*agency: art in the age of artificial intelligence

alvin tan asks the question on everybody's lips: are machines the new 'young creatives' (sorry i-d?)

 

In our modern, post-Digital Revolution world, nothing is sacred. Every conceivable field seems to be pregnable by mechanisation and computerisation, from electronic personal assistants to robot nurses and self-driving cars. While these developments are fascinating, they also raise questions about our notion of personhood, and if machines will eventually develop to have comparable intelligence, perception, morality, and consciousness as humans. One bastion that seems to delineate humans from machines is our capacity for creativity—the innate ability to invent, improvise, and innovate. Since machines are algorithmic in nature, they necessarily have regular and predictable outputs, inasmuch as we can understand the internal reasoning that drives the machine.

 

Some recent advancements in technology, however, have challenged that particular worldview. In particular, the emerging field of machine learning seems to push the boundaries of the decomposability of algorithms. How machine learning works can be briefly summarised like this: the system takes an input, runs it through an initialised algorithm, and then tries to optimise the algorithm through iteration, with the aim of producing better and better outputs. It has become highly successful in many sorts of tasks such as natural language processing and computer vision, due to the parallel advancements in terms of raw computing power and data availability.

 

The key characteristic of machine learning is that the eventual algorithm was not explicitly programmed into the machine. Rather, it simply emerged as a result of the optimisation process. As such, we do not have an understanding of the literal interpretation of the algorithm, making outputs and performance unpredictable and often unexpected. One of the early ways in which this played out in an artistic sense was in Google’s DeepDream, which involved the manipulation of an image classification algorithm to effectively extract what the machine is “seeing” at particular stages. This produced some truly psychedelic and bizarre images, as the network detected very unexpected features and amplified them.

 

Fig. 1: Image produced by DeepDream, generated purely from an input of random noise

 

 

Since then, interest in the applications of machine learning in art has continued to rise, and many other projects that explore this interaction have been developed. These include the “style transfer” network, which combines content from one image and style from another image to create a pastiche that seems like an interpolation of the original images.

 

Fig. 2: Photograph of Stanford’s Hoover Tower in the style of van Gogh’s Starry Night

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

More recently, Google has launched Project Magenta, with the aim of using machine learning for music and art generation. This is significant in two fronts: firstly, the undertaking of such research by a technological giant means that many more resources (in terms of capital, time, and human ingenuity) will be put into this field; secondly, the project’s focus on generation seeks to involve a creative aspect that exceeds merely manipulating existing images. Development in this area means that it may be possible to witness computers creating artwork ab nihilo in under a decade—a prospect that would have seemed absurd a decade ago.

 

What impact would this have on art? If computers too can create art, would this lead to the devaluation of artistry as a discipline? It seems that ascribing ‘creativity’ to computers would diminish yet another aspect of humans’ personhood, and perhaps lead to an upheaval in the industry with the introduction of automation.

 

Against this backdrop of rapid development in computing science, it is perhaps worth examining our notion of creativity more carefully. In doing so, we notice that creative output also incorporates some sort of normative judgement on the ‘goodness’ of such output, which can be described by the idea of aesthetics. Output that is arbitrary or atypical does not necessarily imply that it is creative—rather, creative output also involves some concept of beauty or elegance (even if its manifestation is the subversion of norms regarding these concepts). Furthermore, aesthetics is something ascribed rather than inherent: we regard the images produced by various machines as being interesting not because of intrinsic properties of the images, but because of our perspectives on them. Machines have no way of knowing if something is aesthetic or not; at best, they can attempt to approximate what humans might perceive as aesthetic, and then again, they would be applying external criteria rather than having an internal notion of aesthetics.

  • Spotify
  • Instagram
  • Facebook