Artificial Intelligence Is Making Arts More Artistic

The influence of Artificial Intelligence on our lives has been increasing. However apparently one area that seems untouched by it is arts, i.e. paintings, images, music, poems, stories etc. One reason for this perception is that arts is considered to be in exclusive domain of human creativity and the Mathematics heavy base of Artificial Intelligence makes it look so distant from it. However advances in Artificial Intelligence are impacting arts too.

New technologies

Within Artificial Intelligence, development of neural networks based on Deep Learning (DL) revolutionized its practical applications. They use multiple layers of non linear mapping and are somewhat similar to the way human brains work. 

One important variant of neural networks is Generative Adversarial Networks (GANs). In GANs two networks work together, one creates fake output and second one detects it and sends back feedback leading the first network to improve its output. DCGANs use deep convolutional networks with GANs and hence are computationally more efficient. Another variant of neural networks are Autoencoders that target to reproduce the input.

Yet another development is emergence of Reinforcement Learning which works by providing feedback on each step rather than just setting the end goal as happens in DL. RL has been used in creating algorithms that have bested the world class Chess and Go players.

Images and paintings

DCGANs and to some extent Autoencoders have revolutionized the field of images and paintings. For human faces, it is possible to change the pose, add or remove beard, hairs, glasses, smiles, laughs etc. One website, https://thispersondoesnotexist.com/ generates real looking pictures of non existent persons. In images, the background can be changed, black and whites images or even sketches can be changed to coloured ones. Watermarks or effects of rains can be removed, objects can be changed e.g. one animal to another etc.

Artificial Intelligence is also emerging as an important collaborator in the field of paintings. It is possible to create paintings with style of a particular painter from a photograph or create paintings with merged styles.  AI-da is a mechanical robot that can paint and even sculpt. She has facial recognition capabilities and hence can scan people for input. Her work “Unsecured Features" was exhibited at St. Johns College, Oxford.

The use of Artificial Intelligence in paintings captured public attention when a painting, “A portrait of Edmond Belamy” created using GANs were auctioned by Christie for $ 432,500/-. It is one of a set of 11 paintings of Belamy family. 

A major improvisation to GANs is AICAN, a creative adversarial network developed by Rutgers University. Some of the generated paintings by AICAN have been shown in many art exhibitions including at New York gallery. Its famous “Faceless Portraits Transcending Time” paintings have portraits but without details of faces. It was feed 100,000 paintings to learn. However, it can deviate from the learned styles while drawing and hence can be creative. It can also generate titles for its paintings. 

Music

Music is more complex than images as it is composed of number of instruments playing their own part together in any time period. Any sampled part will be heavily dependent on the previous part and any generated music should preserve melody structure which in turn is composed of pitch and rhythm. Melodies can be further extended by chords which is composed of multiple harmonic pitches.

Reinforcement Learning, GANs and Autoencoders are the key tools here.

AIVA (Artificial Intelligence Virtual Artist) is a known electronic composer for creating sound tracks for movies and commercials and covers wide range of emotions. It can take a score as input and generate music based on a track. It can also create music using various pre set styles e.g. Jazz, Rock etc. 

There are various Artificial Intelligence based software e.g. David Cope’s Emily Howel that can compose music in its own style, Sony’s Flow Machines that can generate melody, Google’s Magenta etc. Emily Howel’s 1st album had compositions for orchestra and multiple pianos. Flow Machines wrote the first Artificial Intelligence song, “Daddy’s Car”. Magenta created a 90 second piano melody.

Artificial Intelligence is helping music industry in other ways too e.g. offering listeners personalized playlists that span across multiple genres e.g. Rock, Jazz etc; audio mastering that involves balancing sonic elements to maximise listener’s experience for different media prior to distribution; finding promising artists based on popularity of their creations etc.

Literature

Deep Learning is the mainstay of NLP today. Introduction of GANs has spurred more development. Natural Language Generation is already used extensively by newspapers e.g. Washington Post, the New York Times, and Norwegian news agency NTB etc to create news. Artificial Intelligence helps in recommending books to the readers based on their preferences and new topics to the writers based on keywords in best sellers and estimated royalties. 

There have been numerous successes in writing poetry. Google’s PoemPortraits generates poetry with just one word as input. Another initiative from Google creates poem and takes the first and the last sentence as inputs. A Chinese company, Cheers Publishing created 10,000 poems after reading 500 Sonnets. It is possible to create a poem based on an image as an input. Many other software can write poems e.g. WASP, Deep-Speare, Pentametron etc.

Now Artificial Intelligence is also writing stories and in fact novels. GTP2 project of OpenAI aims to predict the next word following the existing text. Hence it can create a full story based on a small input. One of the novel written by Artificial Intelligence, “The Day a Computer Writes a Novel” made it beyond the first round in The Nikkei Hoshi Shinichi Literary Award competition. DeepBeat from Aalto University can write lyrics and screenwriting of the movie, “Sunspring” was written by the software, “Benjamin”.

Future

The scope of Artificial Intelligence in arts is increasing and it is also impacting other areas e.g. sculpture and dance and choreography.. However its role is mostly of a collaborator of artists and business driver to the industry as it still lacks emotional intelligence and feelings. It has made the maximum impact in the field of images and paintings as compared to other arts. It will cross the watermark of creating arts autonomously before the middle of century and that will further squeeze the space that humans think they alone have intelligence to handle.