By Andrea Bergonzi, Data Scientist of Dataskills
The combination of Artificial Intelligence and art is now increasingly common and increasingly part of daily digital conversations, stuck at a sort of crossroads where curiosity, amazement and discontent meet. We are, in fact, in a particularly divisive moment in which AI is demonstrating a creative potential that was completely unimaginable until recently and, on the other hand, artists are keen more than ever to underline the difference between art and artifice.
But let’s take a step back and go back, at least in broad terms, to the origin of the current trend of the so-called “AI-generated art”, which has essentially developed in the last two years with the creation of software based on Artificial Intelligence that have the ability to generate images starting from text instructions (this technology is called TTI, i.e. Text To Image).
HOW ART GENERATED BY ARTIFICIAL INTELLIGENCE WORKS: THE EXAMPLE OF STABLE DIFFUSION
One of the brands considered synonymous with AI-based art today is Stability AI, the company that created the Stable Diffusion software and whose philosophy is said to be oriented towards the “democratization of Artificial Intelligence to awaken humanity’s potential”.
Stable Diffusion, we recall, is the open source TTI generator which, according to some, outperforms even formidable competitors such as DALL-E 2. Specifically, Stable Diffusion uses a LAION 5B dataset made by the LAION brand which includes about five billions of images of all kinds, obtained through rather aggressive (and indiscriminate) scraping of the web: this means that the dataset can also scrape material potentially covered by copyright and certainly made by real people, without requiring their authorization.
As far as quality is concerned, AI-generated art is achieving formidable results – especially if one considers that it is precisely the qualitative aspect the “detail” that machines are able to imitate with greater difficulty (not only in the artistic field but also in text language). The reason for this dramatic improvement is, again, to be found in the enormous amounts of data on which these software operate through the GAN machine learning system (Generative Adversarial Network, i.e. “generative adversarial networks”). If, to be clear, the material accumulated in the dataset is of excellent quality and substantial, the artistic results produced by the Artificial Intelligence software will also be considerably better.
Founded in 2020, Stability AI would today have a market value of even $1 billion. However, this was not the progenitor company of Artificial Intelligence applied to art: the seminal version of the software we know today is probably to be found in Deep Dream, the image generator launched by Google even in 2007. Clearly the technology of he era had very little in common with the current one, and the results included almost everywhere psychedelic colored dogs that seemed to have come out of – precisely – a dream. Not surprisingly, the reaction of internet users at the time was simple irony.
Nothing to do with the conflicting opinions on AI-generated Art with which the web is permeated today. To be clear, the one produced by – in addition to Stable Diffusion – Midjourney and the aforementioned DALL- E.
And although there are still some limits (for example, Midjourney does not seem to be very talented at drawing hands), a further and dizzying technological evolution of this software is already expected in the short term.
ARTIFICIAL INTELLIGENCE AND ART: WHAT ARTISTS SAY
Opinion on the art produced by AI software is divided and divisive, and sees two main and opposing factions as protagonists: that of artists in favor and that of creatives opposed to the use of this technology in the creation of works that, in fact, represent a synthesis between imagination and ingenuity.
On the one hand, therefore, we have artists such as the German Mario Klingemann, who considers the art produced by AI as a source of absolute inspiration and a fundamental means for obtaining better results. In this sense, Klingemann adopts a philosophy that sees human beings as “non-original” creatures. We just reinvent, make connections between things we’ve seen.” Conversely, machines “can instead create from scratch.”
And again, among the supporters of AI-generated Art there is also the Italian artist and philosopher Francesco d’Isa, who was the first to publish images and stories created with the Midjourney software in his own magazine L’Indiscreto. According to d’Isa, technologies like the current ones should be considered solely and exclusively as tools: a bit like cameras, which portray what the photographer’s eye sees. In the case of the TTI, the software would therefore only produce what the human being orders it to create.
“These software are not anthropomorphic androids with their own intelligence and personality, but algorithmic models based on enormous amounts of data created by humans, on which they work on a statistical basis in order to successfully respond to our requests”, wrote d’Isa last year, as reported in this interesting article by Artribune.
But is it really so? Maybe only partially. After all, cameras don’t have Big Data that reconstruct the history of art in its broadest and most extensive sense. Furthermore, AI-generated art is different from photography because it does not require the physical time of the latter: in a few seconds, it is possible to generate the “/imagine” command and obtain results that are sometimes so extraordinary as to be almost indistinguishable from a real creation Human.
The detractors of AI-generated Art are very numerous, perhaps superior to its supporters. Some of them are no longer alive: this is the case of US philosopher Hubert Dreyfus (1929-2017) who, with exceptional acumen, had already foreseen the possible evolutions of Artificial Intelligence and reiterated on several occasions the importance of it not accessing to purely human and noble abilities, such as creativity, language and conscience.
This opinion is fully shared by the Italian illustrator and cartoonist Lorenzo Ceccotti, aka LRNZ, who identifies precisely in the art generated by Artificial Intelligence software a serious threat for all graphic professions, not only from a technical and artistic point of view, but also and above all ethical. According to Ceccotti, at the heart of the problem are the datasets and their indiscriminate collection of images – which in practice takes place without permission. But that would not be the only critical point to underline: LRNZ also mentions the risk of indistinguishability between a work created by a human being compared to that produced by a machine, referring to the now numerous artistic competitions won with this kind of approach.
Finally, the issue of copyright should be examined, a complex problem that still does not seem to find a solution. Who owns the works generated with Artificial Intelligence? To the company that produced the software? To the user who physically typed the text command that led to the production of the image? Or, for an infinitesimal part, to the billions of artists who created the images included in the dataset used as “inspiration” by the machine? Since the software does not forget what it learns, and since billions of images have already been acquired, it is sterile for artists to even ask to be excluded from the data collection process.
Therefore, in this complex and controversial context, it is no surprise to discover that Ceccotti, together with other artists of the caliber of Ariel Vittori, Sio, Francesco Artibani, Paola Barbato, Manuele Fior and Elena Casagrande, is collecting sources to “support the legal costs necessary to have the way in which these companies collect their data regulated at European Community level”. And it’s no better abroad either: just last January, the artists Sarah Andersen, Kelly McKernan and Karla Ortiz filed a lawsuit against Midjourney and Stability AI claiming that both companies have infringed the rights of an incalculable number of living artists in the process of machine learning necessary for training their software.
It’s hard to imagine what the sentence will be and how these large companies will choose to regulate themselves. However, it is easy to hypothesize that the art generated by Artificial Intelligence will not disappear. It may perhaps adapt and change on the basis of regulations and limits that will certainly soon be defined, but it will remain – exactly as happened for photography.
