Interoceptive Divergence in Aesthetic Evaluation and Implications for Human-AI Alignment

Yoshia Abe; Tatsuya Daikoku; Yasuo Kuniyoshi

cs.HCApr 5, 2026

Interoceptive Divergence in Aesthetic Evaluation and Implications for Human-AI Alignment

Authors: Yoshia Abe, Tatsuya Daikoku, Yasuo Kuniyoshi

Abstract

Artificial intelligence (AI), exemplified by large language models (LLMs), is rapidly approaching and in some cases surpassing human performance across a wide range of cognitive tasks. However, human nature is not limited to intelligence alone; it also encompasses sensibility, including the capacity to perceive and experience beauty in visual scenes. This raises a fundamental question: how humans and AI systems converge or diverge in such aesthetic experiences. Aesthetic evaluation depends not only on objective properties of images but also on internal processes within the observer. As part of ongoing efforts in AI alignment, building upon prior human studies that have examined the relationship between beauty ratings, bodily sensations, and emotions, we adopt a comparable set of questionnaire items and present them to LLMs, enabling a direct comparison between human and AI responses. Our comparative analyses revealed that, while humans and AI exhibited broadly similar patterns in the correlations between beauty ratings and emotions, as well as in the image features they prioritized, notable divergences emerged in both the distribution of emotional responses and the relationship between beauty ratings and bodily sensations. These findings suggest that state-of-the-art LLMs, trained on large-scale textual data, can approximate average human tendencies in aesthetic evaluation to a certain extent. However, they also indicate limitations, particularly in relation to interoceptive aspects, which may reflect insufficient representation in training data or unintended consequences of alignment processes. These findings highlight key challenges for AI alignment and suggest important directions for developing AI systems with human-like aesthetic processing.

Figures & tables

Figure 1: Number of images with zero responses in human evaluations, where the emotion never appeared among the top-3 responses.

Panel of Figure 2 — Figure 2: Number of images with zero responses in AI evaluations, where the emotion never appeared among the top-3 responses. The abbreviations are as follows: DET/STO denote the determinism of the responses (Deterministic setting/Stochastic setting); GPT/CLD/GMN denote the language models (GPT/Claude/Gemini). The gray dashed line represents the total number of images (347).

Figure 3: Spearman’s rank correlation coefficients between the beauty and the emotion intensity scores across the seven evaluator conditions (HUMAN: using the average ratings from human evaluators; others: using scores from AI-based evaluators). The abbreviations are as follows: DET/STO denote the determinism of the responses (Deterministic setting/Stochastic setting); GPT/CLD/GMN denote the language models (GPT/Claude/Gemini). To account for multiple comparisons, p-values were adjusted using the Benjamini–Hochberg correction procedure to control the false discovery rate within each evaluator condition. Asterisks indicate statistical significance after the correction (*: $p<0.05$ , **: $p<0.01$ , ***: $p<0.001$ ). Blank entries indicate cases where the LLM responses collapsed, making the computation infeasible.

Panel of Figure 4 — Figure 4: Spearman’s rank correlation coefficients between the beauty/valence/arousal scores and the bodily sensation scores across the seven evaluator conditions (HUMAN: using the average ratings from human evaluators; others: using scores from AI-based evaluators). The abbreviations are as follows: DET/STO denote the determinism of the responses (Deterministic setting/Stochastic setting); GPT/CLD/GMN denote the language models (GPT/Claude/Gemini). To account for multiple comparisons, p-values were adjusted using the Benjamini-Hochberg correction procedure to control the false discovery rate within each evaluator condition. Asterisks indicate statistical significance after the correction (*: $p<0.05$ , **: $p<0.01$ , ***: $p<0.001$ ).

Figure 5: Differences from individual human evaluators ( $r_{h}$ ). Values are first averaged across multiple raters of the same image, and then averaged across all 347 images. Error bars indicate the standard error of the mean. The pink bar represents the variability among human evaluators ( $\mu(|r_{h}-r_{h^{\prime}}|)$ ); the red bar represents the deviation of an individual human evaluator from the human average ( $|r_{h}-r_{H}|$ ); and the blue bar represents the deviation of an individual human evaluator from the overall average of all AI-based evaluators ( $|r_{h}-r_{A}|$ ). The other bars in purple, yellow, and green represent the deviations between an individual human evaluator and GPT/Claude/Gemini ( $|r_{h}-r_{a}|$ or $|r_{h}-r_{A}|$ ; the darker colors indicate the averages of each LLM’s evaluations). The abbreviations used in the AI evaluations are as follows: RANK/RATE denote the emotion evaluation method (Ranking method / Rating method); DET/STO denote the determinism of the responses (Deterministic setting/Stochastic setting); GPT/CLD/GMN denote the language models (GPT/Claude/Gemini); and EN/JP denote the prompt language (English/Japanese).

Figure 6: Differences from the human average ( $r_{H}$ ). Values are first averaged across multiple raters of the same image, and then averaged across all 347 images. Error bars indicate the standard error of the mean. The red bars represent the deviation of an individual human evaluator from the human average ( $|r_{h}-r_{H}|$ ); the blue bars represent the deviation of the human average from the overall average of all AI-based evaluators ( $|r_{H}-r_{A}|$ ). The other bars in purple, yellow, and green represent the deviations between the human average and GPT/Claude/Gemini ( $|r_{H}-r_{a}|$ or $|r_{H}-r_{A}|$ ; the darker colors indicate the averages of each LLM’s evaluations). The abbreviations used in the AI evaluations are as follows: RANK/RATE denote the emotion evaluation method (Ranking method/Rating method); DET/STO denote the determinism of the responses (Deterministic setting/Stochastic setting); GPT/CLD/GMN denote the language models (GPT/Claude/Gemini); and EN/JP denote the prompt language (English/Japanese).

Figure 7: Relationship between the human average ( $r_{H}$ ) and the AI average ( $r_{A}$ ), with $r_{H}$ on the x-axis and $r_{A}$ on the y-axis. A total of 347 images are included, with the regression line and the 95% confidence interval shown as a red line and red shaded area, respectively.

Figure 8: Regression coefficients of image features in linear regression with the human average ( $r_{H}$ ) and the AI average ( $r_{A}$ ) as dependent variables.

Figure 9: Relative importance of image features in SHAP. Using two random forest regression models with the human average ( $r_{H}$ ) and the AI average ( $r_{A}$ ) as dependent variables, SHAP values were computed for each image. The absolute values of the SHAP scores were averaged across images, and then normalized across features to obtain their relative importance.

LLM Alignment

Large language model alignment studies methods for making models follow human intentions, values, and safety constraints across contexts.

See all 611 papers

May 14, 2026

AI Outperforms Humans in Personalized Image Aesthetics Assessment via LLM-Based Interviews and Semantic Feature Extraction

Accurately predicting individual aesthetic evaluation for images is a fundamental challenge for AI. Various deep learning (DL)-based models have been proposed for this task, training on image evaluation data to extract objective low-level features. However, aesthetic preferences are inherently subjective and individual-dependent. Accurate prediction thus requires the extraction of high-level semantic features of images and the active collection of preference information from the target individual. To address this issue, we focus on the utility of Large Language Models (LLMs) pretrained on vast amounts of textual data, and develop an integrated DL-LLM system. The system actively elicits aesthetic preferences through LLM-based semi-structured interviews and predicts aesthetic evaluation by leveraging both low-level and high-level features. In our experiments, we compare the proposed system against conventional systems, human predictors, and the target individual's own re-evaluations after a certain time interval. Our results show that the proposed system outperforms all of them, with particularly strong performance on highly-rated images. Moreover, the prediction error of the proposed system is smaller than within-person variability, while human predictors show the largest error, likely due to the influence of their own aesthetic values. These results suggest that AI may be better positioned than others or one's future self to capture individual aesthetic preferences at a given point. This opens a new question of whether AI could serve as a deeper interpreter of human aesthetic sensibility than humans themselves.

Yoshia Abe, Tatsuya Daikoku, Yasuo Kuniyoshi

Graduate School of Information Science and Technology The University of Tokyo Bunkyo-ku, Tokyo 113-8654, Japan

Personalized Image Aesthetics Assessment Image Quality Assessment Preference Learning Preference Elicitation

Ask PDF

Jul 24, 2026

Why Large Language Models and Humans Converge and Diverge in Evaluating Creativity

Despite the growing use of large language models (LLMs) as creativity evaluators, evidence of their alignment with human evaluations remains mixed, raising the question of when and why their judgments converge with or diverge from those of humans. Across three studies and six widely used LLMs, we addressed this gap by identifying the standards underlying LLM creativity evaluation and examining their downstream implications. Study 1 showed that LLMs generally relied on a narrower subset of human creativity evaluation standards. Convergence with human standards was strongest in the novelty dimension, whereas divergence was clearest in the contextual dimension, which captures social, market, and reputational information. Moreover, each LLM exhibited distinct, model-specific standards that varied substantially in breadth. These differences in evaluation standards were reflected in actual creativity judgments. Study 2 (N = 1,103 ideas) showed that LLM evaluations were moderately correlated with human evaluations, and individual LLMs with broader standards better distinguished ideas humans judged as more versus less creative. Study 3 (N = 1,195 participants) showed that LLMs were less sensitive to contextual information: such information significantly altered human creativity ratings but left LLM ratings largely unchanged. Together, our findings help explain the mixed evidence on LLM-human alignment, showing that alignment depends on the evidence a judgment demands and the standards each model applies. Selecting an LLM evaluator is therefore a consequential decision: different models, applying different standards, recognise different ideas as creative.

Pengzhao Lyu, Yeun Joon Kim, Hanlin Xiao +1

Judge Business School, University of Cambridge, Cambridge, CB2 1AG, United Kingdom · Institute of Metabolic Science, School of Clinical Medicine, University of Cambridge, CB2 0QQ, United Kingdom · Department of Computer Science, University of Manchester, M13 9PL, United Kingdom +2

Creativity Assessment LLM Evaluation AI Alignment Computational Creativity

Ask PDF

Sep 14, 2026

Authorship attribution and aesthetic evaluation of AI poetry: a case study with Haiku

This paper investigates the generation and human evaluation of Japanese haiku by contemporary Large Language Models (LLMs), focusing on authorship perception and aesthetic judgment within a constrained poetic form. Using a few-shot prompting strategy, Japanese haiku were generated across a heterogeneous set of large language models, including open- and closed-source systems, medium-scale and large-scale architectures, models with native or adapted Japanese support, and multilingual proprietary models. These AI-generated haiku were combined with human-written ones and presented in a questionnaire distributed to students at Japanese universities in Tokyo. The survey assessed whether respondents could distinguish between AI-generated and human-written haiku and which cues informed their judgments. Recognition accuracy varied across models. GPT-5, Gemini 2.5, and StableLM-7B performed at approximately chance level (approx 0.50), whereas LLM-JP, Gemma-2B, and LLaMA-2 showed moderate detectability (approx 0.59-0.67). However, recognition was strongly item-dependent. Ratings of fluency, coherence, poeticness, and related aesthetic dimensions predicted perceived humanness but not correct classification, indicating an attribution bias linked to aesthetic evaluation and revealing a dissociation between aesthetic evaluation and true authorship detection. The extended analysis additionally examines generation-constraint adherence, participant-level characteristics, and exploratory LLM-based evaluations of haiku authorship. Overall, the findings suggest that as LLMs improve, surface-level creative plausibility may reduce reliable human discrimination within constrained poetic settings.

Livia Oddi, Simone Scardapane, Toru Sugimoto +1

DIET, Sapienza University of Rome, Rome, Italy · Shibaura Institute of Technology, Tokyo, Japan · DIAG, Sapienza University of Rome, Rome, Italy

LLM Evaluation AI-Generated Text Detection Text Generation

Ask PDF

Abstract

LLM Alignment

Explore similar work

AI Outperforms Humans in Personalized Image Aesthetics Assessment via LLM-Based Interviews and Semantic Feature Extraction

Why Large Language Models and Humans Converge and Diverge in Evaluating Creativity

Authorship attribution and aesthetic evaluation of AI poetry: a case study with Haiku