Can AI Help With Classic Art Restoration?

AI is unbelievable when it comes to the enormous scientific projects, while it is even more surprising to watch AI performing daily chores. Well, there is more than that, which is now AI can be seen in performing valuable artworks, such as writing poems, creating marvels in paintings. So we could see that there is now this postmodern classic art era, where art is experiencing a renaissance through AI.

Artificial intelligence worked on “Portrait of Edmond Bellamy” not alone, but together with three 25-year-old Frenchmen who call themselves the art group Obvious. They created a series of such works – they painted a whole non-existent Belamy family. And the work sold at Christie’s was not the first in this series, on which they earned: the painting “Countess Belami” six months earlier brought them 11,430 dollars.

The surname Belami was taken for a reason: the primary Algorithm used to create a series of these works was written by a developer named Ian Goodfellow (Goodfellow), and the entire series of paintings became a kind of homage to the developer.

The very fact of selling paintings created with the help of AI is not uncommon. For example, at a local auction in San Francisco in 2016, 29 works by the AI ​​development team from Google were sold at once. Their total cost was $ 98,000, the most expensive was estimated at 8,000.

In 2017, over 16 thousand dollars bought a painting, the work of an extremely self AICAN program – it generates a picture, assesses the level of their creativity, and the work itself gives a name. On display at India’s Nature Morte Gallery, AI paintings were sold in a price range of $500 to $40,000 per piece.

Online Gallery for AI

The online gallery SuperRare regularly trades in art objects created “with digital tools.” They are sold exclusively on the Internet and for cryptocurrency. Among other things, the site guarantees the protection of the purchased work from counterfeiting: all rights to work are protected by the blockchain, that is, the system constantly checks the Internet for the presence of similar images – it will not be possible to put a duplicate on the network imperceptibly. The online gallery website claims they had sold more than 7,200 works by early July, totalling in the order of a million dollars.

A historical event in the sale of the painting “Portrait of Edmond Bellamy” was that the painting was put up for auction by the auction house Christie’s – one of the two main world sites of the art market. In the art world, this automatically means recognition – both for the artist and a new direction in art.

At the same time, AI has been in demand for creativity for about fifty years. It is believed that the first to use algorithms to create artwork was the British artist Harold Cowen. Since 1973, he has been developing the AARON program, which creates unique paintings by following a set of specific rules. Cowan’s work has been noticed in the art community and has been widely exhibited, including in one of the leading British galleries – Tate.

How AI Works

In the case of the creation of creative works, artificial intelligence is called a neural network (aka machine learning algorithm). Thus, it is a special case of AI.

A neural network is a set of simple, interconnected elements that add up to a primitive semblance of a brain. It analyzes the works uploaded to its database, recognizes images, techniques, and signs of style, and then uses the knowledge gained to create paintings according to the prescribed assignment.

The machine learning algorithm can find dependencies, traits and rules in any set of loaded objects – be it painting, music or poetry. Then, after processing a large amount of data, the neural network summarizes them and concludes in different ways. The process is somehow reminiscent of the RNG of Vulkan Bet slot games.

An average of the results of training a neural network can be, for example, knowledge about the sizes of the most typical brushstroke in loaded paintings of the Impressionists. Or about the frequency of using obscene vocabulary in the poetry of the early Yegor Letov.

What precisely the Algorithm will learn is determined by the data scientist. He also programs the Algorithm on how to use them. For example, write poems similar in size and vocabulary to those uploaded to the database, or it could be dissimilar for novelty. Neural networks are very good at imitating a specific artist or compiling the styles of several at once.

Successful Algorithm in AI

The most successful Algorithm that underlies the most modern programs for creating any artwork is considered Generative Adversarial Network (GAN) Iana Goodfellow.

The GAN architecture consists of two neural networks 

• A generator: acts as an artist – uses mastered techniques to create images.
• A discriminator: it acts as a critic and compares what the generator has done with the original works. If the discriminator cannot distinguish the resulting image from the picture painted by a person, the result is accepted. On the other hand, if the discriminator decides that the proposed picture is a fake, the generator starts over.

You can configure the discriminator so that, based on evaluating the resulting work, it will indicate what caused a doubt. Then the generator will take note of this and will not repeat the error. That is, he will continue to learn and improve.

GAN is not the only Algorithm that successfully copes with creative tasks; it has quite a lot of colleagues in the shop. For example, another popular neural network that creates this kind of works isCAN, a creative adversarial network. It is also based on two neural networks – a generator and a discriminator, but the second, in this case, select the resulting products to not look like any work loaded into the database. Thus, with the help of CAN, stylistically unique works are created.

Concluding Thoughts

A normal human being might think that this could be the limit of AI work, but who can predict what more is coming ahead? However, it should be accepted that reviving classic art through advanced technologies is a huge milestone.