Artistic Creativity: Divine, Profane or Mechanical
Renaissance artist Albrecht Dürer is noted for his arrogance, or what some might call his hubris. In a self-portrait made at age twenty eight Dürer depicted himself in a pose deliberately echoing that of Christ. He was signalling his divine-like powers of creation. But why should he have thought that his creativity was divine? History has produced countless notably creative people, e.g., Leonardo da Vinci, Jane Austen and Vincent van Gogh, none of whose powers we normally associate with the divine. Well, if we think of creativity in general as the power to produce something out of nothing, then it is understandable to associate it with the divine. One definition of a god – or God for the monotheist – is an initiator in this sense.
The ancient Greek philosopher Plato famously propounded an inspirational theory of creation, or poesis to use the Greek term. He conceived of the poet as a medium. The analogy he uses to illustrate this idea is of a chain of suspended metal objects held up by a magnet.[i] The god’s magnetism is the creative force that passes through each person, growing weaker as it ‘travels down’. The greatest poet, therefore, is the one who is directly inspired by the gods, contiguous with the magnet. But it is the gods who provide the something, where before there was nothing. Mortal poets merely receive their inspiration. They are empty vessels. We of course view creativity in more profane terms, thinking of artist, musician or poet as the independent source of original ideas. How do they come by them? No answer readily comes to hand. And certainly saying it is a result of a supernatural external force, i.e., God’s, is by our lights no answer at all.
All the same we are resistant to the idea that creativity is a mechanical power, that is, a power that can be realised by a machine. Many of us feel that the power to create is a distinctively human one; it is not a matter of step-by-step processing of information. Again, this bias is explained by the thought that originality, as a mark of creativity, is a matter of producing something out of nothing. The artist’s painting is original insofar as it is not the result of having looked at what has been painted before and randomly recombining some of the elements of these previous artworks. No, the painting presents something new, something surprising and unexpected. We get joy out of experiencing such novelty in a painting; and we understand it to be relatively rare, and hence valuable.
That said, teams of IT experts sponsored by corporations like Google and Microsoft have been busying themselves attempting to produce genuinely creative machines. Recently, for example, Microsoft teamed up with ING in a project aimed at producing a new Rembrandt van Rijn portrait. Admittedly this is prima facie an absurd project given the artist has been dead for around three hundred and fifty years, but you get the idea. Ron Augustus, a Microsoft marketing director involved in the project, explains that “we haven’t been using data much in a way that touches the human soul. You could say that we use technology and data like Rembrandt used his paints and brushes to create something new.” [ii] The team scanned as many Rembrandt paintings as they could access and used this data as a resource to produce ‘a new Rembrandt’. They decided on producing a typical looking Rembrandt portrait. That consists in a caucasian male who appears to be in his thirties, with facial hair, wearing dark clothing, a collar and a hat, and who is facing the right. Analysing all the data they had on such portraits they calculated the facial proportions and averaged out the various facial features. All these details were collated. They were then able to produce a three-dimensional print of what is effectively an Ersatz portrait by Rembrandt, that is, a picture which the ordinary viewer would readily believe was painted by him.
The project’s spokespersons claim to have created a new Rembrandt no less. But is this handsome picture the product of a genuinely creative process? In one sense the picture is new since no existing Rembrandt portrait looks exactly like it. Nevertheless, given the parameters of the picture, one would have a very good idea how it would eventually turn out. There is nothing surprising or unexpected about it. Further, the process is not creative in terms of producing something from nothing. The team ultimately rearranged the features of various Rembrandt portraits with a high level of detail by using lots of data.
Similar schemes have been devised with respect to music. David Cope, for example, has developed sophisticated algorithms he named ‘Emmy’ that produce music in the style of the various composers whose works Cope has gathered data on. The resulting Emmy compositions are in the style of Chopin, Beethoven, Bach and so on. Again these derivative works offer no surprises. They, like the Ersatz Rembrandt portrait, are predictable. Emmy Howell – a later reincarnation of Emmy made by Cope – is an attempt to produce a less identifiable style of music.[iii] But again there are no surprises.
These works are derivative; that is to say, they are not creations but rather are rearrangements of already existing works by humans. The daughter of Lord Byron, Lady Ada (Augusta) Lovelace, was a pioneer of proto-computer technology in the early Victorian era. She worked with the inventor of the Analytical Engine Charles Babbage. Lady Lovelace famously remarked that given that machines can do only what humans program them to do they cannot be said to be intelligent. The mark of intelligence, she insisted, is originality.
Around a century later the English mathematician and pioneer of modern electronic computers, Alan Turing, objected to Lovelace’s diagnosis.[iv] He argued that because of the primitive nature of the computer that she and Babbage worked on, she could not have anticipated the possibility of machines’ having the capacity to be original. But now we can imagine such a machine. He equates originality with unpredictability. He notes that there is no reason to suppose computers cannot surprise us. Moreover, if computers can be built with the ability to learn, then we expect them to produce original ideas. A machine is said to learn when it develops its own abilities to act, rather than receiving them from without, i.e., from a human programmer. So, humans program the computer to learn and what it learns is outside their control, and hence the results of the computer’s actions become unpredictable. In other words the computer becomes by degrees autonomous.
A recent example of such a machine is DeepMind’s pong-playing machine.[v] DeepMind, a company sponsored by Google, programmed a computer with the aim of getting higher and higher scores while repeatedly going through the process of playing the classic video game ‘pong’. By trial and error the machine ‘honed’ its skills at the game. The operative technology here is called reinforcement learning using recurrent neural networks (RNN). RNN is a non-linear process that mimics the human brain by reinforcing and hence weighting connections between computational units, thereby creating a continually evolving set of behaviours, i.e., input-output relations. One can imagine such technology being applied to build a machine that learns how to compose music or paint pictures. So perhaps machines could be creative. One day a machine might produce artworks as surprising and touching as those of David Hockney.
But it is one thing to produce original work in the sense of being surprising and another to produce work that touches us. Art, music and poetry are quintessentially human endeavours. That is not to suggest other intelligent life-forms could not in principle produce such art. Rather what I mean is that art, music and poetry is such to the extent that we react to it in a particular fashion. It is no good developing machines that can produce artwork that leaves us cold, that bears no relations to our world. American sculptor Carl Andre’s Equivalences – a notorious series of works consisting of specific arrangements of bricks – may leave many viewers cold, but to others these minimalist works resonate. Or again, Mozart’s Requiem touches us. How could a computer ever manage this feat?
One suggestion is to build computers that not only make artworks, but that also gauge how they are received by people. The idea is for computers to develop genetic or evolutionary programs, whereby they continually alter how they make art according to whether people find it interesting or not. That way, at least, their products would bear some relation to our world. Here I think we broach the key idea of artmaking, namely its meaningfulness.
What precisely makes a work meaningful is hard to answer. If we were to imagine time-travelling and taking Judy Chicago’s installation The Dinner Party (1974-79)[vi] to the French royal court, say, of King Louis XVI, it would be difficult to say how they would react to it. This collaborative work of feminist art is a very complex piece and the images and symbolism it presents are multilayered. Its meanings are socially constructed, and French royal society of the eighteenth century is too distant from Chicago’s world for our imagined audience to relate to it in any predictable way. Little of its content would make sense to them. The point is that much art, music and poetry gains its meaningfulness socially. It is meaningful in virtue of its content, what it says to us, namely to those who inhabit the same world as the artist and so have insight into the her perspective.
No artform is more content based in this sense than poetry. Take any piece of verse, e.g., the last lines from Philip Larkin’s poem Here: “Ends the land suddenly beyond a beach/Of shapes and shingle. Here is unfenced existence:/Facing the sun, untalkative, out of reach.” The lines evoke an imprecise image of standing facing the open sea and a sense of the sublime perhaps. To write such poetry requires being a human being. No imitation however sophisticated will do. In the case of language, even for looser poetical writing the gap between machine- and human-generated material is obvious. Here’s a sample of an ingenious attempt to generate ‘new’ Shakespeare verse by computer:
O, if you were a feeble sight, the courtesy of your law.
Your sight and several breath, will wear the the gods
With his heads, and my hands are wonder’d at the deeds,
So drop upon your lordship’s head, and your opinion
Shall be against your honour.[vii]
Apologies to Shakespeare and those familiar with his work. This text may superficially seem to speak to us, but it does not describe a world that we can enter or grasp since ultimately it describes nothing, i.e., it lacks content.
Likewise, to make The Dinner Party requires a very real set of preoccupations and sensibilities that only a human being could possess. Perhaps we could develop computers that integrate in very sophisticated ways with humans on a daily basis, e.g., go on vacations with the family to the seaside, live in an oft grimy city, eat sometimes in fastfood restaurants and at other times in Epicurean style, face discrimination, fall in love and so on. Here I cannot help but think of the futuristic robotic butlers depicted in Woody Allen’s comedy film Sleeper, though there have been many other far more plausible portrayals of human-like machines. We would be less surprised if such human-like machines were able to paint pictures that touch us in the same way as Picasso’s Guernica, compose music that moves us like Arvo Pärt’s Fratres and write a tragedy akin to Shakespeare’s King Lear. The closer to a living human being the machine is, the more meaningful its attempts at artmaking become. For a machine to make genuine art, music or poetry, we must surmise therefore, that it must be nothing but a human being. All attempts by machines to make art are by degrees approximations.
I think these considerations about the nature of artistic creativity are instructive. The ‘something’ that artists, musicians and poets add that makes their works art and not merely artifice comes from a history of our intimate interactions with one another, and by default with our ancestors. Machines can assist in the production of such work, but it would be a mistake to suppose they could partake in it. To do so requires being human. That is the ingredient of art and the creativity that ultimately defines it. What made Dürer a great artist was not his proximity with the divine. It was rather his sensitivity to the world around him as seen through human eyes. Art is not simply made by us, it is part of us. Artistic creativity is an ineluctably human process.
[i] See Plato’s dialogue Ion.
[iii] Listen to Emmy Howell’s From Darkness, Light for example.
[iv] See his ‘Computing Machinery and Intelligence’ in Mind, 59 (1950).
[v] The team in fact constructed reinforcement learning programs for several other Atari games. They modified the screen resolutions compared to the originals in order to reduce the processing demands.
[vi] This piece essentially consists of three long tables joined at their corners to produce a triangle. On the tables are laid out plates with various female-centric imagery, e.g., patterns resembling a vulva.
[vii] From the following blog site: http://karpathy.github.io/2015/05/21/rnn-effectiveness/