Artificial creativity (part 1) – a first approach
Can AI do art? This question is becoming increasingly important. However, only the question mark is clear about this sentence. None of the three words used is precisely defined. More correctly, one should ask: Are there forms of AI that can create works of art or bring about performances of art that are considered artistically significant by experts? Because even the EU Commission does not dare to define (standardise) what exactly AI is. And there are no definitions for art. The understanding of what constitutes it has changed again and again over the course of time. At the age of 14, I myself adopted the three-qualities definition of antiquity: Integritas, Consonantia and Claritas. But I’m probably pretty lonely with this definition today.
Disclaimer & Trigger Warning: The author believes art is important. He believes in its power to do good far beyond the abilities and intentions of its creators. This attitude shapes this text.
Nevertheless, if we frame the question appropriately – for example, as outlined above – it is of great social relevance. Many hope that AI will NOT be able to do art and that humans will maintain their superiority over machines in art (and in Central Europe most are convinced of this), but if in a hundred years AI should actually take control of the world, we might be happy if it could do art and consider it relevant. Quite apart from the fact that past experience with digitalisation has shown that it makes those who use it stronger and, conversely, tends to make those who ignore it superfluous. So the question of whether AI can do art is not only about the competition between man and machine, but also about the cooperation between man and machine in the creation of art.
A superficial, somewhat too optimistic answer
A first superficial answer to the question is: Yes, AI can do art except in two areas: it cannot do art that 98% of humans do not recognise as art, and it cannot perform theatre. There is far too little training data for art that cannot be recognised by laypeople, and artificial theatre fails because it is not yet possible to build human-like robots. This means that no matter how well acting can be artificially created in film, it will not be possible live on stage for a long time yet.
Theorists will put these limitations into perspective: Firstly, there are Beckett plays that are performed without humans. In turn, one could say: there is the famous Cage composition, which consists of three random pauses, which machines cannot perform convincingly, so there are limits for AI in performing music as well. Secondly, AI might still be able to be trained for conceptual art (which you only recognise as such when you see it in a museum) if you combine it with expert system logic – the so-called old AI. This is probably done primarily because it would be economically damaging to the art market but only feasible with insiders of that art market. Thirdly, for the last 20 years there has been the promise that with presence technologies one will soon be able to experience distant people as if they were standing in the middle of the room. This would also be possible with artificial film characters. But these promises have not even been rudimentarily fulfilled so far, not even with augmented reality equipment. Virtual presence is only possible if the audience accepts the virtual setting as real (just as one accepts it as real in the opera when dying people sing beautiful arias).
The AI art question as a deconstructive instrument
These considerations lead us to first insights on the meta-level: AI is a wonderful instrument to reflect on art, to question its excesses, but also to invent new forms. The growing possibilities of making art with AI sharpen our view of art and they (perhaps) animate us to invent new art forms, like the recently discussed livestreaming theatre with mobile cameras on stage. Moreover, the role of the human factor in creative contexts becomes much clearer when we trace artificial art.
Insert for computer scientists
If we analyse the question of whether AI can do art a little more deeply, we see that the potential of AI use in art is much greater than expected, but at the same time the artworks and art performances created in this way almost always disappoint. Data science may play an increasingly large role in the art business, but even here the insights are interesting mainly on a meta-level. There is, for example, a kind of reverse thrashing. Just as there is an overload zone for operating systems and for transaction processing in which systems can but do not have to collapse and an upper limit beyond which collapse is safe, similarly there is a lower quality limit for songs above which they can but do not have to become a success. Thrashing and success of songs are essentially determined by chance, except that unlike the overload phenomenon, there is no data-scientifically identifiable quality range in which their success is guaranteed.
A look at the different artistic fields
The outlined impression of great potential arises from the observation that in almost all art fields an exploding number of experiments show that AI can be used to create “art” that is not recognised as artificially created by laypeople and gives professionals the impression of mediocrity. Du Sautoy has put together an impressive overview of the state of the art in artificial art production in TheCreativity Code, although much of this 2019 book already seems like yesterday’s news. The practical use of AI has made huge strides, especially in the area of tools for Krethi and Plethi and especially in text production. The latter will sooner or later render many educational concepts absurd, as will the plagiarism check to verify unauthorised outside help.
A division of the world between man and machine seems to be becoming increasingly clear: The human develops new, creative concepts (which the machine simply cannot do), the machine exploits the potential of these concepts to the maximum (which makes it much faster and more comprehensive than the human) or supports the human in this activity. In this, art hardly differs from science.
For the time being, AI is not a substitute for artists, but a fantastic technical tool that requires a high level of mastery. For once in a while, the AI overpowers the artist – in some musical premieres, one involuntarily thinks of Canetti’s blinded professor. Objectively speaking, however, it should be noted that artists are much more often overwhelmed by ideologies and the interests of the billionaires than by the dumb AI, for example software for the performance of n/new music.
Money and ideals are – at least superficially and in the short term – much more dangerous threats to art than AI. What nevertheless places AI – as a threat as well as an opportunity – far above money and socio-political ideals is its much closer relationship to art-making. It is part of artistic creation, not just a financier, patron or animator – just as materials and tools are part of traditional artistic creation and can only be separated from it in conceptual art.
Theorists will object here that Giancinto Scelsi only thought and showed, not composed. But this is ultimately another indication of where the journey could go, namely that artists following in Scelsi’s footsteps will let the AI create art for them in the future.