any art that cannot be created without the use of programming” Mazzone & Elgammal (2019). It’s usually created using machine learning techniques, usually Generative Adversarial Networks (GAN) or variations thereof.
- While introducing algorithms reduces an artist’s control somewhat, they still retain a majority of creative control. They choose which images the machine will learn from and curate which outputs make up a final collection.
- There is currently a debate about whether or not machines can be creative. One perspective argues that machines can be creative if their works are novel and indistinguishable from human-made art. Conversely, others argue that creativity is intertwined with an artist’s lived experience, culture, and artistic intent— all of which cannot be replicated by machines.
- However, machines and artists do not have to be adversaries. Instead, they can be collaborative, where the machine enhances the artist’s creativity by developing new types of work.
- While the future of algorithmic art is currently unknown, it will certainly unlock new realms of possibility by allowing artists to explore and discuss society, culture, the environment, and more, in new ways.
Part 1: What is algorithmic art?
According to Mazzone & Elgammal (2019), algorithmic art refers to “any art that cannot be created without the use of programming”. The genesis of algorithmic art opens new realms of possibility for artists as it generates novel ways to discuss society, culture and the environment. Artists —even those with minimal coding know-how— can use open-source software to produce unique images that don’t merely imitate existing art styles. Oftentimes, they’re surreal, uncanny, and beautiful.
There are multiple ways algorithmic art is created. For instance, the Twitter account @images_ai uses an algorithm named CLIP and a Generative Adversarial Network (GAN). More specifically, CLIP is an open-source software trained to identify correlations between words and a given image. For example, it would note a weak correlation between the word “dog” and an illustration of a building. The two algorithms iteratively transform pixels of static into image after image until it proceeds to correlate with the given prompt.
Part 2: How do GANs work?
At the heart of current algorithmic art lies GAN, a machine learning technique that enables computers to extract patterns from sets of images using two neural networks, producing an output complementary to a provided set of images, or a “dataset”. These two neural networks can be termed “artist” and “critic”. Similar to trial and error, one (the artist) extracts patterns from the dataset and generates an image from it. The other (the critic) searches for discrepancies between the generated images and the originals to feed back into the artist. Then, the artist generates new images that incorporate these discrepancies (Still & d’Inverno, 2019). This process repeats over and over until a final output is created.
While it seems like the machine is in pure control of the work, the artist is still essential to the creative process. They pre-curate the artwork by selecting images to feed into the algorithm, tweak the machine, and sift through many outputs to curate a final collection of artworks (Mazzone, & Elgammal, 2019). For example, a GAN can be fed with images of Impressionist artwork exclusively in an attempt to reproduce Impressionist works.
A majority of algorithmic art then, still involves the artist and their intentions to a great degree. So what if you tried to exclude the artist to the greatest degree possible?
Part 3: Can machines replicate creativity?
This question spurred the development of creative adversarial networks (CANs) and AICAN, the machine that uses it. It’s a variant of GAN that attempts to create images that deviate from pre-existing styles. It has two main directives: be novel and different, but adhere enough to conventional art norms to be recognizable as art. According to Mazzone & Elgammal (2017), its creators, “The process simulates how artists digest prior art works until, at some point, they break out of established styles and create new styles.” Interestingly, when people were presented with works made by AICAN, they were assumed to be man-made 75% of the time.
Where AICAN differs from other algorithmic art techniques is in the degree of control the artist takes on. They do not pre-curate and tweak the machine like they do for GANs; they only select which outputs to present. The machine is not fed carefully chosen images, it’s fed a variety of styles and left to discern what “novelty” is on its own.
However, to treat AICAN and other machines as artists, one must assume the product–the image, sound or film– is the most essential part of art and creativity. Additionally, you must assume that the artist’s lived experience, worldview, and socialization are secondary to the product and may be discarded if the machine-made work is indistinguishable from a human made one. Still and d’Inverno (2019) contest this, saying “‘art’ and ‘artist’ is not just the making of novel products that are regarded as creative, but the way of life that accompanies this making”. They argue that the cultural context and artistic intention of a work is fundamental and inextricable from creativity, both of which cannot be replicated by a machine.
Part 4: The path forward is collaborative
“Art has always existed in a complex, symbiotic and continually evolving relationship with the technological capabilities of a culture.”Agüera y Arcas, 2017
Machines can do many things humans cannot do, and humans can do many things machines cannot do. For example, AICAN’s images, though compelling, lack social context– they are isolated from the complex ideologies, politics and conversations happening in the human world. Conversely, art created by humans is constantly engaging with the world. The reconciliation of the two, therefore, relies on collaboration between machines and human artists. The machine is both a tool and a medium. As such, the value of machines in art is largely contingent on its ability to extend the range of a human artist’s abilities.
With digital art as the centerpiece of successful human-machine collaboration, it is beyond question that human and machine may also work synergistically in alternative forms of art: music. An example of a machine that enables this is the Reflexive Looper system. It places the artist at the center of the design process and uses algorithmically generated music to create new types of performances. The musician chooses a repeating chord sequence and records themselves playing (among others) a bass line, a motif or a melody. The machine then generates new melodies and accompaniments in-real-time. It succeeds at being creative because it extends the range of the human artist through dialogue with the machine; it stimulates work that could not be created without it (Still & d’Inverno, 2019).
Part 5: Final thoughts and the future of algorithmic art
AI is fundamentally disruptive to the norms of any industry, even art. Though more importantly, it unlocks new realms of possibility by allowing artists to explore, discuss, and remake the world in new ways. As artists grow more accustomed to these machines and as the developers continually improve them, the eventual vocabulary, styles and aesthetics of algorithmically generated artwork will be decided not by machines in isolation, but by artists, technologists and machines working together in concert. But we are in early days yet– as algorithmic art stretches the horizon of possibility, what work do you think will come next?
WRITTEN BY ANNIKA MARIE YUNQUE
EDITED BY CHRISTIANA REYES
GRAPHICS BY ALIE SANTANA
Still, A. & d’Inverno, M. (2019). Can machines be artists? A Deweyan response in theory and practice. Arts, 8(1). https://doi.org/10.3390/arts8010036
Mazzone, M. & Elgammal, A. (2019). Art, creativity, and the potential of artificial intelligence. Arts, 8(1).https://doi.org/10.3390/arts8010026
Agüera y Arcas, B. (2017). Art in the age of machine intelligence. Arts, 6(4). https://doi.org/10.3390/arts6040018
Chayka, K. (2021, August 13). Appreciating the poetic misunderstandings of AI art. The New Yorker. https://www.newyorker.com/culture/infinite-scroll/appreciating-the-poetic-misunderstandings-of-ai-art
Elgammal, A. (2018, October 17). Meet AICAN, a machine that operates as an autonomous artist. The Conversation. https://theconversation.com/meet-aican-a-machine-that-operates-as-an-autonomous-artist-104381
Images_ai. (2021). A hushed song at the edge of existence [Photograph]. Twitter. https://twitter.com/images_ai/status/1430225515995181061
AICAN. (2018). Alternative Facts: The Multi Faces of Untruth [Series of Portraits]. The Conversation. https://theconversation.com/meet-aican-a-machine-that-operates-as-an-autonomous-artist-104381