Artificial intelligence (AI) has matured and developed rapidly since 2022, bringing new opportunities as well as challenges to various aspects of society. It has forced people to view many things in a new light, including artistic creations such as texts, images, and sounds. These are being redefined or imbued with new meanings, challenging two dimensions primarily: appreciating and understanding.

The significance of the human creator is diminished or even removed in the creative process, raising questions about whether such creations can still be considered art. Roland Barthes’ announcement of “The Death of the Author” has thus become a reality, while Walter Benjamin’s concept of “Aura” is also under threat. In light of this, this article will first categorize different kinds of AI-based creations according to the degree of human input, and then analyze how AI’s involvement affects the appreciating and understanding of artworks within each category. This will help clarify the potential for new receptive and interpretive possibilities arising from these changes.

1. The three categories of AI-based creation

Since the launch of ChatGPT in 2022, artificial intelligence (AI) has matured and developed rapidly. This progress has had profound impacts on artworks across various mediums, including text, image, video, and sound. The influence extends beyond creators to include viewers, listeners, readers etc. as well. To understand the revolutionary effects of this new technology on art better, it’s important to categorize AI-based art creations according to the degree of human input at first. This kind of categorization, which is also an important basis for determining copyright in current legal practice (Perry & Margoni, 2010), addresses the question of the leading role, dividing the contributions of humans into three categories, ranging from significant to minimal: AI-assisted, AI-generated, and AI-originated (Long, 2021, 143).

The first category is driven by human creativity primarily, with AI playing a supporting role, like Adobe Illustrator and Photoshop. In the second category, human input diminishes; and while AI still plays a supporting role, its degree of involvement increases, making it difficult for the human creator to control the final outcome precisely, like in the case of AIVA and DALL-E. The third category is dominated by AI, which acts as an independent creator, with the process being almost entirely automated. In this creative process, human input is minimal and doesn’t play a decisive role, as Ai-Da has shown.

For the viewers, listeners and readers of artworks, the categorization based on the degree of human input in the creative process reflects the different experiences they have when understanding and appreciating artistic creations across the categories. The following sections will explore this subject from their perspective, focusing on the potential receptive and interpretive challenges they may have when engaging with AI-based creations.

2. The challenge of appreciating artworks

When appreciating an artwork, the experience is associated with its “aura”. Walter Benjamin introduced this concept in The Work of Art in the Age of Mechanical Reproduction, describing the “aura” as a unique feature inherent to authentic artworks. It can be compared to the holy light in the religious context, which inspires reverence and a sense of sacredness that deters desecration. An authentic artwork possesses an “aura”, because its existence is restricted by time and space. Reproductions, however, break these restrictions by losing the uniqueness and related characteristics at the same time. This makes them appear ordinary, which in turn affects the experience of the viewers, listeners and readers. As Benjamin wrote: “Even the most perfect reproduction of a work of art is lacking in one element: its presence in time and space, its unique existence at the place where it happens to be. This unique existence of the work of art determined the history to which it was subject throughout the time of its existence.” (Benjamin, 1969, 220)

However, the connection between “aura” and the authenticity of the artwork has faced challenges with the coming of the digital age, due to the various and broader impacts brought by digital reproduction technology (Chen, 2007, 229). The differences between authentic artworks and reproductions have been eliminated; and while cutting off the connection between “aura” and authenticity, a new connection between “aura” and the artistic expression of contents has been established. Now, with the continuous advancement of AI technology, which is also applied to artistic creation, the “aura” of artworks is facing new challenges again. The issues involved are no longer limited to the reproduction, presentation and sharing of artworks, as changes in the “identity of the creator” and the “method of creation” have expanded their scope. Instead, they seriously challenge the definition of art and its value, raising the question: Can works created within the AI-based creative framework be regarded as art?

An answer to this question can be found in the findings of an American research project, which consisted of two parts. The first part – conducted by Porter & Machery (2024) – aimed to examine whether non-expert poetry readers could distinguish between AI-created poems and those written by renowned human poets. The researchers used ChatGPT 3.5 to generate poems mimicking the styles of 10 well-known poets, and 1,634 participants were asked to determine whether the poems were created by AI or humans. The results showed that the participants’ accuracy was slightly below random chance (46.6%), and they were more likely to judge poems created by AI as poems written by humans mistakenly. This suggests that the poetry created by AI has reached a level of verisimilitude that is able to deceive readers, even to the point where non-experts perceive it as “more human than human” (Porter & Machery, 2024).

However, even if AI is capable of creating works that are close to or even surpass human creations in terms of style and quality, it doesn’t mean that these works will be accepted as art, as can be seen in the second part of the research project. It focused on evaluating the quality of poems created by AI and humans, with 696 participants tasked with rating both types of works (Porter & Machery, 2024). The participants were assigned to one of three conditions randomly: “informed it’s written by a human”, “informed it’s created by AI” and “uninformed about the author”. They assessed the poems across 14 dimensions, including overall quality, rhythm, imagery, and sound. The findings revealed that when participants were informed that the poem was created by AI, their ratings decreased significantly. However, the poems created by AI actually received higher ratings than the poems written by humans in 13 dimensions, with the only exception being originality, where no significant difference was found. This shows that “participants are biased against AI authorship” (Porter & Machery, 2024).

The findings of this research largely align with the results of another study conducted in Spain, which focused on the capabilities of small language models (SLMs) in creative writing, particularly in comparison to human writers and large language models (LLMs). The research revealed that “while larger models produce more consistent and coherent text, they also tend to follow more predictable, formulaic patterns” (Marco, Rello and Gonzalo, 2024). On the other hand, compared to human writers, SLMs were “22% more readable, 17% more understandable, 23% more relevant to the title, 11% more informative, and 18% more attractive” (Marco et al., 2024), with only insignificant lower scores in creativity. Furthermore, the researchers conducted an additional experiment with 68 participants, who were tasked with reading and evaluating the texts under the condition that they only knew the texts could have been created by either humans or AI. The results also indicated negative biases against works created by AI and against AI authorship, which impacted readers’ experiences and influenced their assessments of AI-based creations negatively (Marco et al., 2024).

Obviously, in the discussion of the artistic nature of AI creations and whether they can be accepted as art, quality is neither the only nor the most important thing. Instead, people are more concerned about the identity of the creator. This factor, considering the degree of human input, has little to no impact on the first category of creations among AI-assisted, AI-generated, and AI-originated works. For example, many international photography competitions have already accepted specific computer post-processing under strict regulations, rather than insisting on photos output from cameras directly. This demonstrates that minimal technological intervention is not regarded as detrimental to the traditional artistry of creative works. As International Photography Awards (2023) said: “There are many existing categories in which photographers can submit images that have been digitally edited or created using special effects with or without a camera”, while admitting that “the topic of AI-created images has become increasingly controversial amongst photographers, with many voicing concerns over how AI will impact the photography industry.” 

These concerns reflect a sense of unease about the potential disruption of traditional values and the uncertainty of how to engage with works created by AI or with AI authorship. The resulting biases and resistance target AI-generated and AI-originated works with less human input primarily, hindering the recognition of their artistic merit and value, while emphasizing the enduring and unassailable status of “human-made artworks” in contemporary culture at the same time.

“Théâtre D’opéra Spatial” by Jason M. Allen, generated using Midjourney, is in the public domain via Wikimedia Commons.

Although the terms understanding and interpreting are often used interchangeably when referring to artworks, they still have subtle differences. Understanding is a spontaneous and inevitable act, as it is impossible for people to encounter an artwork without trying to understand it. On the other hand, interpreting is an active and conscious act (Mauz & Tietz, 2020, 11), which can be regarded as a form of active understanding. Despite the differences, both acts are linked to authors’ intent closely.

As early as the 1960s, Roland Barthes argued in The Death of the Author that writers are mere vassals to the ideologies shaped by historical developments. Based on Ferdinand de Saussure’s theory, he emphasized that the interpretation of a text should only focus on the language. Two years later, Michel Foucault proposed in The Archaeology of Knowledge that every individual move within a discursive field shaped by the development of history and society. Under these circumstances, people’s unique conditions shape their perspectives, interests and engagement with specific topics and issues, leading to the formation of particular ideas. In other words, everyone, including artistic creators, is the product of the interaction between personal conditions and environmental influences. As a result, people’s thoughts and expressions are inevitably confined within specific frameworks, making complete autonomy unattainable.

In contemporary times, the importance of the author’s intent in interpreting and understanding artworks has not been dismissed entirely and remains a subject of debate. Normally, understanding an artwork involves the questions “what”, “why”, “for what” and “how” (Mauz & Tietz, 2020, 23), while interpreting is an endless process and involves a sixfold relations: the interpreter, the interpretandum, the interpretant, the interpretational perspective, the interpretans, and the recipient (Mauz & Tietz, 2020, 76). The interpreter is the one performing the interpretation, while the interpretandum is what needs to be understood. Additionally, the interpretant is the lens or medium through which the interpretation is made, the interpretational perspective is the viewpoint that guides the interpretation, the interpretans is the result or meaning derived from the interpretation, and the recipient is the person or group with whom the interpretation is shared. There is no doubt that understanding and interpreting associate with each other, or it can even be said that understanding is the goal of interpreting, and interpreting is the means to achieve understanding (Dalferth, 2020, 57). However, before trying to interpret, one must have a certain degree of understanding of the artwork, which can be deepened through further analyses, explanations and clarifications gradually. Nevertheless, interpreting is a subjective process and may lead to misunderstanding or distortion, making the correct approach to achieving true understanding crucial.

Normally, true understanding can be divided into two types. The first type requires grasping the author’s intent accurately, and the second type is more flexible and doesn’t require the understanding to align with the author’s intent completely. Instead, it demands that the understanding meet certain external standards, such as textual coherence, the author’s reasonableness, and comprehensibility (Reinmuth, 2020, 90). However, it’s not easy to capture the author’s intent. For this reason, modest actual intentionalism, which “is a major position on interpretation in contemporary analytic aesthetics” (Lin, 2020, 165), provides a middle way and a moderate method: “Work-meaning is determined by the author’s intention when such intention succeeds, or, when it fails, by convention (and context).” (Lin, 2020, 166) 

On this basis, the capture of the author’s intent can be realized under three conditions: compatibility, meshing, and audience’s uptake (Lin, 2020, 167-169). All three emphasize the author’s clear expression of creative content and the people’s recognition of the author’s intent through that content. In terms of this, not all kinds of AI-based creations seem to be problematic. A comparison between AI-assisted creations and AI-originated creations only shows the transference of dominance in the creative process, namely from the human creator to the AI creator. The identity of the creator is clear and simple in both cases, referring to ideas either from humans or from AI respectively. In the third category, where AI acts as an independent creator, as long as it’s able to express and interact with humans, understanding and interpreting its creative intent shouldn’t pose challenges. As for AI-assisted creations and AI-generated creations, although both result from the instrumental use of AI, AI-generated creations are more controversial, because humans can’t determine the content of the creations completely. In this case, the work can’t reflect the intentions of the human creator as the leading role in the creative process precisely. It only reflects the intentions processed by AI, and there is no doubt that the key issue lies in its “consciousness”.

Traditionally, artists bring their ideas to life directly by mastering their tools, which points to a highly interdependent relationship between artists and their creations. However, when AI with a “consciousness” is used as a creative tool, its involvement in the creative process resembles a third-party intervention, which disrupts the direct and intimate connection between artists and their creations. Setting aside the way AI understands and interprets information, the given instructions are always imperfect and leave room for ambiguity. These ambiguities usually reflect the aspects that the artist isn’t aware of, which would be left unresolved in the past. However, in the works created by AI nowadays, ambiguities and gaps are filled automatically to some extent, creating challenges similar to those encountered when adapting a novel into a film. Just as writers don’t and can’t describe every detail in their novels, filling in these blanks becomes a big challenge for filmmakers during the visualization process. Consequently, there are unavoidable differences between the adapted film and the novel, and these differences are also the reason why people usually have conflicting emotions and complex feelings when engaging with works created by AI.

Furthermore, interpretation means dealing with signs, and a sign is always connected to another sign that explains its meaning. This explanatory sign then becomes a sign itself, which is explained by another sign (Dalferth, 2020, 74). This interpretive process – according to Barthes’ engagement with semiotics in Mythologies – involves the historical and social meanings expressed by cultural symbols through their naturalization. Since every aspect and detail of a creation can serve as a basis for interpreting, the interpreting of the creation will inevitably be influenced to some extent by the content automatically created by AI, which makes true understanding difficult to reach. In this context, understanding and interpreting are the primary challenges of AI-based creations.

4. The possibilities and prospects

There is no doubt that appreciating, understanding and interpreting AI-based creations involve different challenges, depending on the degree of AI involvement in the creative process. Compared to other works, AI-assisted creations are the least affected and the least controversial due to the higher degree of human input. Appreciating AI-generated and AI-originated creations can be challenging and complex, but it’s not impossible, as experiencing beauty is a natural and spontaneous act that, according to Immanuel Kant, involves the internal processing of external objects. This act is triggered by the harmony and consistency between the internal sense mechanism and the presentation of the external object, “meaning that we take pleasure in something because we judge it beautiful, rather than judging it beautiful because we find it pleasurable” (Burnham, n.d.).

The studies mentioned earlier have demonstrated that people can actually appreciate works created by AI to some extent. However, biases disrupt their sense mechanism once they learn about the involvement of AI in the creation process. This leads to decreased ratings and causes the beauty they initially perceived to fade or even disappear. Thus, overcoming biases is a crucial first step in appreciating AI-generated and AI-originated creations or having them recognized and accepted as real artworks.

If people can overcome their biases against AI and come to appreciate its creations, they will inevitably face further challenges in interpreting and understanding when engaging with AI-generated works. After all, appreciating an artwork without understanding it is not only shallow and meaningless, but even impossible to some extent. Given that the greater involvement of AI in the creative process makes it difficult for human creators to control the final outcome, interpreting and understanding such works may require viewing the author’s intent as only partially comprehensible or even unknowable. Therefore, in the process of understanding and interpreting AI-generated works, a different approach is needed compared to creations where the author’s identity is clear and simple, such as AI-assisted or AI-originated works.

In this way, while fully embracing all categories of AI-based creations may be difficult, it’s far from impossible. Although some people hold biases or skepticism toward works created by AI, many others approach them with openness or even enthusiasm, welcoming new possibilities. For example, the International Photography Awards has introduced a new category specifically for “images that have NOT been created using conventional photography methods, or simply digitally altered or edited, but created or generated solely by executing a command to AI” (International Photography Awards, 2023). By separating these creations from traditional artworks and encouraging their independent competition, comparison, exhibition and discussion, this approach may well be an ideal initial response to this artistic revolution.

Benjamin, W. (1969). The Work of Art in the Age of Mechanical Reproduction. In Hannah Arendt (Eds.), Illuminations: Essays and Reflections. (pp. 217-251). New York: Schocken Books.

Burnham, D. (n.d.). Immanuel Kant: Aesthetics. Internet Encyclopedia of Philosophy. https://iep.utm.edu/kantaest/

Chen, C.-H. (2007). The Artificial Aura in the Age of Digital Reproduction, Chinese Journal of Communication Research, 12, 215-251.

Dalferth, I. U. (2020). Die Phänomene des Verstehens und die Praxis des Interpretierens Prolegomena zu einer theologischen Hermeneutik. In Andreas Mauz and Christiane Tietz (Eds.), Die Phänomene des Verstehens und die Praxis des Interpretierens (pp. 57-78). Paderborn: Verlag Ferdinand Schöningh.

International Photography Awards (2023, May 10). IPA’s Official Statement regarding AI-generated images. International Photography Awards. https://www.photoawards.com/ai-generated-images/

Lin, S.-Y. (2020). A Dilemma for Modest Actual Intentionalism. British Journal of Aesthetics, 60(2), 165-181.

Long, X. (2021). A Study on the Legal Status and Protection of AI-Generated Work, Soochow Law Review, 1(33), 139-181.

Marco, G., Rello, L., Gonzalo, J. (2024). Small Language Models can Outperform Humans in Short Creative Writing: A Study Comparing SLMs with Humans and LLMs. arXiv preprint arXiv:2409.11547

Mauz, A., Tietz, C. (2020). Verstehen und Interpretieren. Zur Einleitung. In Andreas Mauz and Christiane Tietz (Eds.), Die Phänomene des Verstehens und die Praxis des Interpretierens (pp. 1-20). Paderborn: Verlag Ferdinand Schöningh.

Perry, M., Margoni, T. (2010). From Music Tracks to Google Maps: Who Owns Computer Generated Works?. Computer Law and Security Review, 26, 621-629. available at SSRN: https://ssrn.com/abstract=1647584

Porter, B., Machery, E (2024, November 14). AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably. Scientific Reports. https://doi.org/10.1038/s41598-024-76900-1

Reinmuth, F. (2020). Wahres und vertretbares Verstehen. In Andreas Mauz and Christiane Tietz (Eds.), Die Phänomene des Verstehens und die Praxis des Interpretierens (pp. 87-125). Paderborn: Verlag Ferdinand Schöningh.

[PDF]