Art generated by neural networks.
It seems that only a couple of years ago neural networks were of interest only to geeks and tech enthusiasts. But already today this technology is breaking through into the mainstream: neural networks are becoming a tool for work, entertainment, and sometimes they themselves perform part of human work. This worries specialists in the creative industry in particular, because neural networks have already learned to write texts and create designs.
Cats in space; portrait of Reagan in the style of Bacon; Mickey Mouse in the image of a real mouse: in the techno trends of 2022-23 a new powerful the phenomenon is art generated by neural networks. Midjourney and DALL-E 2 neurons received the most publicity, the images created thanks to him were on the covers of the popular magazines The Economist and Cosmopolitan. In Ukrainian segment of the Internet, many illustrations on the topic of Russian military aggression, also created using neural networks
Despite the predictions of tech enthusiasts and connoisseurs of generative design, neural networks are currently not functioning in the market creative work as an independent unit. However, all this hubbub had a tangible effect: last year, neural networks finally took their toll place among design tools. A defining role in the creative industry of the future has already been predicted for metauniverses and NFT, so let’s try to soberly assess the prospects for the further development of neural networks. At the same time, these prospects are impressive.
Although neural networks as a tool for creative work have gained popularity relatively recently, there are already things in which neurons have surpassed not only a person, but also any available technology. For example, chess is DeepMind’s AlphaZero neural network, which owns Google has been dominating the best engines in the world for several years now (we are not talking about a person, and I’m sorry). So in the field of computing opportunities, they have already proven their superiority. But how to explain their success in creativity?
A basic logical contradiction arises: neural networks are not capable of doing creative work, at least not yet. We tend to confuse imitation of creativity with creativity - the difference between these things can be debatable except when the author is a person. However, the neural network is not capable of self-expression, because it is not a full-fledged artificial intelligence.
Full-fledged artificial intelligence (strong artificial intelligence) is a probable scientific concept, according to which a computer must be able to perform abstract thought activity similar to human. Such an AI tries to understand itself, surrounding reality and he succeeds.
Currently, artificial intelligence is in a state where it can only compile data, systematize it and output results thanks to pre-created algorithms. It is not capable of meta-processing of data, does not perform really deep analysis. That is precisely why any talk about the self-sufficient characteristics of neural networks — whether they are good or bad — is still pointless.
Let’s focus on technologies that allow creating graphics according to the text-to-image scheme, so we will bypass other phenomena of machine learning like GPT chat. Graphical use of neural networks occurs mainly in two modes:
- Mimicry of a given style (say, some famous artist);
- An attempt to generate something completely new (for this, the neural network will still mix and distort the data it already has).
Neo-Luddites and optimists
At any time, there have been people who are worried about the unstoppable development of technology, and those who, on the contrary, encourage progress. It is not necessarily about two enemy camps. Attitudes towards neural networks also divided colleagues in the design department.
“Neither art nor wisdom can be achieved if they are not learned,” believed Democritus. And what do they do all the time neural networks? - That’s right, they are studying. Every year they will become more and more perfect, but the art created by AI is still the same there will be a lack of author’s research, creative thinking and understanding of artistic techniques. An important question arises: is it important is it in our postmodern world if almost no one will notice the difference?
Delving into this topic raises more questions than there are currently answers. If we try to evaluate the quality of the neural network, what criteria will we apply? If these are the same criteria as those by which we evaluate a person’s work, then what are we even evaluating — the quality of the neural network’s performance or the quality of the few lines of code that a human entered to get the desired result?
Neo-Luddites fear that individual creative contributions will be devalued precisely because entering a query into a neural network can anybody However, reducing the function of the author to writing a few prompts only simplifies the “threshold of entry” into a certain art, and does not remove the role of a person entirely. (A similar debate was heated when electronic music was just emerging). After all, automated means of work have reigned in art for a long time - and still the last word in his creation has only the author.
Generative design is a field at the intersection of design and programming, in which a person transfers part of his work to a computer algorithm. Ukrainian creators such as Twid studio and Banda agency are experimenting with generative design techniques.
In generative design, a person and a computer work “in the same team”, and specialists in this direction calmly perceive neural networks as one of the work tools. At the same time, no program is able to function without human participation.
In truth, generative design is not a new thing. Back in the 1960s, Bell Labs began using computers for creating graphics and animations with repeating digital patterns. However, it was in 2022 that the technology came to an end a new level thanks to machine learning of neural networks.
The company DeepMind Technologies Limited, which was acquired by Google a few years ago, is also engaged in generative approaches in digital design. We mentioned their other invention, the AlphaZero chess phenomenon, earlier. A full-fledged analogue to DALL-E and Google’s Midjourney we’re still waiting for.
It is also interesting that back in 2016, Google launched the Quick, Draw! project, an online game in which people create simple sketches objects by description, and the neural network analyzes the received data and learns. Google isn’t the only one showing a keen interest in the study of the possibilities of neural networks: in particular, Adobe is engaged in similar projects in the field of digital design; Nvidia Canvas turns simple sketches into detailed photorealistic landscapes; OpenAi, an AI research laboratory, received billion investment from Microsoft.
If you perceive design as a craft, then the creative community is still somehow able to take kindly to the fact that neural networks take over the monotonous work of a person. Like, there will be less routine, so there will be more time for ideas. Instead, it is big it is the art created by artificial intelligence that is surprising. Neurons can write songs (although Nick Cave didn’t like it), draw pictures, etc.
Artist Refik Anadol, as part of the experimental project “Unsupervised”, threw data from over 200 years of exhibitions into a neuron at MoMA (Museum of Modern Art in New York) to generate one illustration that currently hangs in the lobby of the museum; exhibition of artificially generated art has already taken place at the Venice Biennale; Jason Allen’s painting Theater D’opéra Spatial, created thanks to AI, won an art competition in the USA. Of course, there are many more such examples.
Art created by artificial intelligence has taken the creative community by surprise. Although the technology was very hyped, but it has not yet become complete. While everyone is already tired of NFT, and the corona crisis has subsided, neural networks are proud took a vacant place in the center of attention of creatives.
Intellectual property in the 21st century is a fragile thing, and the omnivorousness of neural networks is uncompromising. To study, they absorb even copyrighted data. At the same time, neural networks are developing incredibly fast — DALL-E (the name, by the way, combines the names of the artist Dali and the Pixar robot character Wall-E) has been in the public domain for less than a year, and already has become almost the most discussed technological phenomenon in the world.
There was a need to develop a new regulation of intellectual property. Legal chaos during the first months of popularity neural networks have been compared to the early era of music streaming. The fact is that the Napster platform is in the early zeroes was the first to introduce a proto-streaming music service (albeit a pirated one), but soon the company disappeared due to court cases battles with disgruntled musicians, including Metallica.
Changes in copyright legislation are also taking place in our country. Within the framework of adaptation of Ukrainian legal norms, respectively to the pan-European regulation, the Verkhovna Rada adopted a law according to which a legal definition appears for the first time in Ukraine originality The changes will also affect how the authorship of works generated by artificial intelligence will be established.
Neural networks process, learn, compile millions of images and illustrations to create something on their own. Problem is not that people use neural networks for entertainment. But what about the commercial use of such works?
Let’s say you bought a magazine, and on the cover, created thanks to a neural network, you recognized elements of your illustration. Is it possible claim AI stole your job? The creators of Midjourney, by the way, have already faced a class action lawsuit by artists who believe that that neuronka stole their intellectual property.
The problem is hotly debated, and different players in the technology market are solving it in different ways. For example, the largest photo stock Getty Images banned the use of artificially generated images precisely because of the fear of lawsuits. Simultaneously Shutterstock has taken the opposite approach — they encourage the use of generated images, and authors whose work will “eat” the neural network, they promise to compensate for their contribution to the development of the technology.
The world of the wild future
Many people will lose their jobs, but those who are the first to master a new tool will get rich. We expect a burst of creativity, but devaluation of creativity is also likely. As usual, no one can say for sure what the development of a new breakthrough will lead to technologies, but something completely new will certainly appear in creative work. Instead, something like routine can disappear.
“Neither art nor wisdom can be achieved if they are not learned,” believed Democritus. So, we only have to do to learn — in particular, how to use neural networks.