Artificial Intelligence Studies in Society, Science and Systems, Year 2025 Issue 1

Artificial Intelligence: On Language Education, Linguistic Inequality, and Linguistic Justice

Mehmet Dursun ERDEMDeniz DEMİRYAKAN — 2026-07-16

This article examines the transformation brought about by artificial intelligence in applied linguistics and language education in its social, pedagogical, and linguistic dimensions. The central concern is not how advanced these technologies have become, but which languages are brought to the fore in language education, which speakers are placed at the center, which forms of knowledge are recognized as legitimate, and for whom learning opportunities are expanded. On the one hand, artificial intelligence offers important possibilities by increasing learners’ participation in class, providing individualized support, accelerating feedback processes, and opening new pathways for learning, especially for refugees, displaced communities, Indigenous peoples, and students marginalized in linguistic terms. Recent research on Google Translate and generative AI tools shows that these systems can support language learning, expand possibilities for expression, strengthen self-confidence, and foster social participation. On the other hand, the same field also creates conditions for the reproduction of dominant languages, standardized views of language, and centralized cultural assumptions. The marginalization of local languages, dialects, accents, and alternative forms of knowledge stands among the most significant risks of AI-supported language education. In this context, Birhane’s concept of “algorithmic colonialism” and Meighan’s notion of “colonialingualism” provide an important conceptual basis for explaining the centers from which linguistic legitimacy, educational acceptance, and the value of knowledge are defined. The article emphasizes that a more just AI order requires ethical design, collaborative production, cultural representation, equality of access, and careful attention to linguistic diversity. It therefore positions artificial intelligence not simply as a technical innovation, but as a field of power that shapes the direction, scope, and sense of justice within language education.

From Tradition to Dataset: Digital Archiving of Traditional Artistic Heritage and the Ethical and Epistemological Consequences of Its Use as Training Data for Artificial Intelligence

Sema CİVELEK — 2026-07-16

Abstract

This research addresses the profound ethical and epistemological problems arising from the process of high-resolution scanning of traditional art heritage works to transform them into digital archives and the use of these archives as training data for artificial intelligence (AI) models. The study argues that behind the promise of "eternal preservation" and "universal access" frequently offered by digitization projects lies an epistemological transformation that inevitably severs objects from their historical, cultural, and ritual contexts, reducing them to mere visual "style" or "pattern." When a miniature, sculpture, or textile is transformed into RGB pixel values, feature vectors, and statistical distributions, the semantic layers it carries (religious symbolism, social hierarchy, mythological narrative, the sanctity of the material) are typically marginalized or completely lost in the dataset metadata. AI models (GANs, Diffusion Models) trained on this "decontextualized style" can imitate traditional forms, but this imitation occurs in a semantic void, within a state of "meaningless grace." The article analyzes this process through the concepts of "epistemological violence" and "cultural flattening." The proposed methodological framework examines the digitization chain in three stages: 1) Selection and Framing: Which works are digitized, at what resolution and with what cropping? 2) Transformation into Data and Labeling: Under which categories (e.g., "Islamic Art," "16th century") and with what keywords is the visual material defined? 3) Model Training and Reproduction: How do the biases in the dataset solidify and become naturalized in the model's outputs? The analysis, concretized through a fictional yet realistic case study of the "OttomanMiniatures-10k" dataset and the "MinyatürGAN" model trained on it, demonstrates that while AI successfully learns the iconographic language of, for example, a sultan's portrait (the halo, posture, symbols), it completely ignores the function of that language in legitimizing power or visualizing a religious-traditional hierarchy. Consequently, this paper highlights the urgent need in digital heritage management not only for technical improvement but also for developing a contextual integrity-preserving, polyphonic, and transparent epistemology. For AI to learn tradition not merely as a visual resource but as a meaningful knowledge system will only be possible through interdisciplinary collaboration, critical dataset curation, and the enrichment of semantic metadata.

The Role of the Aesthetic Value Function in Evolutionary Art Generation

Mustafa GÜNAY — 2026-07-16

This research examines the application of Evolutionary Algorithms (EAs) in the context of pictorial composition and style optimization, focusing on the fundamental challenge of defining and quantifying the "Aesthetic Value Function" (AVF). EAs, inspired by natural selection and genetic operators, hold the potential for exploration within complex, multi-dimensional artistic search spaces that are difficult to define using traditional analytical methods. However, modeling the AVF—which is needed to guide this process—in an objective, generalizable, and computable form poses a fundamental theoretical and practical challenge. This study systematically classifies approaches to defining the AVF (statistical, psychophysical, learning-based, and interactive) and analyzes how each differently "deforms" the artistic search space and guides the exploration process. Under the proposed methodological framework of "guided exploration," the research investigates how EAs can be configured not merely for simple fitness maximization but to generate diverse, surprising, and yet aesthetically interesting solutions within a multimodal search space. Using a fictional yet realistic experimental setup within a parametric image generation system, the performance of Genetic Algorithms employing different AVF definitions (e.g., fractal dimension, Color Harmony Scale, deep learning-based aesthetic scorers, and human-evaluated interactive evolution) was compared. The findings reveal that while quantitative AVFs accelerate the search, they lead to noticeable convergence and a loss of diversity in the generated works. In contrast, interactive evolution and multi-criteria optimization open up richer and more unforeseen areas of exploration but introduce challenges of scalability and subjectivity. In conclusion, this paper argues that the success of EAs for artistic optimization fundamentally depends on the design of a "guidance function" that captures the aesthetic intent of the artist or designer while preserving opportunities for discovery and surprise. A truly creative system should be able to map meaningful and original paths within the infinite space of possibilities, rather than merely finding an optimum point.

AI-Supported Image Education in Drawing: Adaptive Learning Environments - Personalized Systems Modeling Student Style Development and Generating Feedback

Leyla ÖNEN — 2026-07-16

This research examines the integration of artificial intelligence (AI)-supported adaptive learning environments into painting education to overcome the limitations of the traditional studio instruction model. The study investigates the theoretical foundations, technical possibilities, and pedagogical implications of an integrated system that continuously monitors and models individual student style development while generating personalized visual, verbal, and practical feedback. The central argument is that AI can be conceptualized not only as a technical tool but also as a learning partner—a dynamic "digital mentor"—capable of understanding the student's artistic journey across cognitive, psychomotor, and affective dimensions. The proposed system combines deep learning-based image analysis (Convolutional Neural Networks), temporal modeling of student actions (Recurrent Neural Networks/LSTMs), and recommender system techniques for personalization. The methodology, described through a fictional yet realistic prototype named "ArtTutorAI," outlines the system's three core components: (1) The Style Diagnostic Module: Analyzes student works in light of art historical datasets and pedagogical principles to extract a quantitative profile focusing on composition, color, value, brushwork, and stylistic tendencies. (2) The Developmental Pathway Modeler: Compares works over time to map the pace of progress, recurring challenges, and the emerging artistic "voice." (3) The Adaptive Feedback Generator: Based on diagnostic and modeling outputs, creates tailored exercise suggestions, visual references, corrected composition examples, and constructive critiques suited to the student's current level, goals, and learning style. The findings indicate that such a system could enhance student confidence and self-efficacy by providing a continuous and consistent evaluation framework, while offering instructors in-depth analytical data to strategicize their interventions. However, significant pedagogical and ethical concerns also emerge, including the risk of excessive quantification, the neglect of artistic subjectivity, data privacy issues, and the danger of "algorithmic authority" stifling creative autonomy. In conclusion, this paper argues that the success of AI-assisted adaptive painting education depends on establishing a delicate balance between technological advancement and human-centered, critical pedagogy principles. The system should be a guide that directs the student, not a referee that makes decisions for them.

Ontological and Ethical Boundaries in AI-Based Image Restoration

Abdulkadir Özdemir — 2026-07-16

This research aims to critically analyze the practice of filling historical and material gaps through predictive inpainting algorithms—especially those based on deep learning models—in the field of artificial intelligence (AI)-based painting restoration, from ontological and ethical perspectives. The study argues that these technologies should be understood not only as a technical achievement but also as a practice that reshapes the ontology (conditions of existence) of cultural heritage and the production of historical knowledge. Advanced models (e.g., GANs, Diffusion Models) possess the capacity to produce plausible and visually consistent completions based on the existing context of a damaged or missing artwork. However, this “predictive archaeology” calls into question fundamental concepts such as authenticity, intention, materiality, and historical uncertainty. The article first synthesizes traditional restoration ethics (Brandi, 1963; Riegl, 1903) with computer vision and deep learning literature (Pathak et al., 2016; Iizuka et al., 2017). It then presents a methodological framework, examining how algorithmic decision-making processes, dataset biases, and stochastic outputs limit or distort the “possible worlds” of restoration. By proposing conceptual tools such as “digital palimpsest,” “speculative authenticity,” and “algorithmic nostalgia,” the study focuses on the nature of the historical representations AI can create. The findings reveal that AI-based completion does not merely “restore” a work to its former state but transforms it into a new hybrid object laden with contemporary data and aesthetic assumptions about the past. This process transfers the traditional authority of the restorer to the algorithm, while confronting the viewer with an illusion of “historical probability.” Consequently, this paper highlights the urgent need for a new ethical framework that acknowledges ontological ambiguity and prioritizes transparency and participation. AI restoration should be regarded not as a repair tool but as a field of “applied philosophy of history” that raises profound philosophical questions about the interpretation and representation of cultural memory.