Neurofilosofía y Pensamiento Actual Neurophilosophy and Contemporary Thought

martes, 13 de octubre de 2026

The Art That Teaches Itself: Methodologies of Autonomous Learning in Algorithmic Creation

Bernabé Mallo

Doctor en Filosofía por la Universidad del País Vasco (UPV/EHU)
Investigador en neurofilosofía, evolución humana y origen del arte. / PhD in Philosophy – University of the Basque Country (UPV/EHU)
Researcher in neurophilosophy, human evolution, and the origins of art.

A review of Qian Wang's study (2024) in ACM Transactions on Multimedia Computing, Communications, and Applications: Learning-based Artificial Intelligence Artwork: Methodology and Future Directions

Introduction: the machine that becomes its own teacher

How does a machine learn to create art? Not through explicit instructions, nor through rules programmed by a human. It learns by observing. It learns by analysing. It learns by competing with itself. The study by Qian Wang, published in 2024 in the ACM Transactions on Multimedia Computing, Communications, and Applications, offers an exhaustive technical breakdown of the architectures that make contemporary algorithmic art possible. It does not limit itself to describing what tools exist, but explains how they work internally, unravelling the logic that allows a neural network to generate an image that a human might mistake for a masterpiece.

The article begins with a fundamental distinction between two major paradigms of machine learning applied to art: supervised learning models, which require labelled datasets and external guidance, and autonomous learning models, which generate and evaluate their own creations without direct human intervention. Qian Wang focuses on this second paradigm, and does so with a clarity that turns a technical subject into a window onto profound questions about the nature of creativity, authorship, and aesthetic knowledge.

The dance of GANs: creator and critic in perpetual competition

The heart of contemporary algorithmic art beats in Generative Adversarial Networks (GANs), and Qian Wang devotes a substantial part of his analysis to explaining their internal functioning. The metaphor he uses is especially illuminating: it is a dynamic of "creator and critic".

Let us imagine two neural networks competing with each other. The first, the generator, tries to produce images that appear authentic. The second, the discriminator, tries to distinguish between generated images and real works by human artists. Each time the generator produces an image, the discriminator evaluates it and issues a judgment: is it authentic or fake? The generator adjusts its parameters to deceive the discriminator. The discriminator adjusts its own to avoid being deceived. And so, in an endless competition, both networks improve simultaneously.

This process has a notable consequence: the generator receives no explicit instructions on how to create art. It is not told "paint like Van Gogh" or "use this colour palette". It learns on its own, from adversarial feedback, which configurations produce images the discriminator cannot reject. It is, in a sense, a form of learning by discovery: the machine explores a space of possibilities and finds, through trial and error, the regions that produce convincing results.

Qian Wang notes that this dynamic reproduces, algorithmically, something that occurs in the world of human art: the tension between creation and criticism, between the innovative impulse and the evaluative judgment. But there is a crucial difference. In human art, the critic is a subject with lived experience, with criteria formed by years of exposure and reflection. In the GAN system, the critic is a mathematical function, a set of rules that evaluate authenticity in statistical terms. The machine's "criticism" is not an aesthetic judgment; it is a probability calculation.

From adversarial learning to diffusion models: the technical evolution

Qian Wang's article does not stop at GANs. It also traces the evolution towards diffusion models, the architecture driving contemporary systems such as Midjourney or Stable Diffusion. Unlike GANs, which generate images in one go through adversarial competition, diffusion models operate through a process of gradual ruin and reconstruction.

The principle is as follows: a real image is taken and noise is progressively added until it becomes a chaos of random pixels. Then, a neural network is trained to reverse that process: given a fragment of noise, it must reconstruct the original image step by step. Once trained, the network can generate new images starting from pure noise, guided by textual instructions that orient the reconstruction towards a specific result.

This architecture has significant advantages over GANs. It is more stable during training, produces higher-quality images, and allows finer control over the final result. But, from the perspective of our research, it shares with GANs a fundamental characteristic: the machine does not "see" the world, but learns to reconstruct it from statistical patterns. There is no direct visual experience, no body traversing a scene, no memory associating an image with an emotion. There is, solely, a mathematical space of possibilities that the network has learned to navigate.

Data repositories: the fuel of learning

One aspect Qian Wang analyses in detail is the dependence on datasets. For an autonomous learning system to function, it must be fed with large repositories of labelled images. These repositories, such as the famous WikiArt or the internal datasets of AI companies, contain tens of thousands of works classified by style, author, and period.

The system learns by extracting patterns from this data. It identifies that certain colour combinations are characteristic of Impressionism, that certain compositional structures are typical of the Baroque, that certain themes appear recurrently in Renaissance art. But this learning has an intrinsic limit: the machine can only learn what is in the data. If the repository is biased —towards Western art, towards certain historical periods, towards certain genres— the system will inherit those biases and reproduce them in its creations.

This point connects with a broader reflection on the political economy of algorithmic art. The large image repositories that feed AI systems are composed, for the most part, of works by artists who have not given their consent to be used as training data. Their creations are extracted, fragmented, and recombined without compensation or recognition. The machine learns from the work of thousands of human artists, but does not acknowledge its debt to them. This asymmetry raises ethical and legal questions that still lack satisfactory answers.

The question of originality: what does it mean to create?

One of the most suggestive passages in Qian Wang's article addresses the question of originality in AI-generated art. The question is simple but profound: if a machine learns from thousands of existing works and generates a new image, is that image original? Or is it a mere recombination of pre-existing elements?

The usual answer, from a technical perspective, is that the machine's originality is always statistical. It does not invent from scratch; it explores a space of possibilities defined by the training data. Its "creativity" consists of finding combinations that were not present in the original corpus, but that are, in a sense, derivable from it. It is a form of interpolation, not of creation ex nihilo.

However, this answer raises an uncomfortable question: is human originality of another nature? If we think about it carefully, we too are, to a certain extent, machines that develop and function within a concrete niche. Our ideas, our images, our ways of seeing the world do not emerge from nothing; they derive from our own lived niche —the social, cultural, and natural environment in which we grow— and from our biological characteristics. No human artist invents from absolute emptiness; we all create from what we have seen, felt, learned, inherited.

So can we really assert that human originality implies a rupture with the existing, while the machine's is only recombination? Perhaps the distinction is not so clear-cut. What we call "rupture" in human art —the irruption of a new style, a new sensibility, a new form of expression— does not arise from a probabilistic calculation in the algorithmic sense, but it does arise from a biological calculation: a calculation that our nervous system performs with concrete characteristics that, from our perspective, are survival, symbolon, and wholeness. It is, if you will, a form of internal necessity, biologically internal.

The difference, then, would not lie in the presence or absence of calculation, but in the nature of that calculation. The machine's calculation operates on external data, on statistical patterns extracted from a corpus. The living organism's calculation operates on its own experience, on its needs, on its evolutionary and biographical history. Both are, in a sense, "machines", but one is driven by probability and the other by the need to survive, to symbolise, and to achieve wholeness. And that need, unlike statistical calculation, is not exhausted in the optimisation of a function: it unfolds over time, it transforms with experience, it confronts death and meaning.

Is that a "rupture"? Perhaps yes, perhaps no. Perhaps the difference between human art and algorithmic art is not one of essence, but of degree of embodiment. But what we can affirm is that biological calculation, unlike algorithmic calculation, is traversed by an interiority that is not reducible to statistics. And that interiority is, precisely, what art, in its origin, has sought to express.

A reading from the S/Y/C model

Qian Wang's analysis offers a concrete case in which the functioning of the Law of Biological Coherence S/Y/C that forms the core of our research can be observed (Mallo, 2023, 2025, 2026a, 2026b). The three dimensions of our thesis unfold here with uncommon clarity, precisely because the contrast with the machine allows us to see what we take for granted in the human.

The S (Survival) dimension manifests in the adaptive function of algorithmic art. Qian Wang notes that AI art systems are developed because they solve practical problems: they automate creative tasks, broaden access to design tools, offer new expressive possibilities. This is a legitimate function of cultural survival: technology is adopted when it solves a problem. But the survival of the machine is not that of the organism using it. The system does not need to create art in order to survive; the human who employs it does. AI is a tool for survival, not a subject that survives through it.

The Y (Symbolon) dimension is the point of greatest contrast. GANs and diffusion models manipulate symbols —visual representations, linguistic labels, stylistic patterns— with impressive efficiency. But they do not inhabit the symbol. There is no mutual recognition, no community of lived shared codes. The machine can label a work with the word "Baroque" and detect in it the features that criticism has associated with that period, but it has not grown up in a culture where that word meant something, has not inherited a tradition, has not felt the weight of a shared history. The human symbolon is lived; the machine's is simulated.

The C (Wholeness) dimension manifests in the search for convergence that characterises GANs and diffusion models. The generator and the discriminator compete until they reach an equilibrium, a point at which the generated images are indistinguishable from real ones. Diffusion models, for their part, reconstruct an image from chaos, progressively closing the process until they reach a coherent form. This convergence is a form of algorithmic wholeness. But it is a wholeness without yearning, without the biological drive that in humans seeks to integrate parts into a coherent whole. The machine "closes" a form because its loss function tells it to, not because it needs to make sense of its experience.

Surgical Philosophy invites us to make a precise analytical cut in Qian Wang's study. It is not about rejecting technology nor idealising it, but about distinguishing levels. At the level of technical production, autonomous learning systems are powerful tools that can expand the capacities of the artist and the educator. At the level of aesthetic experience, AI cannot replace the embodied subjectivity that constitutes the heart of art. And at the level of cultural responsibility, an ethical framework is needed to ensure that these tools are used to enhance the human, not to replace it.

Final considerations: art as encounter, not as optimisation

Qian Wang's study has the merit of offering a rigorous and complete vision of the architectures that make algorithmic art possible. His analysis of GANs, diffusion models, and data repositories provides a solid basis for understanding how the machine works internally. But, at the same time, his work reveals —perhaps unintentionally— the limits of that approach.

The machine can learn to generate images that look like art. It can optimise its creations to maximise authenticity, coherence, surprise. It can even surpass humans in specific recognition and classification tasks. But it cannot understand what it does, nor feel what it produces, nor inhabit the process of creation as an act of encounter with the other.

Now, what does this research contribute to the whole of our inquiry into the origin of art? Its most valuable contribution is the comparison it makes possible. By observing how an artificial system, organised by us, learns to generate art through adversarial competition and pattern reconstruction, we are offered an unexpected mirror for thinking about how the biological system itself, tens of thousands of years ago, might have fragmented, assimilated, and reformulated the world to create something that today encompasses us all and everything.

Comparing these two ways of constructing creation —the artificial, guided by algorithms and data; the biological, guided by needs, biases, and yearnings— can be useful for consolidating a possible theory about how biological systems in general came to produce art. It is not about equating both processes, but about using the contrast as a heuristic tool: the machine shows us, through absence, what life contributes. And in that absence —the lack of a body that feels, of a history that hurts, of a need that drives— we find precisely the keys to what art, in its origin, was and continues to be.

Art, in short, is not an optimisation of parameters. It is an encounter between embodied subjectivities. And that encounter, for now, remains the exclusive patrimony of the living.

References

Mallo, B. (2023). La construcción neuro-simbólica. Una aproximación al funcionamiento del cerebro desde una perspectiva multidisciplinar [Doctoral thesis, University of the Basque Country - Euskal Herriko Unibertsitatea]. ADDI Repository. http://hdl.handle.net/10810/62701

Mallo, B. (2025). Arte y biología: Una aproximación neurofilosófica al origen de la experiencia estética. https://www.amazon.com/dp/B0E8Y5WZMK

Mallo, B. (2025). Art and biology: A neurophilosophical approach to the origin of aesthetic experience. https://www.amazon.com/dp/B0E8Y6C2XN

Mallo, B. (2026a). De la filosofía quirúrgica a la ley de coherencia biológica S/Y/C: Hacia una investigación sobre el origen del arte en la especie Homo. https://isbn.bibna.gub.uy/catalogo.php?mode=detalle&nt=57196

Mallo, B. (2026a). De la filosofía quirúrgica a la ley de coherencia biológica S/Y/C: Hacia una investigación sobre el origen del arte en la especie Homo [Kindle edition]. https://www.amazon.com/dp/B0GYGTJD5C

Mallo, B. (2026b). From surgical philosophy to the law of biological coherence S/Y/C: Toward a study of the origin of art in the Homo lineage. https://isbn.bibna.gub.uy/catalogo.php?mode=detalle&nt=57197

Wang, Q. (2024). Learning-based artificial intelligence artwork: Methodology and future directions. ACM Transactions on Multimedia Computing, Communications, and Applications, *20*(11), Article 347. https://doi.org/10.1145/3698105

Autor / Author

Bernabé Mallo
Doctor en Filosofía – Universidad del País Vasco / Euskal Herriko Unibertsitatea (UPV/EHU)
Investigador independiente en neurofilosofía, evolución humana y origen del arte.

Bernabé Mallo
PhD in Philosophy – University of the Basque Country / Euskal Herriko Unibertsitatea (UPV/EHU)
Independent researcher in neurophilosophy, human evolution, and the origin of art.

Enlaces / Links

Página de autor Amazon / Amazon Author Page: https://www.amazon.com/author/bernabemallo
ORCID: https://orcid.org/0000-0001-9002-9728
Plataforma EHUenRed / Link EHUenRed: https://www.ehu.eus/es/web/masterrak-eta-graduondokoak/red-latinoamericana-de-posgrados
Canal YouTube / Channel YouTube: https://www.youtube.com/@neuroideas815
Canal YouTube / Channel YouTube: https://www.youtube.com/channel/UCBsf6OZ482NjST6QA-hvYtQ
Publicaciones y proyectos en desarrollo / Publications and projects:
https://www.amazon.com/author/bernabemallo
https://ehuenred.theglocal.network/ideas/el-origen-del-arte-en-el-cerebro-de-makapansgat-al-moma-del-primate-al-sapiens

El arte que se enseña a sí mismo: metodologías del aprendizaje autónomo en la creación algorítmica

Bernabé Mallo

Una reseña del estudio de Qian Wang (2024) en ACM Transactions on Multimedia Computing, Communications, and Applications: Learning-based Artificial Intelligence Artwork: Methodology and Future Directions

Introducción: la máquina que se convierte en su propio maestro

¿Cómo aprende una máquina a crear arte? No mediante instrucciones explícitas, ni mediante reglas programadas por un humano. Aprende observando. Aprende analizando. Aprende compitiendo consigo misma. El estudio de Qian Wang, publicado en 2024 en las ACM Transactions on Multimedia Computing, Communications, and Applications, ofrece un desglose técnico exhaustivo de las arquitecturas que hacen posible el arte algorítmico contemporáneo. No se limita a describir qué herramientas existen, sino que explica cómo funcionan por dentro, desentrañando la lógica que permite a una red neuronal generar una imagen que un humano podría confundir con una obra maestra.

El artículo parte de una distinción fundamental entre dos grandes paradigmas del aprendizaje automático aplicado al arte: los modelos de aprendizaje supervisado, que requieren conjuntos de datos etiquetados y una guía externa, y los modelos de aprendizaje autónomo, que generan y evalúan sus propias creaciones sin intervención humana directa. Qian Wang se centra en este segundo paradigma, y lo hace con una claridad que convierte un tema técnico en una ventana hacia preguntas profundas sobre la naturaleza de la creatividad, la autoría y el conocimiento estético.

La danza de las GAN: creador y crítico en competencia perpetua

El corazón del arte algorítmico contemporáneo late en las Redes Generativas Antagónicas (GAN), y Qian Wang dedica una parte sustancial de su análisis a explicar su funcionamiento interno. La metáfora que utiliza es especialmente iluminadora: se trata de una dinámica de "creador y crítico".

Imaginemos dos redes neuronales que compiten entre sí. La primera, el generador, intenta producir imágenes que parezcan auténticas. La segunda, el discriminador, intenta distinguir entre las imágenes generadas y las obras reales de artistas humanos. Cada vez que el generador produce una imagen, el discriminador la evalúa y emite un juicio: ¿es auténtica o falsa? El generador ajusta sus parámetros para engañar al discriminador. El discriminador ajusta los suyos para no dejarse engañar. Y así, en una competencia sin fin, ambas redes mejoran simultáneamente.

Este proceso tiene una consecuencia notable: el generador no recibe instrucciones explícitas sobre cómo crear arte. No se le dice "pinta como Van Gogh" o "usa esta paleta de colores". Aprende por sí mismo, a partir de la retroalimentación adversarial, qué configuraciones producen imágenes que el discriminador no puede rechazar. Es, en cierto sentido, una forma de aprendizaje por descubrimiento: la máquina explora un espacio de posibilidades y encuentra, por ensayo y error, las regiones que producen resultados convincentes.

Qian Wang señala que esta dinámica reproduce, de manera algorítmica, algo que ocurre en el mundo del arte humano: la tensión entre la creación y la crítica, entre el impulso innovador y el juicio evaluador. Pero hay una diferencia crucial. En el arte humano, el crítico es un sujeto con experiencia vivida, con criterios formados por años de exposición y reflexión. En el sistema GAN, el crítico es una función matemática, un conjunto de reglas que evalúan la autenticidad en términos estadísticos. La "crítica" de la máquina no es un juicio estético; es un cálculo de probabilidad.

Del aprendizaje adversarial a los modelos de difusión: la evolución técnica

El artículo de Qian Wang no se detiene en las GAN. Traza también la evolución hacia los modelos de difusión, la arquitectura que impulsa sistemas contemporáneos como Midjourney o Stable Diffusion. A diferencia de las GAN, que generan imágenes de una vez mediante una competencia adversarial, los modelos de difusión operan mediante un proceso de ruina y reconstrucción gradual.

El principio es el siguiente: se toma una imagen real y se le añade ruido progresivamente hasta que se convierte en un caos de píxeles aleatorios. Luego, se entrena a una red neuronal para invertir ese proceso: dado un fragmento de ruido, debe reconstruir la imagen original paso a paso. Una vez entrenada, la red puede generar imágenes nuevas partiendo de ruido puro, guiada por instrucciones textuales que orientan la reconstrucción hacia un resultado específico.

Esta arquitectura tiene ventajas significativas sobre las GAN. Es más estable durante el entrenamiento, produce imágenes de mayor calidad y permite un control más fino sobre el resultado final. Pero, desde la perspectiva de nuestra investigación, comparte con las GAN una característica fundamental: la máquina no "ve" el mundo, sino que aprende a reconstruirlo a partir de patrones estadísticos. No hay una experiencia visual directa, no hay un cuerpo que recorra una escena, no hay una memoria que asocie una imagen con una emoción. Hay, únicamente, un espacio matemático de posibilidades que la red ha aprendido a navegar.

Los repositorios de datos: el combustible del aprendizaje

Un aspecto que Qian Wang analiza con detalle es la dependencia de los conjuntos de datos. Para que un sistema de aprendizaje autónomo funcione, necesita ser alimentado con grandes repositorios de imágenes etiquetadas. Estos repositorios, como el famoso WikiArt o los datasets internos de las empresas de IA, contienen decenas de miles de obras clasificadas por estilo, autor y época.

El sistema aprende extrayendo patrones de estos datos. Identifica que ciertas combinaciones de colores son características del impresionismo, que ciertas estructuras compositivas son típicas del Barroco, que ciertos temas aparecen recurrentemente en el arte renacentista. Pero este aprendizaje tiene un límite intrínseco: la máquina solo puede aprender lo que está en los datos. Si el repositorio está sesgado —hacia el arte occidental, hacia ciertos períodos históricos, hacia ciertos géneros—, el sistema heredará esos sesgos y los reproducirá en sus creaciones.

Este punto conecta con una reflexión más amplia sobre la economía política del arte algorítmico. Los grandes repositorios de imágenes que alimentan a los sistemas de IA están compuestos, en su mayoría, por obras de artistas que no han dado su consentimiento para ser utilizadas como datos de entrenamiento. Sus creaciones son extraídas, fragmentadas y recombinadas sin compensación ni reconocimiento. La máquina aprende del trabajo de miles de artistas humanos, pero no reconoce su deuda con ellos. Esta asimetría plantea preguntas éticas y legales que aún no tienen respuesta satisfactoria.

La cuestión de la originalidad: ¿qué significa crear?

Uno de los pasajes más sugerentes del artículo de Qian Wang aborda la cuestión de la originalidad en el arte generado por IA. La pregunta es simple pero profunda: si una máquina aprende de miles de obras existentes y genera una imagen nueva, ¿es esa imagen original? ¿O es una mera recombinación de elementos preexistentes?

La respuesta habitual, desde una perspectiva técnica, es que la originalidad de la máquina es siempre estadística. No inventa desde cero; explora un espacio de posibilidades definido por los datos de entrenamiento. Su "creatividad" consiste en encontrar combinaciones que no estaban presentes en el corpus original, pero que son, en cierto sentido, derivables de él. Es una forma de interpolación, no de creación ex nihilo.

Sin embargo, esta respuesta plantea una pregunta incómoda: ¿acaso la originalidad humana es de otra naturaleza? Si lo pensamos detenidamente, también nosotros somos, en cierta medida, máquinas que se desarrollan y funcionan dentro de un nicho concreto. Nuestras ideas, nuestras imágenes, nuestras formas de ver el mundo no surgen de la nada; se derivan de nuestro propio nicho vivencial —el entorno social, cultural y natural en el que crecemos— y de nuestras características biológicas. Ningún artista humano inventa desde el vacío absoluto; todos creamos a partir de lo que hemos visto, sentido, aprendido, heredado.

Entonces, ¿podemos realmente asegurar que la originalidad humana implica una ruptura con lo existente, mientras que la de la máquina es solo recombinación? Quizás la distinción no sea tan nítida. Lo que llamamos "ruptura" en el arte humano —la irrupción de un nuevo estilo, de una nueva sensibilidad, de una nueva forma de expresión— no surge de un cálculo probabilístico en el sentido algorítmico, pero sí de un cálculo biológico: un cálculo que nuestro sistema nervioso realiza con unas características concretas que, desde nuestra perspectiva, son supervivencia, symbolon y completitud. Es, si se quiere, una forma de necesidad interna, biológicamente interna.

La diferencia, entonces, no residiría en la presencia o ausencia de cálculo, sino en la naturaleza de ese cálculo. El cálculo de la máquina opera sobre datos externos, sobre patrones estadísticos extraídos de un corpus. El cálculo del organismo vivo opera sobre su propia experiencia, sobre sus necesidades, sobre su historia evolutiva y biográfica. Ambos son, en cierto sentido, "máquinas", pero una está movida por la probabilidad y la otra por la necesidad de sobrevivir, de simbolizar y de completarse. Y esa necesidad, a diferencia del cálculo estadístico, no se agota en la optimización de una función: se despliega en el tiempo, se transforma con la experiencia, se enfrenta a la muerte y al sentido.

¿Es eso una "ruptura"? Quizás sí, quizás no. Quizás la diferencia entre el arte humano y el arte algorítmico no sea de esencia, sino de grado de encarnación. Pero lo que sí podemos afirmar es que el cálculo biológico, a diferencia del algorítmico, está atravesado por una interioridad que no se reduce a la estadística. Y esa interioridad es, precisamente, lo que el arte, en su origen, ha querido expresar.

Una lectura desde el modelo S/Y/C

El análisis de Qian Wang ofrece un caso concreto donde se puede observar el funcionamiento de la Ley de coherencia biológica S/Y/C que constituye el núcleo de nuestra investigación (Mallo, 2023, 2025, 2026a, 2026b). Las tres dimensiones de nuestra tesis se despliegan aquí con una claridad poco común, precisamente porque el contraste con la máquina nos permite ver lo que en el humano damos por sentado.

La dimensión S (Supervivencia) se manifiesta en la función adaptativa del arte algorítmico. Qian Wang señala que los sistemas de IA artística se desarrollan porque resuelven problemas prácticos: automatizan tareas creativas, amplían el acceso a herramientas de diseño, ofrecen nuevas posibilidades expresivas. Esta es una función legítima de supervivencia cultural: la tecnología se adopta cuando resuelve un problema. Pero la supervivencia de la máquina no es la del organismo que la utiliza. El sistema no necesita crear arte para sobrevivir; el humano que lo emplea, sí. La IA es una herramienta de supervivencia, no un sujeto que sobreviva a través de ella.

La dimensión Y (Symbolon) es el punto de mayor contraste. Las GAN y los modelos de difusión manipulan símbolos —representaciones visuales, etiquetas lingüísticas, patrones estilísticos— con una eficacia impresionante. Pero no habitan el símbolo. No hay un reconocimiento mutuo, una comunidad de códigos compartidos vividos. La máquina puede etiquetar una obra con la palabra "Barroco" y detectar en ella los rasgos que la crítica ha asociado a ese período, pero no ha crecido en una cultura donde esa palabra significaba algo, no ha heredado una tradición, no ha sentido el peso de una historia compartida. El symbolon humano es vivido; el de la máquina, simulado.

La dimensión C (Completitud) se manifiesta en la búsqueda de convergencia que caracteriza a las GAN y a los modelos de difusión. El generador y el discriminador compiten hasta alcanzar un equilibrio, un punto en el que las imágenes generadas son indistinguibles de las reales. Los modelos de difusión, por su parte, reconstruyen una imagen a partir del caos, cerrando progresivamente el proceso hasta alcanzar una forma coherente. Esta convergencia es una forma de completitud algorítmica. Pero es una completitud sin anhelo, sin la pulsión biológica que en el humano busca integrar las partes en un todo coherente. La máquina "cierra" una forma porque su función de pérdida se lo indica, no porque necesite dar sentido a su experiencia.

La Filosofía Quirúrgica nos invita a aplicar un corte analítico preciso al estudio de Qian Wang. No se trata de rechazar la tecnología ni de idealizarla, sino de distinguir niveles. En el nivel de la producción técnica, los sistemas de aprendizaje autónomo son herramientas poderosas que pueden ampliar las capacidades del artista y del educador. En el nivel de la experiencia estética, la IA no puede reemplazar la subjetividad encarnada que constituye el corazón del arte. Y en el nivel de la responsabilidad cultural, es necesario un marco ético que garantice que estas herramientas se utilizan para potenciar lo humano, no para sustituirlo.

Consideraciones finales: el arte como encuentro, no como optimización

El estudio de Qian Wang tiene el mérito de ofrecer una visión rigurosa y completa de las arquitecturas que hacen posible el arte algorítmico. Su análisis de las GAN, de los modelos de difusión y de los repositorios de datos proporciona una base sólida para comprender cómo funciona la máquina por dentro. Pero, al mismo tiempo, su trabajo revela —quizás sin proponérselo— los límites de ese enfoque.

La máquina puede aprender a generar imágenes que parecen arte. Puede optimizar sus creaciones para maximizar la autenticidad, la coherencia, la sorpresa. Puede incluso superar a los humanos en tareas específicas de reconocimiento y clasificación. Pero no puede comprender lo que hace, ni sentir lo que produce, ni habitar el proceso de creación como un acto de encuentro con el otro.

Ahora bien, ¿qué aporta esta investigación al conjunto de nuestra indagación sobre el origen del arte? Su contribución más valiosa es la comparación que hace posible. Al observar cómo un sistema artificial, organizado por nosotros, aprende a generar arte mediante la competencia adversarial y la reconstrucción de patrones, se nos ofrece un espejo inesperado para pensar cómo el propio sistema biológico, hace decenas de miles de años, pudo haber fragmentado, asimilado y reformulado el mundo para crear algo que hoy nos abarca a todos y a todo.

Comparar estas dos maneras de construir la creación —la artificial, guiada por algoritmos y datos; la biológica, guiada por necesidades, sesgos y anhelos— puede ser útil para afianzar alguna posible teoría sobre cómo los sistemas biológicos en general llegaron a producir arte. No se trata de equiparar ambos procesos, sino de utilizar el contraste como herramienta heurística: la máquina nos muestra, por ausencia, lo que la vida aporta. Y en esa ausencia —la falta de un cuerpo que sienta, de una historia que duela, de una necesidad que impulse— encontramos precisamente las claves de lo que el arte, en su origen, fue y sigue siendo.

El arte, en definitiva, no es una optimización de parámetros. Es un encuentro entre subjetividades encarnadas. Y ese encuentro, por ahora, sigue siendo patrimonio exclusivo de lo vivo.

Referencias bibliográficas

Mallo, B. (2023). La construcción neuro-simbólica. Una aproximación al funcionamiento del cerebro desde una perspectiva multidisciplinar [Tesis doctoral, Universidad del País Vasco - Euskal Herriko Unibertsitatea]. Repositorio ADDI. http://hdl.handle.net/10810/62701

Mallo, B. (2025). Arte y biología: Una aproximación neurofilosófica al origen de la experiencia estética. https://www.amazon.com/dp/B0E8Y5WZMK

Mallo, B. (2025). Art and biology: A neurophilosophical approach to the origin of aesthetic experience. https://www.amazon.com/dp/B0E8Y6C2XN

Mallo, B. (2026a). De la filosofía quirúrgica a la ley de coherencia biológica S/Y/C: Hacia una investigación sobre el origen del arte en la especie Homo [Versión Kindle]. https://www.amazon.com/dp/B0GYGTJD5C

Autor / Author

Bernabé Mallo
Doctor en Filosofía – Universidad del País Vasco / Euskal Herriko Unibertsitatea (UPV/EHU)
Investigador independiente en neurofilosofía, evolución humana y origen del arte.

Bernabé Mallo
PhD in Philosophy – University of the Basque Country / Euskal Herriko Unibertsitatea (UPV/EHU)
Independent researcher in neurophilosophy, human evolution, and the origin of art.

Enlaces / Links

jueves, 1 de octubre de 2026

The Algorithmic Eye: How Artificial Intelligence Learns to See Art

Bernabé Mallo

A review of the study by Yali Liu and Can Zhu (2025) in Scientific Reports: The use of deep learning and artificial intelligence-based creative teaching in art education

Introduction: when the machine looks at a painting

What does a machine see when it contemplates a painting? It experiences no aesthetic pleasure, is not moved by colours, evokes no memories. But can it learn to recognise what makes a work what it is? Can it decompose a brushstroke into its essential components —texture, colour, composition— and, from there, generate something new?

The study by Yali Liu and Can Zhu, published in 2025 in Scientific Reports (Nature Portfolio), addresses precisely this question. The authors propose and optimise an innovative artistic creation system called the Creative Intelligence Cloud (CIC), which combines generative adversarial networks (GANs) and convolutional neural networks (CNNs) to explore how deep learning can be applied to art education. The goal is not only to generate images, but to understand the process by which a machine acquires what we might call an "aesthetic knowledge".

This approach places us before a question that runs through our research on the origin of art: what does it mean to know a work of art? Is it enough to identify patterns, styles, and compositional structures, or does it require something more —a lived experience, a body that feels, a symbol that is inhabited— that the machine, for now, does not possess? The answer, as we shall see, forces us to distinguish between two forms of knowledge: the one based on statistical correlation and the one emerging from the single function of the nervous system.

The CIC system: anatomy of a visual learning process

The system proposed by Liu and Zhu is structured around two fundamental technical components: generative adversarial networks (GANs) and convolutional neural networks (CNNs). GANs, introduced by Ian Goodfellow in 2014, consist of two networks competing with each other: a generator, which produces images, and a discriminator, which tries to distinguish generated images from real ones. This adversarial competition forces the generator to progressively improve the quality of its creations.

CNNs, for their part, are the tool through which the system "reads" images. Unlike a traditional neural network, CNNs are designed to process data with a grid structure —such as the pixels of an image— and are capable of detecting local features (edges, textures, patterns) that are then combined into higher-level representations (forms, objects, compositions). It is, in a sense, an algorithmic imitation of how the human visual cortex processes information: from simple features to complex configurations.

The CIC system is trained on massive art repositories, such as the WikiArt dataset, which contains tens of thousands of paintings labelled by style, artist, and period. From this training, the model learns to decompose works into their constituent elements: brushstrokes, colour palettes, compositional structures. It does not "see" the painting as a meaningful whole, but as a matrix of features that it can analyse, compare, and recombine.

The "data lens": how the machine learns the visual language

The learning process of the CIC system illustrates what, in another review on this blog, we have called the "data lens" of artificial intelligence. Unlike a photographic camera, which captures the light of a real scene, AI does not see the world directly. It fragments billions of pre-existing images into minimal units of visual information —the so-called tokens— and associates them with linguistic labels (words such as "impressionist", "portrait", "blue"). In this way, the machine learns a "visual language of the world" that allows it, once trained, to move fluidly and autonomously across artistic movements, styles, and eras.

This process has profound implications. On the one hand, it allows unprecedented versatility: a model trained on WikiArt can generate an image in the style of Van Gogh, a hyperrealistic photograph, or a geometric abstraction with equal ease. On the other hand, it raises the question of what "originality" means when the system is trained on millions of existing works. The machine does not invent from scratch; it recombines what it has learned. Its "creativity" is, at bottom, a sophisticated form of statistical interpolation.

What the machine cannot learn: the symbolon gap

Here is where technical analysis meets philosophy. Liu and Zhu demonstrate that the CIC system is capable of analysing works of art, identifying styles, and generating new images. But does it understand what it does? The answer, from the perspective of our research, is negative. The machine can manipulate visual and linguistic tokens, but it does not inhabit the symbol. There is no one "inside" who recognises or is recognised in the act of creation.

This distinction is crucial. Symbolon, in its original etymological sense, is an act of recognition through shared codes. When a human artist creates, they do not merely combine forms: they project their interiority, communicate an experience, seek an encounter with the other. The work of art is a bridge between subjectivities. The machine, by contrast, produces objects that can be interpreted as art, but it does not generate the living process —the survival substrate, the symbolic dimension, the drive toward wholeness— that gives meaning to human creation.

The CIC system, by decomposing works into statistical features, loses precisely what makes a work a work: its character as an encounter. The machine can learn about art, but it cannot learn from art. It has no body that feels, no history that hurts, no need that drives it. Its knowledge is, at best, a shadow of human aesthetic knowledge.

A reading from the S/Y/C model

The study by Liu and Zhu offers a concrete case in which the functioning of the Law of Biological Coherence S/Y/C that forms the core of our research can be observed (Mallo, 2023, 2025, 2026a, 2026b). The three dimensions of our thesis unfold here with uncommon clarity, precisely because the contrast with the machine allows us to see what we take for granted in the human.

The S (Survival) dimension manifests in the usefulness of the CIC system. Its stated goal is to improve art education, provide automated feedback, and broaden access to creative tools. This is a legitimate adaptive function: technology, like Neolithic pottery, is adopted when it solves a practical problem. But the survival of the system is not that of the organism using it. The machine does not need to learn art in order to survive; the human who uses it does. AI is a tool for cultural survival, not a subject that survives through it.

The Y (Symbolon) dimension is the point of greatest contrast. The CIC system manipulates symbols —labels, styles, visual representations— with impressive efficiency. But it does not inhabit the symbol. There is no mutual recognition, no community of lived shared codes. The machine can classify a painting as "impressionist", but it does not know what it means to be impressionist, has not felt the light of southern France, has not argued with an academic about brushwork. The human symbolon is lived; the machine's is simulated.

The C (Wholeness) dimension manifests in the machine's search to close a form, to generate a coherent image. GANs, in particular, operate through a process of convergence: the generator and the discriminator compete until they reach an equilibrium, a point at which the generated images are indistinguishable from real ones. This convergence is a form of algorithmic wholeness. But it is a wholeness without yearning, without the biological drive that in humans seeks to integrate parts into a coherent whole. The machine "closes" a form because its loss function tells it to, not because it needs to make sense of its experience.

Surgical Philosophy invites us to make a precise analytical cut in the study by Liu and Zhu. It is not about rejecting technology nor idealising it, but about distinguishing levels. At the level of technical production, the CIC system is a powerful tool that can expand the capacities of the artist and the educator. At the level of aesthetic experience, AI cannot replace the embodied subjectivity that constitutes the heart of art. And at the level of cultural responsibility, an ethical framework is needed —such as the technological humanism we have discussed in other reviews— to ensure that these tools are used to enhance the human, not to replace it.

Final considerations: art as encounter, not as classification

The study by Yali Liu and Can Zhu has the merit of offering a rigorous and well-documented vision of how deep learning can be applied to art education. Their CIC system demonstrates that machines are capable of analysing, classifying, and generating images with a degree of sophistication that a decade ago would have seemed impossible. But, at the same time, their work reveals —perhaps unintentionally— the limits of that approach.

The machine can learn to recognise a style, but not to understand why that style matters. It can decompose a brushstroke, but not to feel what the brushstroke expresses. It can generate a new image, but not to inhabit the process of creation as an act of encounter with the other.

Art, in its origin and in its function, is not a classification of visual features. It is an encounter between embodied subjectivities. And that encounter, for now, remains the exclusive patrimony of the living. AI can be an ally in art education, a tool that broadens access and enhances creativity. But it cannot occupy the place of the artist, nor of the spectator, nor of the educator. Because art, like survival, symbol, and wholeness, is something that is lived before it can be measured.

Now, what does this research contribute to the whole of our inquiry into the origin of art? We believe that its most notable contribution is not only what it reveals about AI's capacities, but the comparison it makes possible. By observing how an artificial system, organised by us, decomposes and recomposes humanity's visual legacy, we are offered an unexpected mirror for thinking about how the biological system itself, tens of thousands of years ago, might have fragmented, assimilated, and reformulated the world to create something that today encompasses us all and everything.

References

Liu, Y., & Zhu, C. (2025). The use of deep learning and artificial intelligence-based creative teaching in art education. Scientific Reports, *15*, Article 15763. https://doi.org/10.1038/s41598-025-00892-9

Mallo, B. (2025). Arte y biología: Una aproximación neurofilosófica al origen de la experiencia estética. https://www.amazon.com/dp/B0E8Y5WZMK

Mallo, B. (2025). Art and biology: A neurophilosophical approach to the origin of aesthetic experience. https://www.amazon.com/dp/B0E8Y6C2XN

Yu, C., & Liu, Z. (2026). Multi-task convolutional network for artwork analysis and personalized guidance. Scientific Reports, *16*, Article 70478. https://doi.org/10.1038/s41598-026-70478-6

Autor / Author

Bernabé Mallo
Doctor en Filosofía – Universidad del País Vasco / Euskal Herriko Unibertsitatea (UPV/EHU)
Investigador independiente en neurofilosofía, evolución humana y origen del arte.

Bernabé Mallo
PhD in Philosophy – University of the Basque Country / Euskal Herriko Unibertsitatea (UPV/EHU)
Independent researcher in neurophilosophy, human evolution, and the origin of art.

Enlaces / Links

El ojo algorítmico: cómo la inteligencia artificial aprende a ver el arte

Bernabé Mallo

Una reseña del estudio de Yali Liu y Can Zhu (2025) en Scientific Reports: The use of deep learning and artificial intelligence-based creative teaching in art education

Introducción: cuando la máquina mira un cuadro

¿Qué ve una máquina cuando contempla una pintura? No experimenta placer estético, no se conmueve con los colores, no evoca recuerdos. Pero, ¿puede aprender a reconocer lo que hace que una obra sea lo que es? ¿Puede descomponer una pincelada en sus componentes esenciales —textura, color, composición— y, a partir de ahí, generar algo nuevo?

El estudio de Yali Liu y Can Zhu, publicado en 2025 en Scientific Reports (Nature Portfolio), aborda precisamente esta cuestión. Los autores proponen y optimizan un sistema innovador de creación artística denominado Creative Intelligence Cloud (CIC), que combina redes generativas antagónicas (GAN) y redes neuronales convolucionales (CNN) para explorar cómo el aprendizaje profundo puede aplicarse a la educación artística. El objetivo no es solo generar imágenes, sino comprender el proceso mediante el cual una máquina adquiere lo que podríamos llamar un "conocimiento estético".

Este planteamiento nos sitúa ante una pregunta que atraviesa nuestra investigación sobre el origen del arte: ¿qué significa conocer una obra de arte? ¿Es suficiente con identificar patrones, estilos y estructuras compositivas, o se requiere algo más —una experiencia vivida, un cuerpo que sienta, un símbolo que se habite— que la máquina, por ahora, no posee? La respuesta, como veremos, nos obliga a distinguir entre dos formas de conocimiento: la que se basa en la correlación estadística y la que emerge de la función única del sistema nervioso.

El sistema CIC: anatomía de un aprendizaje visual

El sistema propuesto por Liu y Zhu se articula en torno a dos componentes técnicos fundamentales: las redes generativas antagónicas (GAN) y las redes neuronales convolucionales (CNN). Las GAN, introducidas por Ian Goodfellow en 2014, consisten en dos redes que compiten entre sí: una generadora, que produce imágenes, y una discriminadora, que intenta distinguir las imágenes generadas de las reales. Esta competencia adversarial obliga al generador a mejorar progresivamente la calidad de sus creaciones.

Las CNN, por su parte, son la herramienta mediante la cual el sistema "lee" las imágenes. A diferencia de una red neuronal tradicional, las CNN están diseñadas para procesar datos con una estructura de rejilla —como los píxeles de una imagen— y son capaces de detectar características locales (bordes, texturas, patrones) que luego se combinan en representaciones de mayor nivel (formas, objetos, composiciones). Es, en cierto modo, una imitación algorítmica de cómo la corteza visual humana procesa la información: desde rasgos simples hasta configuraciones complejas.

El sistema CIC se entrena con repositorios masivos de arte, como el dataset WikiArt, que contiene decenas de miles de pinturas etiquetadas por estilo, artista y época. A partir de este entrenamiento, el modelo aprende a descomponer las obras en sus elementos constituyentes: pinceladas, paletas de colores, estructuras compositivas. No "ve" el cuadro como un todo significativo, sino como una matriz de características que puede analizar, comparar y recombinar.

El "lente de datos": cómo la máquina aprende el lenguaje visual

El proceso de aprendizaje del sistema CIC ilustra lo que en otra reseña de este blog hemos denominado el "lente de datos" de la inteligencia artificial. A diferencia de una cámara fotográfica, que captura la luz de una escena real, la IA no ve el mundo directamente. Fragmenta miles de millones de imágenes preexistentes en unidades mínimas de información visual —los llamados tokens— y las asocia con etiquetas lingüísticas (palabras como "impresionista", "retrato", "azul"). De este modo, la máquina aprende un "lenguaje visual del mundo" que le permite, una vez entrenada, moverse de manera fluida y autónoma entre corrientes artísticas, estilos y épocas.

Este proceso tiene implicaciones profundas. Por un lado, permite una versatilidad sin precedentes: un modelo entrenado con WikiArt puede generar una imagen en el estilo de Van Gogh, una fotografía hiperrealista o una abstracción geométrica con la misma facilidad. Por otro lado, plantea la pregunta de qué significa "originalidad" cuando el sistema se entrena con millones de obras existentes. La máquina no inventa desde cero; recombina lo que ha aprendido. Su "creatividad" es, en el fondo, una forma sofisticada de interpolación estadística.

Lo que la máquina no puede aprender: la brecha del symbolon

Aquí es donde el análisis técnico se encuentra con la filosofía. Liu y Zhu demuestran que el sistema CIC es capaz de analizar obras de arte, identificar estilos y generar imágenes nuevas. Pero, ¿comprende lo que hace? La respuesta, desde la perspectiva de nuestra investigación, es negativa. La máquina puede manipular tokens visuales y lingüísticos, pero no habita el símbolo. No hay nadie "dentro" que reconozca o sea reconocido en el acto de creación.

Esta distinción es crucial. El symbolon, en su sentido etimológico originario, es un acto de reconocimiento mediante códigos compartidos. Cuando un artista humano crea, no solo combina formas: proyecta su interioridad, comunica una experiencia, busca un encuentro con el otro. La obra de arte es un puente entre subjetividades. La máquina, por el contrario, produce objetos que pueden ser interpretados como arte, pero no genera el proceso vivo —el sustrato de supervivencia, la dimensión simbólica, la pulsión de completitud— que da sentido a la creación humana.

El sistema CIC, al descomponer las obras en características estadísticas, pierde precisamente lo que hace que una obra sea una obra: su carácter de encuentro. La máquina puede aprender sobre el arte, pero no puede aprender desde el arte. No tiene un cuerpo que sienta, una historia que duela, una necesidad que impulse. Su conocimiento es, en el mejor de los casos, una sombra del conocimiento estético humano.

Una lectura desde el modelo S/Y/C

El estudio de Liu y Zhu ofrece un caso concreto donde se puede observar el funcionamiento de la Ley de coherencia biológica S/Y/C que constituye el núcleo de nuestra investigación (Mallo, 2023, 2025, 2026a, 2026b). Las tres dimensiones de nuestra tesis se despliegan aquí con una claridad poco común, precisamente porque el contraste con la máquina nos permite ver lo que en el humano damos por sentado.

La dimensión S (Supervivencia) se manifiesta en la utilidad del sistema CIC. Su objetivo declarado es mejorar la educación artística, proporcionar retroalimentación automatizada, ampliar el acceso a herramientas de creación. Esta es una función adaptativa legítima: la tecnología, como la cerámica neolítica, se adopta cuando resuelve un problema práctico. Pero la supervivencia del sistema no es la del organismo que lo utiliza. La máquina no necesita aprender arte para sobrevivir; el humano que la usa, sí. La IA es una herramienta de supervivencia cultural, no un sujeto que sobreviva a través de ella.

La dimensión Y (Symbolon) es el punto de mayor contraste. El sistema CIC manipula símbolos —etiquetas, estilos, representaciones visuales— con una eficacia impresionante. Pero no habita el símbolo. No hay un reconocimiento mutuo, una comunidad de códigos compartidos vividos. La máquina puede clasificar una pintura como "impresionista", pero no sabe lo que significa ser impresionista, no ha sentido la luz del sur de Francia, no ha discutido con un académico sobre la pincelada. El symbolon humano es vivido; el de la máquina, simulado.

La dimensión C (Completitud) se manifiesta en la búsqueda de la máquina por cerrar una forma, por generar una imagen coherente. Las GAN, en particular, operan mediante un proceso de convergencia: el generador y el discriminador compiten hasta alcanzar un equilibrio, un punto en el que las imágenes generadas son indistinguibles de las reales. Esta convergencia es una forma de completitud algorítmica. Pero es una completitud sin anhelo, sin la pulsión biológica que en el humano busca integrar las partes en un todo coherente. La máquina "cierra" una forma porque su función de pérdida se lo indica, no porque necesite dar sentido a su experiencia.

La Filosofía Quirúrgica nos invita a aplicar un corte analítico preciso al estudio de Liu y Zhu. No se trata de rechazar la tecnología ni de idealizarla, sino de distinguir niveles. En el nivel de la producción técnica, el sistema CIC es una herramienta poderosa que puede ampliar las capacidades del artista y del educador. En el nivel de la experiencia estética, la IA no puede reemplazar la subjetividad encarnada que constituye el corazón del arte. Y en el nivel de la responsabilidad cultural, es necesario un marco ético —como el humanismo tecnológico que hemos discutido en otras reseñas— que garantice que estas herramientas se utilizan para potenciar lo humano, no para sustituirlo.

Consideraciones finales: el arte como encuentro, no como clasificación

El estudio de Yali Liu y Can Zhu tiene el mérito de ofrecer una visión rigurosa y bien documentada de cómo el aprendizaje profundo puede aplicarse a la educación artística. Su sistema CIC demuestra que las máquinas son capaces de analizar, clasificar y generar imágenes con un grado de sofisticación que hace una década habría parecido imposible. Pero, al mismo tiempo, su trabajo revela —quizás sin proponérselo— los límites de ese enfoque.

La máquina puede aprender a reconocer un estilo, pero no a comprender por qué ese estilo importa. Puede descomponer una pincelada, pero no a sentir lo que la pincelada expresa. Puede generar una imagen nueva, pero no a habitar el proceso de creación como un acto de encuentro con el otro.

El arte, en su origen y en su función, no es una clasificación de características visuales. Es un encuentro entre subjetividades encarnadas. Y ese encuentro, por ahora, sigue siendo patrimonio exclusivo de lo vivo. La IA puede ser una aliada en la educación artística, una herramienta que amplíe el acceso y potencie la creatividad. Pero no puede ocupar el lugar del artista, ni del espectador, ni del educador. Porque el arte, como la supervivencia, el símbolo y la completitud, es algo que se vive antes de que se pueda medir.

Ahora bien, ¿qué aporta esta investigación al conjunto de nuestra indagación sobre el origen del arte? Creemos que lo más destacado no es solo lo que revela sobre las capacidades de la IA, sino la comparación que hace posible. Al observar cómo un sistema artificial, organizado por nosotros, descompone y recompone el legado visual de la humanidad, se nos ofrece un espejo inesperado para pensar cómo el propio sistema biológico, hace decenas de miles de años, pudo haber fragmentado, asimilado y reformulado el mundo para crear algo que hoy nos abarca a todos y a todo.

Referencias bibliográficas

Mallo, B. (2025). Arte y biología: Una aproximación neurofilosófica al origen de la experiencia estética. https://www.amazon.com/dp/B0E8Y5WZMK

Mallo, B. (2025). Art and biology: A neurophilosophical approach to the origin of aesthetic experience. https://www.amazon.com/dp/B0E8Y6C2XN

Yu, C., & Liu, Z. (2026). Multi-task convolutional network for artwork analysis and personalized guidance. Scientific Reports, *16*, Article 70478. https://doi.org/10.1038/s41598-026-70478-6

Autor / Author

Bernabé Mallo
Doctor en Filosofía – Universidad del País Vasco / Euskal Herriko Unibertsitatea (UPV/EHU)
Investigador independiente en neurofilosofía, evolución humana y origen del arte.

Bernabé Mallo
PhD in Philosophy – University of the Basque Country / Euskal Herriko Unibertsitatea (UPV/EHU)
Independent researcher in neurophilosophy, human evolution, and the origin of art.

Enlaces / Links

domingo, 20 de septiembre de 2026

Art as Agency: When Things Act Upon Us

Bernabé Mallo

A review of Art and Agency: An Anthropological Theory by Alfred Gell (1998)

Introduction: the art that makes us act

What makes an image, an object, or a building move us, compel us to stop, provoke an emotion, or incite us to act? The traditional answer of Western aesthetics has been: beauty, form, symbolic meaning. The British anthropologist Alfred Gell (1945–1997) proposed a radically different answer in his posthumous work Art and Agency: An Anthropological Theory, published by Clarendon Press in 1998. For Gell, art is not fundamentally a vehicle of meaning nor an object of aesthetic contemplation, but a form of instrumental action: a means of influencing the thoughts and actions of others.

Gell died in January 1997, just a few months before the book saw the light. The text, which remained in a state of "work in progress" according to his editor, nevertheless constitutes one of the most influential legacies of the anthropology of art of the last quarter century. His proposal challenges the most deeply rooted conceptions of the discipline and forces us to rethink what art is, who produces it, and what effects it has on the social world.

This review explores the central ideas of Gell —agency, index, abduction, and the art nexus— and connects them with our research on the origin of art, the S/Y/C model, and Surgical Philosophy.

The critique of passive aesthetics: art as action

Gell starts from an uncomfortable observation for the anthropological and aesthetic tradition: existing theories adopt an overwhelmingly passive attitude towards art. Art is contemplated, interpreted, deciphered, but its capacity to do things in the world is not recognised. Against this passivity, Gell proposes an inversion: art is, above all, action. "Instead of symbolic communication, I place all the emphasis on agency, intention, causation, result, and transformation. I view art as a system of action, intended to change the world rather than to encode symbolic propositions about it."

This reformulation has profound consequences. The anthropology of art should not concern itself with a special class of objects —those we call "beautiful" or "artistic"— but with a category of action: the making of things that influence others. Any object, in the appropriate context, can become an index of artistic agency. What defines art is not its form or its meaning, but its function within a system of social relations.

Agency and index: the fundamental concepts

Gell's theoretical edifice rests on two interconnected concepts: agency and index.

Agency is the capacity to initiate a series of events in the world, to cause effects, to influence the thought and action of others. In the context of art, agency does not reside solely in the human artist who creates the object. The objects themselves —paintings, sculptures, tattoos, idols— mediate and embody social agency. They are "things that act": not in a literal sense, as if they had intentions of their own, but in the sense that they participate in networks of social relations, acting as vehicles for the agency of their creators or users.

The index, in Gell's terminology (borrowed from the semiotics of Charles Sanders Peirce), is the object that mediates agency. An index is something that, in its mere existence and configuration, points towards something else: towards the artist who made it, towards the model it represents, towards the recipient who receives it. Gell argues that art objects are indexes of the agency of those who produced or used them. They are not symbolic signs that refer to conventional meanings; they are traces, vestiges, proofs of an action.

This distinction is crucial. Gell rejects the idea that art is fundamentally a symbol, in the Saussurean sense of an arbitrary sign that refers to a concept. Art does not "say" things; it "does" things. It does not communicate meanings; it produces effects. An idol is not a symbol of the deity; it is an instrument for acting upon the divine world. A tattoo is not an ornament; it is a mark that binds a person to a lineage, to a status, to a system of social obligations.

Abduction and the art nexus: how agency works

If art is action mediated by indexes, how exactly does this mediation work? Gell turns here to another Peircean concept: abduction. Abduction is the inferential process by which, from an index, we infer the agency that produced it. We see a footprint in the sand and deduce that someone walked there. We see a painting and deduce that someone painted it, with certain intentions, in certain circumstances.

For Gell, art operates precisely through this abduction of agency. The object leads us to infer an agency that transcends it: the artist, the model, the recipient. But abduction is not a mere logical deduction; it is a psychological process, often automatic and unconscious, that leads us to attribute agency to the object. Art "captures" our attention, "enchants" us (enchantment), makes us experience the object as if it had its own power.

Gell organises these relations into what he calls the art nexus: a network of relations between four terms: the artist (who produces the index), the index (the object), the prototype (what the index represents or evokes), and the recipient (who experiences the index). Combinations of these terms generate a series of "formulas" that describe the different types of artistic agency. The art nexus is thus an analytical tool for mapping the complex networks of agency that unfold around art objects.

The distributed person and the extended mind

Two final concepts complete Gell's theoretical edifice. The first is the distributed person: the idea that a person's agency is not limited to their body, but extends to the objects they produce, the marks they leave, the things that belong to them. An artist, in this sense, is distributed across their works; their agency continues to act through them long after their death.

The second is the extended mind: Gell suggests that art objects function as extensions of the mind, as external vehicles of cognition and agency. This idea, which anticipates contemporary debates in the philosophy of mind, connects directly with our research on the origin of art and the role of external symbols in cognitive evolution.

Connection with research on the origin of art (S/Y/C)

Gell's theory offers a conceptual framework that illuminates crucial aspects of the Law of Biological Coherence S/Y/C that forms the core of our research (Mallo, 2023, 2025, 2026a, 2026b). His concepts of agency, index, and abduction allow us to reread prehistoric art not as a system of symbolic meanings, but as a device of social action.

The S (Survival) dimension manifests in the instrumental function of art that Gell emphasises. Art objects are, in many ethnographic contexts, tools of social survival: amulets that protect, masks that mediate with spirits, marks that ensure group belonging. Art is not a luxury; it is a technology for acting upon the world.

The Y (Symbolon) dimension requires qualification in light of Gell. For our thesis, symbolon is the act of recognition through shared codes. Gell, by rejecting the symbolic dimension of art, emphasises instead indexicality: the object as a trace of an agency. But both aspects are not mutually exclusive. The Gellian index can be the materialisation of symbolon: the object is the physical support through which symbolic agency is exercised. An idol not only acts; it is also recognised as the presence of the deity, and that recognition is a symbolic act.

The C (Wholeness) dimension manifests in the abduction of agency that the viewer performs. The art nexus, with its four terms, is a system that seeks coherence: the viewer integrates the index into a network of relations that give it meaning. The artist's agency, that of the prototype, that of the recipient, are articulated in a whole that transcends the mere materiality of the object. This integration is a form of wholeness: the object is understood as part of a broader system of action.

Surgical Philosophy invites us to make a precise analytical cut in Gell's work. It is not about accepting his rejection of aesthetics without more, but about recognising that his approach opens a field of analysis that traditional aesthetics had neglected: art as action, as agency, as power. At the same time, we must resist the temptation to reduce art to agency. The symbolic dimension, which Gell excludes, is also constitutive of artistic experience. Surgical Philosophy allows us to distinguish levels: the level of agency (what art does), the level of the symbol (what art means), and the level of experience (what art makes us feel). The three are irreducible to one another.

Critique and reception

Gell's work has generated intense debate. His critics, such as Howard Morphy, argue that his approach diverts attention from human agency by attributing agency to the objects themselves. The aesthetic and semantic properties that Gell excludes from his analysis are, for Morphy, integral to the understanding of art as a mode of action. Robert Layton maintains that Gell fails to explain the distinctive ways in which art objects extend the agency of their creators, and that his rejection of Saussurean semiotics is untenable.

Other critiques point to Gell's ethnographic bias, with his emphasis on Asia and the Pacific, and his scant attention to Western art. The internal coherence of his theory has also been questioned, especially regarding the relationship between agency and indexicality.

Despite these critiques, the impact of Art and Agency has been profound. The book has reformulated the anthropology of art, inspired new research on materiality, agency, and power, and built bridges between anthropology, archaeology, art history, and philosophy.

Final considerations: art as action and as encounter

Alfred Gell's posthumous work is one of those rare contributions that transform a field by changing the questions we ask. His proposal to study art as agency, as index, as action, forces us to look at objects not only for what they represent or signify, but for what they do in the social world.

For our research on the origin of art, Gell's theory offers a valuable complement to the neuroscientific and philosophical perspectives we have explored. Art, in its origin, was not only a symbolic expression or an aesthetic response; it was also —and perhaps above all— a tool of social action. Cave paintings, body ornaments, ivory figurines: all of them were indexes of an agency that sought to influence the world, the spirits, the others.

But art is not only agency. It is also encounter. The abduction that Gell describes, that process by which we attribute agency to an object, is an act of recognition. And recognition is the essence of symbolon. Art objects speak to us because we recognise in them the presence of a subjectivity, an intention, a history. That presence is not only an effect of power; it is also an invitation to encounter. And in that encounter, art fulfils its deepest function: that of connecting the living with the living, the living with the dead, humans with the sacred.

Gell taught us to see art as action. We add: art is action because it is encounter. And encounter is, at its root, the fullest form of agency and of symbol.

References

Gell, A. (1998). Art and agency: An anthropological theory. Clarendon Press. https://philpapers.org/rec/GELAAA

Layton, R. (2003). Art and agency: A reassessment. Journal of the Royal Anthropological Institute, *9*(3), 447–464. https://doi.org/10.1111/1467-9655.00158

Mallo, B. (2025). Arte y biología: Una aproximación neurofilosófica al origen de la experiencia estética. https://www.amazon.com/dp/B0E8Y5WZMK

Mallo, B. (2025). Art and biology: A neurophilosophical approach to the origin of aesthetic experience. https://www.amazon.com/dp/B0E8Y6C2XN

Morphy, H. (2009). Art as a mode of action: Some problems with Gell's Art and Agency. Journal of Material Culture, *14*(1), 5–27. https://doi.org/10.1177/1359183508100006

Thomas, N. (1998). Foreword. In A. Gell, Art and agency: An anthropological theory (pp. vii–xiii). Clarendon Press.

Autor / Author

Bernabé Mallo
Doctor en Filosofía – Universidad del País Vasco / Euskal Herriko Unibertsitatea (UPV/EHU)
Investigador independiente en neurofilosofía, evolución humana y origen del arte.

Bernabé Mallo
PhD in Philosophy – University of the Basque Country / Euskal Herriko Unibertsitatea (UPV/EHU)
Independent researcher in neurophilosophy, human evolution, and the origin of art.

martes, 13 de octubre de 2026

The Art That Teaches Itself: Methodologies of Autonomous Learning in Algorithmic Creation

Bernabé Mallo

A review of Qian Wang's study (2024) in ACM Transactions on Multimedia Computing, Communications, and Applications: Learning-based Artificial Intelligence Artwork: Methodology and Future Directions

Introduction: the machine that becomes its own teacher

The dance of GANs: creator and critic in perpetual competition

From adversarial learning to diffusion models: the technical evolution

Data repositories: the fuel of learning

The question of originality: what does it mean to create?

A reading from the S/Y/C model

Final considerations: art as encounter, not as optimisation

References

Enlaces / Links

El arte que se enseña a sí mismo: metodologías del aprendizaje autónomo en la creación algorítmica

Bernabé Mallo

Una reseña del estudio de Qian Wang (2024) en ACM Transactions on Multimedia Computing, Communications, and Applications: Learning-based Artificial Intelligence Artwork: Methodology and Future Directions

Introducción: la máquina que se convierte en su propio maestro

La danza de las GAN: creador y crítico en competencia perpetua

Del aprendizaje adversarial a los modelos de difusión: la evolución técnica

Los repositorios de datos: el combustible del aprendizaje

La cuestión de la originalidad: ¿qué significa crear?

Una lectura desde el modelo S/Y/C

Consideraciones finales: el arte como encuentro, no como optimización

Referencias bibliográficas

Enlaces / Links

jueves, 1 de octubre de 2026

The Algorithmic Eye: How Artificial Intelligence Learns to See Art

Bernabé Mallo

A review of the study by Yali Liu and Can Zhu (2025) in Scientific Reports: The use of deep learning and artificial intelligence-based creative teaching in art education

Introduction: when the machine looks at a painting

The CIC system: anatomy of a visual learning process

The "data lens": how the machine learns the visual language

What the machine cannot learn: the symbolon gap

A reading from the S/Y/C model

Final considerations: art as encounter, not as classification

References

Enlaces / Links

El ojo algorítmico: cómo la inteligencia artificial aprende a ver el arte

Bernabé Mallo

Una reseña del estudio de Yali Liu y Can Zhu (2025) en Scientific Reports: The use of deep learning and artificial intelligence-based creative teaching in art education

Introducción: cuando la máquina mira un cuadro

El sistema CIC: anatomía de un aprendizaje visual

El "lente de datos": cómo la máquina aprende el lenguaje visual

Lo que la máquina no puede aprender: la brecha del symbolon

Una lectura desde el modelo S/Y/C

Consideraciones finales: el arte como encuentro, no como clasificación

Referencias bibliográficas

Enlaces / Links

domingo, 20 de septiembre de 2026

Art as Agency: When Things Act Upon Us

Bernabé Mallo

A review of Art and Agency: An Anthropological Theory by Alfred Gell (1998)

Introduction: the art that makes us act

The critique of passive aesthetics: art as action

Agency and index: the fundamental concepts

Abduction and the art nexus: how agency works

The distributed person and the extended mind

Connection with research on the origin of art (S/Y/C)

Critique and reception

Final considerations: art as action and as encounter

References

Enlaces / Links

Vistas de página la semana pasada