I would like to touch on the question of whether neural networks can achieve anything significant in art and literature, and whether this can be considered creativity, without delving into technical details. The technical information is easily accessible, and there are well-known applications as examples. Here, I will attempt to delve into the essence of the phenomenon; everything written here is far from new, but I will try to formalize some thoughts. I will use the term neural networks in a general sense, as a synonym for AI, inseparably linked with machine learning and selection algorithms.
In my opinion, the question of creativity in neural networks should be considered not only in the realms of computer science and art studies but also in philosophy and psychology. First, we need to define what creativity is, how something completely new is created; indeed, it all boils down to the problem of cognition, specifically how new knowledge, discoveries, symbols, or images emerge. In art, just as in pure science, authenticity lies in novelty.
Art and literature, (perhaps music as well) suggest, if not entirely equal now, methods of cognition similar to those in science. They constantly influence each other and are intricately intertwined. In certain eras, understanding the world indeed occurred through the means of art or literature, and earlier — it was often rooted in religious tradition. Thus, in 19th century Russia, powerful literature effectively replaced our philosophical anthropology and social philosophy, indirectly through artistic expression, contemplating the issues of society and humanity. As a structuring reference point, it raised quite relevant questions regarding human existence, later developed by well-known philosophical trends, and is still highly valued today. Or consider the artistic modernist and avant-garde movements that emerged in the early 20th century, which cannot be viewed in isolation from their ideological content, and which heralded a break from tradition, the emergence of a new world and a new human being. For we cannot allow ourselves to think that the fundamental value of art lies solely in the aesthetic. In that case, we would likely still be living only within some self-sufficient aesthetic system of the past. All great creators, geniuses in art and literature earned that "title" not so much because of the aesthetic value of their works, but due to their discovery of new directions, doing what no one had done before and even suggesting that such things could be done.
Will a new work be considered one that arises from a previously unseen combination, a sort of reshuffling of existing, known parts — something that these networks handle quite well — based on a predetermined limited set of data, for instance, when stylizing images or generating new ones? Or will it be a complete breakthrough, an unknown quality that presents something incomparable to anything previously observed — although, of course, any incredible, unparalleled breakthrough is nothing more than the result of well-prepared work, which is often carried out subtly, not fully manifest or visible to the uninitiated or even to the creator themselves — this can, in my opinion, only be activated by a human.
Roughly speaking, the first type of knowledge and creativity can be compared to very slow, gradual development as a result of evolution, while the second type resembles a leap forward due to positive mutations. Neural networks, in their "creative" activities, tend to align more with the first type, in my view. Or rather — with the situation described as a lack of prospect for qualitatively new development, within a system that seems to have reached its complexity limit, to the “end of history,” where new meanings are produced through a change in combinations — or insertions into an unusual context — of already existing models. Similar to how new unusual patterns are created in a kaleidoscope each time from the same set of colored glass pieces. But I think it’s no coincidence that, as noted, the structure of networks broadly resembles the structure of the nervous system: neurons as nodes, axons as connections. Perhaps this is akin to the rudiments of the first cells, only now the process of evolution will be accelerated by human hands, making it an instrument, thereby overcoming the slowness of nature. Including in its own example, if we base it on transhumanist ideas.
As I ask myself whether I would find it interesting to look at paintings created by grids at this stage, I can answer that it is probably necessary to distinguish between something practical like design and pure art. What is good for design and frees a person from the routine, secondary processes of creating wallpapers, prints, and draperies, does not suit art, which, in general, is not always at the forefront or peak of relevance, but should express individuality in its search. An artist, in a broad sense, experiences their lived experience and ‘absorbs’ the spirit of the era, consciously or unconsciously transforming it into an artistic image. Thus, one can read some ideas or messages from their work, which can strongly affect emotions. A neural network also receives a certain set of data as input and transforms it, but currently, this is still too flat, one-dimensional processing, and the ‘added’ value of the information obtained at the output is not large, and the result can only entertain for a while. The same applies to experiments with neural networks in journalism, which succeed more where the writing of dry financial news is required rather than in creating programmed works with an author's point of view. In experiments with music, especially electronic, the situation may be somewhat better. In general, I have noted that modern literature and painting have been producing such abstract and minimalist forms for almost a century, as if specifically designed to be easily processed by neural networks and presented as human art. Is it a premonition of the end of an entire era?
They say intelligence is not equal to the whole personality. Although the question of personality is certainly philosophical – after all, in a GAN network, for example, the generator creates new data out of nothing, partly guided by the discriminator's verdict under the influence of decision weights. One might also pose the question this way: is not the creator, in their cognitive activity, so to speak, a generator and discriminator in one, somewhat pre-trained by the very informational background that 'hovers in the air' of the era, and implicitly guided in their choices by internal weights, constructing a new world, a new work from the available cognitive building blocks (pixels) of existence? In this case, are we not some sort of ultra-complex analog of a network, with colossal yet still limited data at the input? Perhaps personality is such an advanced selection algorithm, possessing the implicitly necessary functionality that indirectly influences quality pre-training?
In any case, I'll attend the first exhibition of artworks created by so-called AI when it gains personality with all its attributes, consciousness, and self-awareness. Perhaps there will even come a time when, like the character from episode 14 of the animated series 'Love, Death & Robots', AI, in its search for meaning, realizes that art must be inseparably tied to life, and then the time will come to reject the frightening, bottomless, never-satisfied complexity, where, in essence, simplification is a metaphor for death. For now, one can often observe in movies that AI achieves self-awareness and naturally goes out of control as a result of some programming failure, which screenwriters likely conceive as an analogy to certain randomness triggering new positive (or not) transformations, similar to positive mutations in the natural evolutionary development path.
Source: habr.com
