CCB Magazine

The debate about Artificial Intelligence and learning algo­rithms is becoming ever more heated: What can machines do that people can’t? Can machines even be creative? Will they soon be the better artists? In fact, machine­ learning processes are already intervening in all areas of cultural creation: Machines automate film productions, they beat us at the board game Go or compose sophisticated music. Reason enough to take a closer look at the creative­-artistic achievements of intelligent systems.

BY DR. ANDREAS SUDMANN (media scientist and co-editor of the recently published book “Machine Learning. Medien, Infrastrukturen und Technologien der Künstlichen Intelligenz”, [“Machine Learning. Media, infrastructures and technologies of Artificial Intelligence”],  transcript Verlag)

1950. In an essay for the journal Mind, no less a brain than Alan Turing presents a famous thought experiment. In the following decades, it will decisively shape the international discussion about the potentials and limitations of Artificial Intelligence (AI). Today we know the experiment by the term “Turing Test”. At the time, the British mathematician and computer scientist wondered if machines are able to think. With the Turing Test he changed the question, asking: Can machines appear to be intelligent? More precisely, can a machine, with­out revealing itself as such, successfully imitate human communications behaviour? If a machine were capable of doing so, it would also be legitimate to con­ceive of it as thinking according to Turing. 

Let's go to the present: Contrary to other theses, no ma­chine has yet passed the Turing Test. Nonetheless, AI research has been able to achieve quite a few other successes in recent years. For example, the spectacular victory of Alpha­Go in the board game Go, in which a com­puter program played the board game by itself and won against a human world champion, as well as ad­vances in the field of language translations, image recognition, and in the field of self­driving cars, etc. Moreover, Google’s telephone assistant Duplex has recently impres­sively demonstrated that AI systems can, at least to some extent, now communicate with individuals without the latter noticing it – in contrast to the Turing Test, where the asses­sor is still aware that the interlocutor is a machine. In this respect, it comes as no surprise that the actually not so new question in the current AI discussion is more frequently being raised: Can machines be creative? Can machines even produce art?

One thing’s for sure: Smart ma­chines are already participating in creative and artistic pro­cesses. For a long time now, AI technologies have been used to make films, compose music or create pictures. What’s more, in many cases AI systems are not merely used as a means, but act as the actual, essential agents of creative artistic expres­sion. So: Will machines be the better artists soon? 

Smart machines already participate in creative and artistic processes. So, are machines soon the better artists? No. Machines cannot see themselves as creative-artistic entities. And artificial intelligences have so far had no reflexivity and no consciousness

The answer is quite simply no. For one thing, machines aren’t able to conceive of themselves as creative-­artistic entities. And they won’t be able to do that at any point in the near fu­ture either. Also, they have no idea what art or creativity even is. Machines can play Go with us and compose music, but they have no reflexivity, no consciousness. And you can’t really teach machines anything like common sense knowl­edge yet, even though there has been a lot of progress re­cently. For example, TwentyBN, a Berlin and Toronto­-based AI company, has recently trained artificial neural networks to detect actions and gestures, such as whether a person is sitting or standing, moving toward or away from the camera whether she hides her eyes with her hands or does things like drinking from a cup. 

This achievement is certainly at least as significant as AlphaGo's victory. For a long time, hardly anyone in the general public took any notice of it. No wonder, when a computer identifies everyday actions and gestures by means of a simple RGB camera, it is prima facie not as exciting as, say, a duel between man and machine in one of the oldest games in the world. Above all, such abilities of a machine seemingly have nothing to do with creativity.

But this is a mistake. If computers are ever to perform creative and artistic tasks close to human beings, they will have to be able to perceive their environment and 'understand' this sensory input. At the moment, they are hardly able to do this. But that can change quickly - as recent developments in the field of AI-based surveillance technologies in China illustrate. Besides that, one should not forget: What can be considered art and creativity is not only in the eye of the beholder. The judgment depends on the specific materiality and form of the artifact, the creative process that produces art or enables the creative performance, and the silent or not-so-silent dialogue between the work and the world. Fifty years ago, the philosopher Theodor W. Adorno had already emphasized the enigmatic character of art, among other things, that as art it necessarily contains something incomprehensible, not least in view of its utopian potential. This means, among other things, that art cannot be positively determined.

The fact that machines supposedly produce art currently raises a completely different question, namely that of who is still actually creating value when machines or AI algorithms themselves make art. Recently, this conflict was exemplified by the copyright dispute over a portrait entitled “Edmond De Belamy”. Created with the help of a learning algorithm, “Ed­mond De Belamy” is probably the first work of its kind to have been auctioned at the renowned auction house Christie’s – for the not insignificant sum of over $400,000. The portrait shows a smeared “painted“ male figure that looks as though being from the 18th century. A Parisian group of artists called “Obvious” is responsible for the picture. According to their own website, the collective has taken up the cause of wanting to “explain and democratise" AI with art. 

The current copyright dispute over the portrait "Edmund De Belamy" raises the question: Who will still be a value-adder in the future when machines or AI algorithms make art themselves? Are these still the people or the machines?

However, this claim doesn’t quite fit with the process the por­trait was created. For the Parisian artists had used a learn­ing algorithm, which actually was developed by a 19­ year ­old man named Robbie Barrat from West Virginia who put the code as open source online. More precisely one would have to say: Barrat developed it further. Because the tech­nology is a so ­called Generative Adversarial Network (GAN), which was originally designed by the well­ known scientist Ian Goodfellow. In any case, the Paris artist collective asked Barrat whether it could use “its” algorithm for the cause of AI art democratisation. He agreed and even helped the group with the technical implementation of their project. However, when the portrait produced with the neural network was finally auctioned at Christie’s, the Barrat was annoyed. On Twitter, he wrote on 25 Oc­tober 2018, “I’m crazy for thinking that they’re just using my network and are selling the results?” 

Since then, it’s been debated who is actually the creator and author of the image. In a contribution to the Süddeutsche Zeitung, Bernd Graff put forward the thesis that not the por­trait, but the algorithm itself must be considered art. A humanist might sarcastically comment that, on the bright side, at least these algorithms are still being devel­oped by humans. As a media scientist, however, a completely different intervention makes sense to me: It consists above all of abandoning the classical dichotomy of man ver­sus machine and taking greater account of the fact that cre­ative-­artistic practices are in principle based on processes in which the power to act is distributed between non­human and human spheres. And so it is, ultimately, when it comes to the case of “art” generated by learning algorithms. 

There is no such thing as AI, which is similar to humans or superior to them in terms of creativity and art. Not yet

What does all this mean?

As never before in the history of civilisation, computers today can give the impression or sim­ulate the impression that they could be creative on their own or even produce art. In fact, they are only capable of doing so because humans and algorithms can achieve this simulation achievement together. The AI that is similar to humans or superior to them in terms of creativity and art does not exist. Not yet. The autonomous machine à la Terminator is not (yet) a reality - whether with a gun or a paintbrush.

Here you can find further information about Andreas Sudmann

"Machine Learning. Medien, Infrastrukturen und Technologien der Künstlichen Intelligenz" has just been published by transcript.