What happens when algorithms, aesthetic theory, machine learning, and architectural design are treated not as separate fields, but as parts of one long conversation? “Beyond Data-Driven Aesthetics,” an exhibition by MIT Architecture alumnus and researcher Alexandros Haridis, takes that question into the gallery.
On view at the MIT Keller Gallery through June 30, the exhibition examines 20th- and 21st-century attempts to turn computing into a medium for creative production and aesthetic judgment in architecture and the applied arts. Its central move is direct: take ideas often trapped in papers, books, formulas, and systems, then make them visible through physical installations and interactive visualizations.
AI did not invent the question of aesthetic judgment
The exhibition grew from three intersecting lines of research. The first came from Haridis’s PhD work in design and computation in the MIT Department of Architecture around 2022, when data-driven machine learning systems such as ChatGPT and Stable Diffusion were moving quickly into public debates about creativity, design, aesthetic judgment, and high-profile art auctions.
Those debates often made the questions feel new. Haridis’s research points in another direction. The problem of whether computation can participate in creation and evaluation has a much longer history across the 20th century.
One example in the source is the 1956 Dartmouth Summer Research Project, described as a foundational event for AI. There, creation and evaluation processes were named as one of seven key dimensions of human intelligence that future AI research should address.
That matters because it changes the frame. The exhibition is not simply asking whether current AI systems can produce attractive images or useful designs. It is asking how architecture, design, mathematics, computer science, philosophy, and aesthetic theory have repeatedly tried to describe, formalize, or test judgment itself.
Rule-based design still matters
A second influence on the exhibition comes from design computation and shape grammars. These areas investigate the relationship between human insight and computation through rule-based methods, rather than through purely data-driven learning.
That distinction is important in the context of contemporary AI. Machine learning systems are one path into computational aesthetics, but they are not the only one. The exhibition also looks at how rules, comparisons, and interpretative structures can shape design thinking.
Haridis also draws on interpretative studies of aesthetic theories connected to figures such as Samuel Taylor Coleridge, Oscar Wilde, and John von Neumann. These studies ask whether ideas of aesthetic value and comparison found in philosophical and literary texts can reveal both possibilities and limits in contemporary digital computation and AI for architecture and design.
In plain terms, the exhibition treats older theories as active material. They are not background decoration. They become ways to test what computation can and cannot explain when design moves beyond function alone.
Turning abstract systems into spatial experience
The third line of research behind “Beyond Data-Driven Aesthetics” involves design, fabrication, and data visualization as tools for interpreting mathematical concepts, algorithms, and “black box” machine-learning systems.
Across disciplines, researchers use reconstruction and visualization to make computational systems easier to understand. The source connects this work to neural network visualization in computer science, software reconstruction, digital fabrication in architecture, and curatorial practice.
The exhibition uses a similar logic. It asks what the most important idea in a book or research paper is, then translates that idea into a visual, spatial, and experiential format. Software reconstruction, physical making, and data visualization become methods for turning dense academic material into stories in space.
This is the exhibition’s key contribution: design is not only the subject being studied. It is also the method of interpretation.
Five ways to read computational aesthetics
The exhibition is organized around five thematic areas: Aesthetic Measure, Aesthetic Guidelines, Algorithmic Aesthetics, Aesthetic Appropriation, and Aesthetic Novelty. Each theme works as a selective window into a computational approach to aesthetic judgment drawn from a specific publication, either a book or a research paper.
The titles come from concepts central to those publications. “Measure,” for example, refers to mathematician George Birkhoff’s work in the 1930s to quantify aesthetic value mathematically. “Novelty” examines how the machine learning system AICAN judges generated images according to a theory in cognitive aesthetics that balances familiarity and deviation from known artistic styles.
Together, the five areas show that computation can approach aesthetic judgment from different angles:
- Aesthetic Measure considers the effort to quantify aesthetic value.
- Aesthetic Guidelines points to structured ways of directing design judgment.
- Algorithmic Aesthetics focuses on computation as part of aesthetic production and assessment.
- Aesthetic Appropriation addresses how computational approaches engage existing aesthetic material.
- Aesthetic Novelty looks at how generated images may be judged through familiarity and deviation.
The result is not a single answer to whether machines can judge beauty or taste. It is a map of approaches, each with its own assumptions about what judgment means and how design can make those assumptions legible.
Why this matters for architecture and the built environment
Haridis describes “Beyond Data-Driven Aesthetics” as both a research exhibition and an ongoing platform. Its focus is how computational systems participate in aesthetic judgment, generation, and transformation across architecture and the applied arts.
One central question is computational evaluation beyond purely performative or functional requirements. The source applies this to many design spaces, including buildings, structural forms, and everyday products.
That point is especially relevant for architecture and engineering because design decisions are not only about whether something performs a task. They also affect how people experience spaces and objects they inhabit and use. Haridis is interested in how computation, whether rule-based or data-driven, can help designers and engineers understand what contributes positively to human experience.
The exhibition also raises a broader question about research communication. Through software reconstruction, visualization, and physical making, it turns opaque computational systems into artifacts that are more legible, tangible, and experiential. That opens a path beyond the traditional preoccupation with mechanizing “beauty” or “taste.” It also suggests that scholarship itself can move through spatial, visual, and public-facing formats.
For the future of data-driven aesthetics, the exhibition’s message is measured but significant. The newest AI systems may have intensified the debate, but the deeper questions were already there: how judgment works, what computation can reveal, and how design can make complex ideas understandable.