It is difficult to be objective about ourselves and even more so about the things that we make. Individual aesthetic choices are by nature subjective in the moment of creation. At the same time, it is necessary to be objective about the world around us, even if that world consists largely of our own constructions. Artificial Intelligence has existed as a way to explore our own brains as much as it has been to replicate them. Nowhere is this clearer than if we walk back the derivation of artistic creations. AI, as a thing itself, is subjectively invented, but the existence of the AI is objective. Art made by an AI must be explored through this lens of human experience and perspective. It is two layers removed from our subjective experience of creating, not just as individuals but as an entire species.
The centuries of discourse over the objectivity of human created art are yet to be resolved, though the need to do so will have a profound impact on the utility functions of learning systems. We are training AI agents, but they will be doing the learning. Most people relegate art to the subjective to avoid metrics that make the human perspective less valuable. The success of art as an opinion is akin to philosophical post-modernism as it relieves us of the burden of accepting a universal aesthetic. While science is often seen as hard, art appreciation is more democratic, as long as it remains subjective. Science as a testable and mathematically derived process of discovery leaves more than just a cultural chasm. It confirms that our bias in the search for truth is only acceptable through the process of the scientific method. It is yet to be seen, but as our machines begin to supplement and replace the physical and mental labor that humans are less capable of, they also aim to create art, mainly music, in ways that are independently good. AI science has therefore concluded that good art is objective, and due to that objectivity sufficiently intelligent AI should be capable of creating better art. This may very well be the most underappreciated aspect of a technology with the ambition to be better than humans. It is also impossible to recognize. Human bias blinds us from noticing beauty in aesthetic purity. The science of building an optimized AI for music is not challenging due to its subjectivity, but rather to the objectivity that the creators of the algorithms put in place. There is nothing as formal as the Scientific Method to objectively evaluate Art or the artistic practice.
An objective AI has an advantage in its assessment of art. The trajectory of art history involves a process where an original idea leads to generations of iterations which are often loose copies of the original. Whether this is Cubism, network television, or Top 40 pop music we see a regression to the mean. Great art can arise from the clutter, but it is the process of observing and continuing to create the status quo, the mediocre, and the bad that helps humans objectively recognize a groundbreaking work or new movement. The time horizon for determination of good art is jittery at best, with some quality art fading into obscurity. The creation of bad art is also the way we understand the values of our time in contrast to the quality work of our time. Extreme optimization will inevitably fail at this. The choices and creations of the machine will be objective, but not accurate in the eyes of a human agent. For art, objectively good is distinguished from subjectivity only by universal human values.
As these values evolve, Artists often search for the undiscovered genius in the past. Objectively good art represents the voices of its contemporaries. Historical context is perhaps a feature where AI can excel, but the commonality of artistic success does not evolve in a way that is consistent with human evolution, nor with technological achievements. It is a chance occurrence of complex inspiration. Detection of past and unrecognized brilliance may be achievable, but inspired creation of new human experiences by an AI is undesirable.
Art and Science need a process of learning through instinctual, and often poorly optimized choices for the results to lead to progress. The process is not objective nor are the results of the combined attempts that are being made. Furthermore, the non-instinctive ability to view those things that we have made with any objectivity can only occur with a type of amnesia to the process. The deconstruction of the failures and successes of experiments is where objective advancement lives. An AI will never go down the wrong path to see where it may lead.
If we normalize art made by people and art made by machines, we will have lost something fundamental to our being. This is a much greater threat to humanity’s purpose than the automation of other tasks that intelligent systems can and will do better. Art rests between the knowns of physics and the unknowns of consciousness, but appears easier than both. The illusion of this simplicity relegates AI art to an infancy which may never, and maybe should never, be brought to maturity. This is an unscientific view of a scientific question that everyone should consider for the same reason we explore Nature itself. We do it to be human, not to diminish it. AI is not the builder of the science itself. Good art is not the servant of nature, but its reflection, sometimes pure, often distorted. The object as art is not within an AIs grasp. This should preserve one of humanities greatest objective potentials.