ISSUE 1 ⎯ FEBRUARY 15, 2020
CREATIVE MACHINES

____________________________

 

All about emerging artists exploring the creative power of AI
and the most prominent personalities in the field of AI and generative art. 

 

 

WHAT DO MACHINES WANT?

by Anne-Marie Boisvert

“The devil, commenting on the first man’s first drawing on Earth, whispered into his ear: “That’s good, but… is it art?” (Orson Welles, F for Fake, 1972).

 

Portrait of Edmond of Belamy, 2018, de l’artiste IA « min G max D x [log (D(x))] + z [log(1 – D (G(z)))] ».

 

Throughout the twentieth century, many artists have taken on the task of questioning and constantly stretching the limits of what art is, or could be; especially so in so-called conceptual art. Less an artistic movement as such, the latter corresponds with the artist’s adoption of a critical point of view that leads him to take position in his work, or rather to make the work itself a positional stance with respect to the art world, and to the world itself. The border lines between art and business, in particular, between the object traditionally called “art object” and the manufactured object, have been examined and their legitimacy questioned. But what makes a work of art a “work of art” in this context is the flawless simplicity of the typical gesture of conceptual art at its best, as well as its innovative and provocative character. The value of the “work” here merges with the assurance with which the artist moves from intention to action. This value is also a function of the way in which said intention and said action are received perceived in the world of art in which they are embedded.

 

Intention. So the word have been uttered. Can there be a work of art without intention? “I want to be a machine,” said Andy Warhol in 1963. Nowadays, machines are becoming artists. But can they really want it?

 

Andreas Mueller, For All Seasons, 2005

 

Computer-generated art dates back to the 1950s and is far from being a new phenomenon. The term “computer art” itself dates precisely (is it a coincidence?) from 1963; and the first article on the subject appeared in 1964 [1]. This article is aptly entitled “The Electronic Computer as an Artist”[2]. In the 1990s and 2000s, the advent of the web allowed “web” or “digital” art to flourish in multiple forms, emphasizing in particular the interactivity with the spectator / visitor facilitated by this new medium.

 

Raymond Queneau, Cent mille milliards de poèmes, 1961

 

The CIAC’s Electronic Magazine published between 1997 and 2014 forty thematic issues which each explored a facet of this type of art. In theTwo of those issues in particular are worth mentioning in the context of AI art: issue 19 devoted to randomness in art (published in 2004) and issue 24 (published in 2006) to the concept of constraint. In these issues, we often went back to pre-computer time to look at the interplay between chance and necessity, the intentional and the unintentional typical of many artistic projects throughout history, and in particular since the beginning of the twentieth century – from exquisite corpses to OULIPO[5] to computer generative art – in order to recall the roots of web art and to understand its nature. Thus Evelyne Boudroux, lecturer in Information and Communication Sciences at the National Institute of Documentation Techniques (INTD) in Paris, observed in issue 24 that “today, generation has become a part of many digital works. From the concept of the program text (as Perec employed it in his organisational chart “L’art et la manière d’aborder son chef de service pour lui demander une augmentation” (“How, irrespective of sanitary, psychological, climatic, economic and other circumstances to stack the odds in your favour when approaching your Head of Department with a view to requesting an increment”)11, we have moved on to that of the programmed text, a text produced by a program. This concept of the program is what the authors Philippe Bootz, Alexandre Gherban, Tibor Papp, Jean-Pierre Balpe and Antoine Schmidt have brought to the group Transitoire Observable, around the concept of “programmed art and literature”. But is it still literature? [6]

 

The same question arises today about “artificial” art, that is, artworks created by or with the assistance of “artificial intelligence”. Is it still art?

 

Artificial intelligence is best defined, in the widest sense possible, as the area devoted to the construction of “intelligent agents” who receive environmental perceptions and perform actions, and who imitate and sometimes surpass human behavior. Although artificial intelligence has its roots in logic and philosophy of mind, the field was officially born at a conference sponsored by the Defense Advanced Research Projects Agency (DARPA) in 1956[7], whre the term “artificial intelligence” was coined[8]. Artificial intelligence covers many sub-fields, the main ones being robotics, expert systems (knowledge-based) and machine learning.

 

Frank Rosenblatt’s perceptron (1957)

 

In the early days of artificial intelligence, rule-based systems dominated the field. They have proven to be very effective at solving problems and performing tasks such as arithmetic operations which can be described as a list of explicit formal rules hand-coded by computer experts. But many more intuitive tasks, such as speech or image recognition, could not be approached this way, because they are difficult to describe formally. In contrast, machine learning systems (so-called “learners”) do not need an explicit list of instructions to complete a task or solve a problem: they learn by themselves by identifying patterns, by making generalizations and inferences from the data, on the basis of which they are able to make precise predictions on material never encountered before. Machine learning is not a new idea. Precursors include Arthur Samuel’s checkers program created in 1952[9] and Frank Rosenblatt’s perceptron in 1957[10]. Inspired by brain neurons, it is the first artificial system able to learn by experience. Machine learning systems have experienced a resurgence in recent years, thanks to the explosion of big data and the dramatic increase in computer power. Machine learning should not be solely assimilated to neural networks (and whose deep learning – “deep learning” – is a particularly effective form). Machine learning also encompasses other types of learning, each with its own languages ??and tools of choice [11].

 

In the art world, very recently (in October 2018), a sale at Christie’s created a stir and raised a great deal of controversy. The cause of all this was the “portrait” of an imaginary gentleman, Edmond de Belamy, whose “author” is a generative adversarial network (known as GANs for short) (see the portrait at the beginning of this article). This type of system has been developed by Ian Goodfellow et al. in 2014. GANs take advantage of the fact that neural networks used in deep learning, and that are proven to be so efficient in many cases, in particular for image and language recognition, are nevertheless easy to deceive. Hackers have not been the last to notice. The originality of GANs is to use this weakness by transforming it into a force to improve the performance of the network. GANs thus put two neural networks into competition. The first network (the generator), trained with a massive database provided to it during learning as examples, generates an image. The second network (the discriminator) must determine the origin of the image: is it an image taken from the database, or an image created by the generator from the examples that have been provided? The goal is to generate an image as close as possible to the examples contained in the database, in order to trick the discriminator. In the case of the “portrait” of Edmond de Belamy, the generator was fed during its training with a huge database made up of painted portraits from the 14th to the 20th century. At the end of the process, the GAN generated thousands of portraits from which the one sold at Christie’s was selected, printed, framed and signed using a fragment of the code used (“min G max D x [log (D (x))] + z [log (1 – D (G (z)))] ”) by Obvious, a group of French students (Hugo Caselles-Dupré, Pierre Fautrel and Gauthier Vernier). The “work” was sold to an anonymous buyer for $ 432,500.

 

The controversy was sparked by several factors. First, the members of the collective are not the authors of the code used in the production of the “work”. The author of the code is Robbie Barrat, a young developer and artist who took the opportunity to start a discussion on Twitter about this[12].

 

The other controversial factors are of particular interest to us here, because they again raise the questions discussed above. Obvious members attribute the creation of their work to the algorithm. Their motto is: “Creativity is not just for humans.”

 

But how true is this statement? And if it is true, does it change the nature of portrait painting? Can a machine be the creator of anything? And what does it feel like to be a computer that makes art? Can the machine be considered to have intentional states? Must a machine to be creative have to be considered to have intentional states? In what sense can computer generated art be considered representational? And what does it represent? What can computer generated art teach us about creativity?

 

We will pursue this reflection in future issues.

 

Notes :

 

[1] See Grant D. Taylor. When the Machine Made Art. The Troubled History of Computer Art. 2014, introduction and chapter 1.

 

[2] Leslie Mezei and Arnold Rockmann, “The Electronic Computer as an Artist”, Canadian Art Magazine, vol. 21, no. 6, 1964, pp. 365–367.

 

[3] CIAC’s Electronic Magazine no 19, 2004, https://ciac.ca/documents/magazine/no_19/index.html

 

[4] CIAC’s Electronic Magazine no 24, 2006, https://ciac.ca/documents/magazine/no_24/index.html

 

[5] Ouvroir de littérature potentielle, founded in 1960 by Raymond Queneau, Italo Calvino, Georges Perec, and others. Official website: https://www.oulipo.net/

 

[6] Evelyne Boudroux, “Programming constraint”, CIAC’s Electronic Magazine no 24, 2006, https://ciac.ca/documents/magazine/no_24/en/dossier.htm

 

[7] The Defense Advanced Research Projects Agency (DARPA) is an agency of the United States Department of Defense responsible for the development of emerging technologies for use by the military founded in 1958. Voir : https://www.darpa.mil/

 

[8] See Selmer Bringsjord et Naveen Sundar Govindarajulu, « Artificial Intelligence », Stanford Encyclopedia of Philosophy, 2018.

 

[9] See Arthur Samuel’s page at https://en.wikipedia.org/wiki/Arthur_Samuel

 

[10] In machine learning, the perceptron is an algorithm for supervised learning of binary classifiers. A binary classifier is a function which can decide whether or not an input, represented by a vector of numbers, belongs to some specific class. The perceptron is the ancestor of the artificial neural networks used today in deep learning. See ttps://fr.wikipedia.org/wiki/Perceptron

 

[11] See Pedro Domingos, The Master Algorithm, Basic Books, 2015.

 

[12] See James Vincent, « How three French students used borrowed code to put the first AI portrait in Christie’s », The Verge, 2018, https://www.theverge.com/2018/10/23/18013190/ai-art-portrait-auction-christies-belamy-obvious-robbie-barrat-gans.

WHAT IF WE SUCCEED?

by Anne-Marie Boisvert

The threat posed to the future of humanity by the advent of a hypothetical “singularity” in artificial intelligence – that is, the advent of systems with “superintelligence” and getting out of control – has often made headlines in recent years…

Read more

NEWS

Publication: Olliver DYENS, La terreur et le sublime : humaniser l’intelligence artificielle pour construire un nouveau monde.

 

Event: The AI Art Lab, an international multidisciplinary laboratory using artificial intelligence (AI) in the creation of audiovisual works.

Read more