On self-relevance as the core of aesthetic judgement and its machine boundary
In the two preceding texts, I maintained two points. First, that aesthetic comprehension rests upon regimes of relevance, and that misunderstanding—between cultures, between humans and machines—is a misalignment of those regimes. Second, that the divergence between human and machine attention is measurable and concentrates where human relevance ceases to be formal and becomes social, semantic, or vital. The second text concluded with a caveat: the social aspect of this divergence is well-documented; the self-referential aspect—the vital weight, that which touches the self—remained, in my argument, more an inference than an established result.
This text addresses that component. The question is specific: what makes a work not merely appear beautiful or interesting to us, but actually matter? My thesis, supported by recent empirical research, is that the best candidate to explain this leap is self-relevance: the degree to which a work enters into a relationship with the observer's memory, identity, and history. Not the form itself, not mere processing fluency, not novelty: self-relevance. And I maintain that this mechanism is precisely what a system without a self cannot replicate, not due to a transient technical limitation, but because it lacks that in relation to which something could be self-relevant.
Precision of scope from the outset marks the difference between rigour and enthusiasm. The available evidence is promising but not yet conclusive. It strongly suggests that self-relevance explains the most intense and memorable forms of aesthetic judgment. It does not yet demonstrate that all aesthetic valuation depends upon it. I maintain that distinction throughout the text.
1. The third stratum, once more
In the first text of this series, I distinguished three strata of aesthetic relevance. Formal salience—what stands out perceptually: colour, contrast, rhythm, composition. Historical-contextual pertinence—authorship, technique, tradition, intention; that which is only perceived by knowing the work's provenance. And vital weight—that which touches memory, identity, attachment, vulnerability; that which makes something not only informationally rich but existentially significant.
The first stratum is well-studied and, in part, modelable: salience systems reasonably predict where a human looks for physical reasons. The second is the classic territory of art history and aesthetics. The third is the most difficult to operationalise and therefore the easiest to treat with romantic vagueness. The contribution of recent research is to have begun to measure it.
The term that the literature offers to approach the third stratum is self-relevance, and the behavioural phenomenon that allows for its study is the self-reference effect: the memory advantage that appears when we process information in relation to ourselves. It is a robust and long-standing effect in cognitive psychology: we remember words, images, and facts that we have related to our own person better than those we have processed neutrally. The hypothesis that runs through this text is that this same mechanism operates at the heart of intense aesthetic judgement.
2. What the evidence shows, without exaggeration
Four studies mark the current state of the matter. We read them together because their value lies in how they correct one another.
The first is a null result, and it is important precisely for that reason. Kasdan and Belfi (2020) investigated the self-reference effect in music: they asked participants to judge brief instrumental fragments according to how much they liked them—a condition linked to the self—versus other conditions, and then measured recognition. No self-referential advantage appeared. Simple musical 'liking' was not sufficient to produce the effect. This result establishes a boundary: not every aesthetic judgement recruits the self, and superficial 'liking' is not the same as deep self-relevance. Starting here protects the argument from becoming dogma.
The second, Lee et al. (2023), worked with painting. They compared three encoding conditions—beauty judgement, personal memory, and a symmetry control—and measured subsequent recognition. The result was partial and revealing: the beauty judgement did not generally match the memory advantage of the self-referential condition, but paintings with extreme beauty judgements were remembered as well as those encoded in relation to the self. The convergence between beauty and self-reference appears, above all, when the aesthetic reaction is intense. Not all beauty touches the self; the beauty that strikes us most deeply, does.
The third is the strongest in favour of the thesis. Vessel et al. (2023) measured whether self-relevance predicts aesthetic appeal, using both real works and synthetic works personalised through style transfer based on a biographical questionnaire for each participant. The result: self-relevance strongly predicted appeal, explained more variance than eighteen combined image traits—28% versus 8%—and the effect was not mediated by familiarity. That is to say: it is not that we like what we recognise; it is that we are attracted to what concerns us. The stimulus control in this work is excellent and it is the piece of evidence that best sustains the view that self-relevance is not an epiphenomenon of other variables.
The fourth, Salgues et al. (2024), articulated the sharpest formulation. Working with a hundred representational paintings and separating the conditions of beauty, personal memory, and emotion into two studies, they found that aesthetic judgement produces a memory advantage, albeit smaller than self-reference or emotion, and that emotional intensity predicts both recognition and judgements of beauty and self-relevance. Their explicit conclusion: the appraisal of self-relevance could be a basic mechanism of aesthetic judgement. Not identical to emotion, not identical to beauty, but the functional common denominator of both.
To this is added a recent finding that reinforces the conclusion. Smith, De Brigard, and Marsh (2025) showed that spontaneous autobiographical recollection—not induced, but emerging on its own during contemplation—increases the degree to which a work moves us. Personal meaning does not need to be prompted: when it emerges of its own accord before the work, it amplifies the aesthetic experience.
Read in conjunction, the five works draw a clear and prudent figure. Self-relevance does not explain every aesthetic judgment—the null result of Kasdan and Belfi precludes this. But it explains, better than any alternative, the intense, memorable, and transformative forms of judgment: those in which a work ceases to merely please us and begins to matter to us.
3. Why self-relevance is distinctive
There are other explanations for aesthetic value. Confronting them helps to identify what self-relevance possesses that they do not.
Novelty mobilises attention and curiosity, but it does not explain the idiosyncrasy of taste: why the same new work moves one person and leaves another cold. Processing fluency—the ease with which we perceive something—predicts part of the pleasure, but Vessel's data are uncomfortable for reducing beauty to fluency: the mediation by familiarity was small and left a large direct effect of self-relevance intact. Expertise modulates evaluation, but the studies most favourable to self-relevance were conducted with non-expert participants: the mechanism operates even without specialised interpretive capital.
Self-relevance is distinctive for three reasons. It connects perception, memory, emotion, and self-awareness into a single mechanism, which other hypotheses treat separately. It explains the idiosyncrasy of taste: why intense aesthetic value is so personal. And it generates stronger, testable predictions: if a work becomes more relevant to the self, it should be liked more, remembered better, and elicit more emotion; this is precisely what the evidence finds. The other hypotheses capture real components of the aesthetic episode; self-relevance better explains how they are integrated.
Operationalisation to avoid remaining in the vague. I call vital weight the degree to which an aesthetic object enters into a relationship with the subject's self-referential system and acquires value for their life, not merely for their perception. Observable through measurable variables: declared self-relevance, autobiographical evocation, congruence with identity, personally attributed affective intensity, and memory advantage compared to non-self-referential conditions. It is not an ineffable essence. It is a construct with indicators.
4. The machine frontier
I arrive at the point that connects this series to its underlying question. If vital weight rests upon self-relevance and self-relevance is the relationship of a work to the observer's self, then a system that lacks a self lacks that in relation to which something could be self-relevant.
This is not a statement regarding what machines cannot do today due to a lack of power. It is a structural statement. A system can model with increasing precision which works humans tend to like, which traits correlate with high ratings, and which images generate the most interaction. It may even predict, sometimes better than the subject themselves, what they will like. But predicting what someone will like is not the same as caring oneself. Self-relevance is not a property of the work that the system can detect; it is a relationship between the work and a biography, an identity, a vulnerability. Without the second term of that relationship, the operation has nowhere to anchor itself.
Precision here avoids falling into the errors I critiqued in previous texts. I am not asserting that machines lack an inner experience—that is an open question which serious research treats with caution and which I need not resolve. I am asserting something more modest and firm: that the specific mechanism which evidence points to as the core of intense aesthetic judgement, self-relevance, requires a self with autobiographical memory and identity, and that current systems operate through proxies of that mechanism—frequencies, correlations, aggregated preferences—which approximate the result without reproducing the relationship.
It is the same structure I described in the case of the agent that organised a birthday party and relegated the most important person in the user's life. The system could retrieve traces; it could not prioritise existences. Applied to art: a system can retrieve which works resemble those you liked; it cannot know which one will shatter you because it touches something that only you know you carry within. Self-relevance is that knowledge, and it is not transferable.
5. What this does not authorise us to conclude
Three closing remarks, for the sake of rigour.
This does not authorise the conclusion that art made by machines lacks aesthetic value. Aesthetic value, according to the argument itself, does not lie in the object or in the one who produces it, but in the relationship with the one who receives it. A work generated by a system can deeply touch a human: the vital weight is provided by the observer, not the producer. What the system cannot do is experience that weight, not prevent it for another.
This does not authorise the conclusion that self-relevance is already fully demonstrated as the core mechanism. The evidence favours it as an explanation for intense forms of aesthetic judgement; it leaves open whether it explains all of them. The direct corpus is small, samples tend to consist of young students, and the precise relationship between emotion and self-relevance remains unresolved. I state this because the rigour of this line of reasoning depends upon not overstating the case.
This does not authorise the conclusion that the boundary is eternal. I assert that current systems lack a self in the relevant sense. I do not assert that it is impossible to build future systems with persistent autobiographical memory, identity, and something analogous to vulnerability. If such a thing were to arrive—and nothing indicates it is near—the nature of the question would change. Today the boundary is real; I do not declare it impassable, I describe where it lies.
6. Conclusion
Three conclusions.
The first is empirical. Recent evidence points to self-relevance as the best candidate to explain the core of aesthetic judgement: not form, not fluency, not novelty, but the degree to which a work enters into a relationship with the observer's memory and identity. The evidence is robust for intense forms of aesthetic judgement and remains open for the rest.
The second is conceptual. Vital weight—the third stratum of relevance I have been defending in this line of reasoning—can be operationalised through self-relevance, with measurable indicators, without resorting to ineffable essences. It is what transforms 'I like this' into 'this concerns me'.
The third concerns the machine boundary. Self-relevance requires a self in relation to which something is relevant. Current systems lack this and operate through proxies that approximate the result without reproducing the relationship. They can predict what we will like; they cannot experience that something matters to them. The difference is not one of degree of sophistication: it is one of the nature of the operation.
The self-reference effect fundamentally denotes a simple asymmetry. A work of art matters to us when it touches us; it touches us when it enters into a relationship with who we are; and who we are is a narrative, not a collection of data. The machine can read the data. For the moment, we provide the narrative. That is the distance between predicting what we will like and sharing why it matters to us.
On the open conversation
This text is the third in the series on relevance regimes that I have been developing in NeuroArt: Cognitive Surplus. It develops the third stratum—vital weight—which the two previous texts identified but did not substantiate. The next planned step addresses generative art as a territory where human and machine share a process, even if they do not share meaning.
Should anyone wish to intervene from the perspectives of neuroaesthetics, the psychology of memory, the philosophy of mind, or artistic practice, this notebook remains open.
Sources
Kasdan, A., & Belfi, A. M. (2020). Investigating a self-reference effect in musical aesthetics. Experimental Results, 1, e6. https://doi.org/10.1017/exp.2020.6
Lee, J., et al. (2023). The beauty and the self: A common mnemonic advantage between aesthetic judgment and self-reference. Cognitive Neuroscience. https://doi.org/10.1037/cns0000345
Vessel, E. A., et al. (2023). Self-relevance predicts the aesthetic appeal of real and synthetic artworks generated via neural style transfer. Psychological Science. https://doi.org/10.1177/09567976231188107
Salgues, S., Jacquot, A., Makowski, D., Tahar, C., Baekeland, J., Arcangeli, M., Dokic, J., Piolino, P., & Sperduti, M. (2024). Self-reference and emotional reaction drive aesthetic judgment. Scientific Reports, 14, 19699. https://doi.org/10.1038/s41598-024-68331-9
Smith, A. P., De Brigard, F., & Marsh, E. J. (2025). Aesthetic experience is supported by spontaneous autobiographical memory recollection. Memory & Cognition, 54, 1037–1051. https://doi.org/10.3758/s13421-025-01795-w
Vessel, E. A., Starr, G. G., & Rubin, N. (2013). Art reaches in: The default mode network and intense aesthetic experience. (Activation of the default mode network in response to intensely moving works.)
Arcangeli, M., Dokic, J., & Sperduti, M. The Beautiful, the Sublime, and the Self. (Philosophical framework of the SublimAE project on the involvement of the self in aesthetic experience.)
Leder, H., & Nadal, M. (2014). Ten years of a model of aesthetic appreciation and aesthetic judgments. British Journal of Psychology. (Model of aesthetic appreciation as a multi-stage episode.)
Esteban Ruiz, J. A. What matters and what is relevant. Public notebook at juanesteban.art, 2026.
Esteban Ruiz, J. A. Where the machine looks. Public notebook at juanesteban.art, 2026.
Esteban Ruiz, J. A. Art as Structural Surplus: Toward a Relational Ontology Beyond Human Authorship (V2.3). PhilArchive and Zenodo, 2026.