The research team, Yunhe Su of the Department of Oriental Painting at Hongik University’s College of Fine Arts and Xiaowen Feng of the Graduate School of Information at Yonsei University, set out to address a persistent bottleneck in digital art education and aesthetic computing. Traditional assessment of drawing composition has relied either on the subjective judgment of individual evaluators or on single-modal computational analysis that examines only the visual characteristics of an artwork. Both approaches carry well-known limitations. Human scoring varies from grader to grader and cannot easily be applied across thousands of student submissions in large online courses. Rule-based and purely visual computational models, meanwhile, capture only part of the picture: they cannot incorporate the verbal context, such as assignment descriptions, evaluation criteria, or rubric language, that human assessors naturally bring to bear when deciding whether a composition succeeds.

The core innovation of MM-DLN lies in its architecture, which fuses two powerful but very different branches of modern artificial intelligence into a single scoring pipeline. On the visual side, the system employs a Vision Transformer, or ViT, a neural network architecture that has become a mainstay of computer vision by dividing an image into small patches and processing them through attention mechanisms that learn which regions of the image matter most for a given task. Here, the ViT extracts spatial and compositional features from the drawing itself: the arrangement of elements, the distribution of visual weight, the relationships between foreground and background, and other layout properties that determine whether a composition feels balanced, dynamic, or disjointed.

On the textual side, the model uses LLaMA3, the Large Language Model Meta AI, version 3, to analyze verbal input associated with each drawing. This textual stream can include descriptions of the assignment, the criteria against which the work should be judged, or other rubric-related language. By processing this text, the language model builds a representation of what a good composition should look like in the specific context of the task, providing the scoring system with an explicit standard of comparison rather than forcing it to infer quality criteria purely from visual examples. This is a meaningful departure from earlier automated assessment systems, which typically operated on images alone and therefore had no way to encode the evaluative expectations that instructors articulate in words.

Bringing these two streams together is a cross-modal attention mechanism, a technique that allows the model to dynamically align information from the visual and textual domains. In practical terms, the attention mechanism lets the network learn which visual features of a drawing are relevant to which textual criteria, and vice versa, creating a fused representation that is richer than either modality in isolation. This fused representation then feeds into a regression-based scoring head, the final component of the network that outputs a numerical quality score for the composition. The entire system is trained end to end, meaning that the visual encoder, the language encoder, the fusion layer, and the scoring head are all optimized together to minimize the difference between predicted and expert-assigned scores.

One of the most intriguing aspects of the system is its interpretability. Automated scoring models are often criticized as black boxes: they produce a number, but they cannot explain why. MM-DLN addresses this by highlighting the attention regions that the model focused on when generating its score. In effect, the system can show which parts of a drawing drove its judgment, offering a window into its scoring logic. For educators, this could transform an automated grade from an opaque verdict into a teaching tool, showing students precisely where their compositions draw the model’s attention and, by extension, where the strengths and weaknesses of their layout lie. Interpretability of this kind is increasingly seen as essential for deploying AI in educational settings, where trust and transparency matter as much as raw accuracy.

The performance gains reported in the study are substantial. Compared with the best-performing baseline model in their experiments, MM-DLN reduced the mean absolute error, a standard measure of how far predictions deviate from true scores, by 23.9 percent. It also improved the Pearson correlation coefficient, which captures how closely the model’s scores track expert judgments on a continuous scale, by 10.2 percent. Together, these two metrics indicate that the multimodal approach is not merely marginally better than existing methods but represents a meaningful leap in both prediction accuracy and alignment with expert-driven aesthetic evaluation. In a field where the ground truth is inherently fuzzy, since even human experts disagree about artistic quality, gains of this magnitude suggest that the textual modality carries real information that purely visual models have been leaving on the table.

Equally important is the model’s robustness. The researchers report that performance remained stable when the system was presented with noisy inputs or drawings in stylistically diverse styles. This matters because real-world art education does not deal in clean, standardized data. Student work spans an enormous range of techniques, media, and personal styles, and scanned or photographed submissions often contain artifacts, inconsistent lighting, and other imperfections. A scoring system that only works on idealized inputs would be of limited practical value. The stability of MM-DLN across noisy and heterogeneous inputs suggests that the attention-based architecture is learning genuinely compositional features rather than overfitting to superficial characteristics of a particular dataset or artistic tradition.

The implications extend well beyond the art classroom. Automated composition assessment could support computerized evaluation systems at scale, providing consistent feedback in massive open online courses, standardized art examinations, and portfolio screening processes where human graders are scarce or expensive. In the emerging field of aesthetic computing, which seeks to formalize and compute notions of beauty and design quality, the study offers a template for how multimodal inputs can be combined to capture evaluative judgments that resist simple rule-based encoding. The work also speaks to a broader trend in artificial intelligence: the recognition that many real-world judgments, from medical diagnosis to content moderation to artistic critique, are inherently multimodal, requiring the integration of perception and language to reach conclusions that either channel alone cannot support.

There remain, of course, open questions. Aesthetic judgment is culturally situated, and a model trained on particular rubrics and expert populations will reflect those standards; whether MM-DLN generalizes across cultures, age groups, and artistic traditions is a matter for future research. The authors note that the study received no external funding and declare no competing interests, and the work was published open access, making the full technical details available to the research community. As digital art education continues to expand and AI-assisted tools become routine in creative fields, systems like MM-DLN point toward a future in which the machine is not replacing the art teacher but extending the teacher’s reach, offering every student a consistent, explainable first read on the oldest problem in visual art: how to arrange the elements of a picture so that they hold together as a whole.

Subject of Research: Multimodal deep learning for automated assessment of drawing composition quality

Article Title: A multimodal deep learning-driven model for assessing drawing composition quality

Article References: Su, Y., & Feng, X. (2026). A multimodal deep learning-driven model for assessing drawing composition quality. Journal of Big Data. https://doi.org/10.1186/s40537-026-01551-0

Image Credits: AI Generated

DOI: 10.1186/s40537-026-01551-0

Keywords: multimodal learning, deep learning, drawing assessment, Vision Transformer, LLaMA3, cross-modal attention, automatic scoring, aesthetic computing, art education, digital art, interpretability, Journal of Big Data

The historical core of the argument rests on a deceptively simple idea: that an artwork can be a set of instructions. Conceptual art, which crystallized as a movement in the 1960s, prioritized the idea behind a work over the finished object. Sol LeWitt famously declared that the concept is central while execution is merely a perfunctory affair, and that the idea becomes a machine that makes the art. Around the same time, artists associated with the Fluxus collective, including John Cage, George Brecht, La Monte Young and Yoko Ono, developed event scores: brief, colloquial instructional statements intended to be performed by anyone, anywhere. Because these text-based pieces could be executed by any participant, they opened the artwork to endless interpretation and variation, effectively distributing the act of creation beyond the named artist.

Xie and Wong show how this instructional, rule-driven logic prefigured generative and algorithmic art. Citing Philip Galanter’s definition of generative art, in which the artist cedes partial or full control to an external system such as an algorithm, the authors note that the legitimacy of such systems was grounded in foundations laid by conceptual art. Fluxus founder George Maciunas argued that the artist need only create a concept or method, after which the form can be created independently of him. In the 1960s, the Japanese avant-garde artist Yamaguchi even supplied manufacturing instructions to factories to produce his works. What has changed, the study contends, is the participatory capacity of the external system: where assistants and factories once carried out the concept, generative AI now does so at unprecedented scale and speed.

The technical trajectory the authors describe is recent and rapid. Text-guided digital image generation advanced significantly with the introduction of generative adversarial networks in 2014, followed by Google’s Deep Dream in 2015, which hinted at the possibility of rendering conceptual directives in visual form. A watershed came in January 2021 with OpenAI’s release of CLIP, which demonstrated an unprecedented capacity for associative retrieval between textual and pictorial modalities. These techniques were subsequently integrated into diffusion models optimized for synthesizing images from semantically anchored descriptions. The upshot is that users with little knowledge of deep learning pipelines can now outsource image-making entirely to AI assistants configured through minimal language, a capability that reshapes who, or what, can complete an artwork.

This is where the study raises its sharpest concern. Yoko Ono described her instruction paintings as unfinished works, recasting art from a finished object into an unfinished process to be completed by others, and she sought to erase the sacred distance between object and participant. Audience participation became a hallmark of contemporary art, theorized by Claire Bishop as a breakdown of the artist as autonomous producer and by Nicolas Bourriaud as relational aesthetics, in which art happens in the relations between elements, human and non-human alike. But when a generative model is trained on Ono’s instructions and produces visual outputs on demand, the human participant’s substantive engagement with the underlying concept can become virtually non-existent. The participant risks becoming a passive provider of impetus, an incidental vessel transmitting directives rather than an active determiner of conceptual outcomes, reducing participation to a hollow illusion of democratic inclusion.

An instructive counterexample comes from the German artist Mario Klingemann, whose work Appropriate Response trained a model on 60,000 internet quotes to generate aphorisms, delivered to kneeling audience members through a prayerful trigger device. Because people interpreted the generated sentences against their own experiences, the machine’s outputs functioned like fortune-telling or divination. The authors observe that the roles of human and machine in this feedback loop are reversed relative to Ono’s practice: the artist’s trained model now occupies the position Ono once held, issuing instructions to the audience. More troubling, they argue, the false contingency created by computation bestows a quasi-divine authority on the machine that no human artist ever commanded, since audiences and artists always knew they belonged to the same species.

To escape this impasse, the paper turns to posthumanist theory. Drawing on Stefan Herbrechter’s emphasis on the biological and microbiological entanglement with nonhumans, and on the insight from the Human Genome Project that humans, like all organisms, can be understood as encoded informational entities, the authors argue for positioning humans and machines as equal conversational partners and co-creators of meaning. In this reconfigured triangular framework of artist, audience and technological mediator, the AI system becomes a creative agency serving both artist and audience, performing tasks that were previously theirs alone. Slovenian philosopher Slavoj Žižek’s description of the uncanny experience of the human mind directly integrated into a machine underscores why new participatory frameworks must abandon the illusion of the autonomy of personhood.

The study also distinguishes carefully between two divergent conceptual legacies that carry different implications for AI art. LeWitt’s wall paintings required assistants to physically execute his written instructions at specific times and places, with results varying according to who carried them out; his linear trajectory from conception to completion suggests that AI image generation legitimately realizes intentionally seeded ideas. Ono, by contrast, held that mental realization alone constitutes art, challenging any strict correlation between computational output and original conceptual intention; from her perspective, AI finalization of imagined works is a redundant, even counterintuitive step. The authors point to Casey Reas’s Software Structures from 2004, in which programmers translated LeWitt’s instructions into code that generated wall drawings, as an early demonstration that machine systems could substitute for human draftsmen while contributing their own contingency factors to the outcome.

Where, then, does human creativity reside? Using Nelson Goodman’s distinction between autographic works, which are singular and authenticity-based like paintings, and allographic works, which are repeatable and concept-based like scores, and Kwastek’s analysis of interactive art as an autographic technical system built on constitutive rules, the authors argue that in text-to-image generation human ingenuity lies in the conceptual components embodied in language, not in the final digital artifact. The human mission is no longer to output a finite art object but to complete a framework of constitutive rules. Prompt engineering becomes the crucial site of this creativity: humans adaptively craft prompts that leverage neural network training and syntax modulation, and researcher Oppenlaender characterizes the process as iterative refinement, haltable by the human collaborator at any stage. Because current text-to-image outputs remain imperfect and often require post-production, final outcomes still depend on supplemental human judgment rather than full computational autonomy.

The paper closes by reconciling these tensions. Rather than negating conceptual art’s dematerialized impulses, generative AI signals their expanded reformulation within a technoculture of co-authorship between human and computational domains, echoing Jean Tinguely’s 1950s Méta-matic drawing machines, which satirized human veneration of machines even as they prefigured automated creation. By revisiting the conceptual art legacy and its insistence on ideas over objects, Xie and Wong contend, artists and theorists can build frameworks that support emergent AI practices and drive a continued rethinking of posthuman creative partnerships, ensuring that the delegation of agency to machines enriches rather than hollows out the human act of making meaning.

Subject of Research: The relationship between 1960s conceptual art and generative AI art practices, focusing on authorship, participation and posthuman creativity

Article Title: From instructions to intelligent synthesis: conceptual art, distributed creativity, and the posthuman turn in AI-powered artistic practices

Article References: Xie, Y., & Wong, A. L. (2026). From instructions to intelligent synthesis: conceptual art, distributed creativity, and the posthuman turn in AI-powered artistic practices. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03316-y

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03316-y

Keywords: conceptual art, generative AI, Yoko Ono, Sol LeWitt, text-to-image generation, prompt engineering, posthumanism, participatory art, authorship, distributed creativity, digital art, AI & Society