Of the many strategies to understand art, one of the most often employed is examining the intentions of the author. This involves an exploration of the techniques which the author used and the reasons which may have motivated those techniques. Debates on the importance or even the relevance of authorial intent are by no means settled, but looking at the ways in which artists construct their work is often useful for trying to understand art. However, the question of authorial intent is complicated by art generated seemingly without an author—art generated by algorithms. If the question of authorial intent only makes sense when there is an author who could have intention, how could it be applied to art generated by algorithms?
For the purposes of this essay I would like to leave open what art is so that the phenomenon of algorithmically generated art can be examined through the lenses of many different views. I am also assuming that algorithms are not conscious in the way that humans are—whatever human consciousness is—although I am open to the possibility that the lack of human consciousness in algorithms is only due to their primitiveness.
Computer composers and computer artists already exist with many works credited to them. As computing power and programming techniques grow, these algorithms will also become more complex as well. In the near future, perhaps even the beloved writers of the Texas Orator will be replaced by programs that report the news. There are even algorithmically written screenplays which—although far from Citizen Kane—are pretty good considering how they were written.
An algorithm is “a precisely defined set of mathematical or logical operations for the performance of a particular task,” according to the Oxford English Dictionary. They have been around for a long time, ancient Greek mathematicians created an algorithm called the “Sieve of Eratosthenes” to find prime numbers. The sets of operations which comprise these algorithms are complicated and varied, ranging from knowledge-based systems to systems which process music genetically, although often systems are combined together. Each requires a programmer to write the algorithm, and provide some data set—anything from an audio recording to a composer’s entire canon of work—which the algorithm transforms into something else.
Algorithmically generating music is not new, and many of the programmers who create the algorithms stress their innocuousness. Mozart is credited with the creation of the “Musical Dice” game, in which preselected fragments are organized according to fair dice tosses—though it is telling that none of his most loved pieces were created this way. Arguably, this process is similar to how humans create art. We are taught certain rules of composition by mentors, take in the body of available work, and then create something new.
Unlike humans, algorithms are not conscious actors. At no point does it make sense why an algorithm makes a particular decision. Nor would it make sense to ask about how the algorithm interprets or reinterprets the work given to it by its programmer in the way a human author might recreate a work of art into something new and original. There cannot be an emotional connection, as Tolstoy suggests in “What is Art?,” because algorithms do not have emotional responses or experiences. As he defines it, “Art begins when a man, with the purpose of communicating to other people a feeling he once experienced, calls it up again within himself and expresses it by certain external signs.”
Tolstoy’s definition precludes any sort of algorithm because algorithms do not have feelings, ideas or experiences to communicate. When an algorithm faces the hypothetical choice between two words with similar meanings such as “best” and “unrivaled” it does not pick one in an attempt to communicate a particular idea. It chooses because the works in its dataset or its rules prescribe what decision to make. Roger Ebert’s criticism of Battlefield Earth sums up the point: “The director…has learned from better films that directors sometimes tilt their cameras, but he has not learned why.” Algorithms cannot know why a particular technique is effective or what the effect of a technique is. They simply do what they have been programmed to do.
One possible avenue for avoiding the problem of authorial intent is by assigning authorship to the programmers. After all, the programmer consciously curated the dataset which the algorithm uses and set the rules by which the algorithm functions. The conception of the programmer-as-artist requires a broad understanding of the relationship between the artist and the art, as the programmer may not literally know what is being created. Even in improvisational approaches to art like jazz, musicians may not know ahead of time what they are going to do, but they still draw from years of experience and training as well as their aesthetic and emotional inner lives. Even somewhat chaotic techniques like Jackson Pollock’s action painting are subject to a level of control that programmers do not have over their algorithms.
A programmer attempting to provide artistic justifications for his or her work would be forced to significantly limit the algorithm’s choices. Such an algorithm would not be capable of adding much to the artistic product. The alternative would be for the programmer to have no significant connection at all to what an algorithm produces beyond curator and teacher. Both of these are important to the growth of an artist, but are not sufficient for actually being an artist—not unless the credit for Mozart’s symphonies should actually go to his father and teacher, Leopold Mozart.
Another possible way to address this problem is by changing the definition of art to one that does not include authorial intent. What would count as art according to this view is a mystery. Could anything count as art, even removed from context? If there is no authorial intent at all in art then the pattern of rainfall becomes a song generated by the processes of weather as much as any chorale created by the EMI algorithm. Weather patterns are, after all, patterns as much as the mathematical patterns in algorithms are and that might include too much.
In order to account for authorial intent in art, algorithmically generated art would either require that the programmer of the algorithm becomes the artist, or that authorial intent be removed from art entirely. The first view requires seriously reevaluating what relationship artists have with their art, and the second requires opening up the field of art to almost anything which is problematic in that it includes too much. But this leaves a baffling question: What is algorithmically generated ‘art’ if it cannot be art?