AI Research

AI Outperforms Humans in Personalized Image Aesthet... | AI Research

Key Takeaways

AI Outperforms Humans in Personalized Image Aesthetics Assessment via LLM-Based Interviews and Semantic Feature Extraction This research addresses the challe...
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.

Paper AbstractExpand

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.

AI Outperforms Humans in Personalized Image Aesthetics Assessment via LLM-Based Interviews and Semantic Feature Extraction
This research addresses the challenge of predicting how specific individuals evaluate the aesthetics of images. While traditional AI models often rely on objective, low-level image features—such as brightness or texture—these methods struggle to account for the subjective, personal nature of aesthetic taste. The authors developed an integrated system that combines deep learning with Large Language Models (LLMs) to actively interview users about their preferences, allowing the AI to incorporate high-level semantic and contextual information into its predictions.

How the System Works

The proposed system functions through two main components: an interview process and a prediction engine. First, an LLM-based interview system conducts a semi-structured conversation with a participant. Two AI agents work in parallel: an "Interviewer" that asks questions and an "Analyzer" that interprets the user's responses to build a profile of their aesthetic values.
Second, the prediction system uses this profile to identify high-level features—such as narrative, emotion, or cultural context—that are meaningful to that specific individual. These high-level features are combined with low-level image data processed by a deep learning model. A machine learning module then integrates these inputs to generate a personalized aesthetic score for any given image.

Key Findings

The researchers tested their system against conventional deep learning models, other LLM-based predictors, and human participants. The results showed that the proposed system outperformed all other methods. Notably, the AI demonstrated its strongest performance when predicting scores for images that users rated highly, suggesting that the system is particularly effective at capturing the specific, subjective reasons why a person finds an image appealing.

AI as an Interpreter of Taste

A significant finding of the study is that the AI’s prediction error was smaller than the natural variability found within a single person’s own evaluations over time. In contrast, human predictors—who were asked to guess the preferences of others—showed the largest errors, likely because they were biased by their own personal aesthetic values. These results suggest that AI may be uniquely positioned to act as a precise interpreter of human aesthetic sensibility, potentially understanding an individual's preferences better than other humans or even the individual themselves at a different point in time.

Comments (0)

No comments yet

Be the first to share your thoughts!