Alexandros Haridis: Beyond Data-Driven Aesthetics

Simone Brauner. One could imagine aesthetics as a whole orchard. Many different fruits hang there, beautiful ones, ugly ones, sublime, kitschy, grotesque. And the orchard also includes its soil, its climate, its care, its pests, its harvest, its market. Beauty would then be the apple in the orchard, one particular fruit loaded with a long mythology, charged with hopes and projections.

Does this account, at least in part, for the fact that the figures in your lineage consistently say aesthetic rather than beauty? Or is there another reason? How do you distinguish between aesthetics and beauty?

Alexandros Haridis. This is an interesting observation and points directly to a key distinction in this research. I use the terms aesthetic judgment or evaluation as opposed to beauty or taste because they draw attention to a process rather than a particular outcome or verdict. By aesthetic judgment, I mean the experience or process of evaluating, comparing, expressing a preference, or discerning something meaningful in an art or design object. Beauty, or ugliness, may be one possible result of that process, but it is certainly not defining or determining that process.

In that sense, I agree with your orchard metaphor. Beauty is only one fruit in the orchard and is one that is in many ways molded by custom or social conditioning and changes with the times. Aesthetic experience encompasses a much wider range of responses. This includes novelty, familiarity, surprise, association with past or present experience, discernment between the symbols or ideas expressed in one or another object, ambiguity in meaning, transformation, and so forth. When seen in this way, the issue is less about defining “what is beautiful?” so much as it’s about: How do people make judgments about or discern between the things they perceive in ways other than attending purely to a design function or an end-goal, and can those judgments be described, represented, or even modeled from a computational perspective?

Historically, philosophical accounts of aesthetics were indeed framed in terms of beauty which presumably required a special kind of educated taste. For example, in the 18^th century figures such as Immanuel Kant and David Hume formulated their treatises on aesthetic judgment from a taste-based point of view. There is a long-standing precedent for theorizing beauty as a category of aesthetic experience (along with the related category of the sublime) in the philosophical literature as well as in architecture and art criticism. My own interest in this subject, however, comes from a different direction.

Around 2022, while finishing my PhD in design and computation at MIT’s Department of Architecture, I watched the rapid emergence of data-driven machine learning systems such as ChatGPT and Stable Diffusion into public discussions about creativity, aesthetics, design, and even high-profile art auctions. It became increasingly clear to me that many of the questions presented publicly as “new” in relation to AI had a much longer history. Questions about how machines might create, evaluate, compare, or make selections among alternatives had already occupied researchers throughout the twentieth century especially in computational design theory and practice, my own disciplinary area. Each case study, from Birkhoff to AICAN, proposes a particular visual and logical language for describing how judgments can be made computationally specifically with objects of architecture or the applied arts.

Alexandros Haridis. Moreover, the exhibition was influenced by research in design computation and shape grammars that investigates relationships between human insight and computation through rule-based methods, rather than purely data-driven learning. More recent interpretative studies of aesthetic theories— drawing from figures such as Samuel Taylor Coleridge, Oscar Wilde, and even John von Neumann—have been especially important to me (Stiny, 2022). These studies examine whether theories of aesthetic value and comparison articulated in philosophical and literary texts may reveal possibilities or limitations in contemporary models of digital computation and AI in architecture and design.

Figure 1: Dartmouth Hall Commemorative Plaque for the 1956 Dartmouth Summer Research Project on Artificial Intelligence.

What interested me most was that aesthetic judgment could be understood not simply as a matter of taste, but as a form of intelligence. In fact, at the 1956 Dartmouth Summer Research Project (Figure 1), an event commonly associated with the birth of artificial intelligence, creation and evaluation processes in humans were explicitly identified as one of seven key dimensions of human intelligence. This perspective resonates strongly with the exhibition.

Interestingly, this is also one reason I find Kant still relevant. In a recent article on Kant's concept of “free play,” I argue that this understanding of aesthetics as a distinct form of intelligence is already implicit in his philosophical system (Haridis 2026). In the third Critique, the Critique of the Power of Judgment (1790), he describes aesthetic judgment as a distinct and autonomous form of judgment – “an entirely special faculty for discriminating and judging” (§1, 5:204). What is striking is that he treats this faculty as fundamental rather than secondary. He places it alongside logical reasoning (e.g., scientific understanding) and practical judgment (ethics or moral understanding) to complete a tripartite view of how human beings make sense of the world.

From this perspective, the central question is not whether a machine can recognize beauty. The more interesting question is whether the processes through which we evaluate, compare, interpret, and assign significance to things can themselves be represented computationally. Computing has traditionally been associated with logic, calculation, and problem solving. The research behind the exhibition asks what happens when computational systems are used to model another dimension of human intelligence: our capacity for aesthetic judgment.

The term aesthetics is also useful here in another way. When it is seen in terms of experience or perception, it implies a process of engaging with objects we encounter with our senses, that is to say, with objects that are real and present, and that we can see, touch, hear, or experience spatially. Beauty, for good or ill, is often applied more broadly to things that may not be perceptually present at all—take, for example, mathematical theorems, abstract ideas, or our internal imaginary thoughts. I’m much more interested in the former: the things we encounter through our senses, and which are present around us, especially those deliberately shaped through acts of design in our built or unbuilt environments.

Simone Brauner. Your exhibition translates abstract models of aesthetic judgment into sensory experience. George D. Birkhoff's polygons become a sculpture with lighting that highlights exactly the boundary lines. Vera Molnár's guidelines become a tablet application that visitors execute themselves. Lillian F. Schwartz's morphing becomes a video monitor. How did you arrive at these curatorial decisions?

Alexandros Haridis. The approach of the exhibition is to ask what exactly in a particular research paper or book captures its most salient idea and then use design to interpret that idea in a visual, spatial, and experiential format. Drawing on design techniques such as software reconstruction, physical making, and data visualization, the exhibition takes written sources that are dense with algorithmic ideas, abstract concepts, and mathematical formulas, and translates them into stories in space that include interaction, material form, and digital visualization. How can we make the invisible/abstract in a paper or book into something that’s visible or tangible?

The case studies I selected for the exhibition are organized around five thematic areas: Aesthetic Measure, Aesthetic Guidelines, Algorithmic Aesthetics, Aesthetic Appropriation, and Aesthetic Novelty. Each theme functions as a selective “window” into a distinct computational approach to aesthetic judgment drawn from a specific publication—a book or research paper. The titles of these themes are derived from concepts central to each publication.

For example, “measure” refers to mathematician George Birkhoff’s 1933 publication of Aesthetic Measure (Harvard University Press), a key reference on formalist aesthetics of the 20th century. In Birkhoff’s system, aesthetic value is defined as a ratio between order (O) and complexity (C) in an object that belongs to a particular class, e.g., the class of polygons, class of ornaments, class of vases. The physical sculpture in Figure 2 is a wall-mounted relief presenting all 90 polygonal forms from Birkhoff’s original plates of the class of polygons in Aesthetic Measure (Birkhoff 1933, Chapter II). The polygons are manufactured as negative reliefs from a white substrate and arranged in a grid, ordered in terms of their aesthetic score—from highest (top left) to lowest (bottom right).

The subtractive, negative relief emphasizes the boundary lines of the polygons as physical edges that cast shadows under lighting, make viscerally present a key psychophysiological premise of Birkhoff's aesthetic measure: that the “effort of attention,” which is necessary for the act of perception, increases in proportion to the complexity C of the polygon; in this case the complexity is measured by the number of distinct lines that contain all sides of a polygon (Birkhoff 1933, 34). The easier a boundary is to trace visually, the simpler its shadow in the physical sculpture; the more irregular the polygon, the greater the perceptual effort.

Figure 3: Diagram of Birkhoff’s formulation of the aesthetic experience in terms of automatic eye movements and correlative sensory input leading to his formula.

Alexandros Haridis. This embodied reading of complexity as attention was central to late 19th- and early 20th-century psychophysics and is illustrated in Birkhoff's own diagram in Figure 3 as the automatic adjustment of the eye traversing a polygon’s edges. This translation demonstrates how interpretive reconstruction can expose the perceptual assumptions embedded in a mathematical formula by materializing complexity as a measurable yet embodied experience.

I followed analogous interpretative approaches for the other case studies, too, choosing design techniques that represented key ideas in a publication in a meaningful way. One design technique that’s used in two case studies, in Vera Molnár (Aesthetic Guidelines) and Lillian Schwartz (Aesthetic Appropriation), is software reconstruction, a technique that researchers, scholars, and curators often use for reconstructing historic computer systems in formats accessible to contemporary audiences.

Figure 4: Six physical prints, framed, derived from algorithmic implementations of three of Molnár’s works. Each work is paired with its underlying procedure or guidelines. Photo: Beyond Data-Driven Aesthetics (2026).

Alexandros Haridis. In her article, “Toward Aesthetic Guidelines for Paintings with the Aid of a Computer” (Leonardo, 1975), Molnár describes how new works of art and design can emerge through procedures in which simple geometric shapes are successively altered into more elaborate arrangements—by hand or with the aid of digital computers. These procedures are computer-aided “guidelines”: step-by-step heuristics and numerical parameters that can be translated into code but also executed manually. Guidelines are iterative and experimental; they are meant to facilitate active attention and judgment as a work develops. The intention here was to allow visitors to use a digital pen and tablet that implements in a modern programming language the guidelines derived from three works by Molnár (Figure 4): (Dés)Ordres ((Dis) Orders) (1974), Quatre éléments distribués au hasard (1950), Signes sans Signification B (1975). Each implementation guides users through a sequence of operations that generates variations of a pattern to expose the “experimental method” for art-making and art-appreciation that Molnár describes in her writings.

In an analogous way, I was interested in Lillian Schwartz’s article “Computers and appropriation art: the transformation of a work or idea for a new creation” (Leonardo, 1996). In this article she explores concepts of image and identity transformation that are quite common in the visual arts, but she does so through early computer graphics techniques. Unlike rule-based abstract compositions, the works she describes show how aesthetic value can emerge by appropriating existing images from known artworks–such as Duchamp’s Nude Descending a Staircase or Leonardo da Vinci’s Mona Lisa–into novel yet recognizable forms.

Figure 5: Two original works by Lillian F. Schwartz on loan from the Henry Ford Museum of American Innovation and a video monitor with a digital reconstruction of Schwartz’s “Mona-Leo” studies. Photo: Adrian Yu (2026).

Alexandros Haridis. One of the computational techniques Schwartz applied in her artistic practice is called image morphing or interpolation. It’s used in particular in the Mona-Leo studies that she describes in the article “Morphing the three faces of Mona: the decision-making steps Leonardo used to create his Mona Lisa” (Computers & Graphics, 1995). In the exhibition, this particular technique is shown in two mediums: an original cover design for a book and a video monitor (Figure 5). The cover design is from ‘The Computer Art Book’ (c 1992), by Lillian Schwartz and Laurens R. Schwartz, which I obtained on loan from the Ford Museum of American Innovation. The video monitor reconstructs a computer-aided morphing algorithm that references the transformation of Isabella, Duchess of Aragon, into the Mona Lisa (c. 1503–1506), a process often associated with Da Vinci’s own facial features.

Also on display is second original piece I obtained on loan from the Henry Ford called Homage to Duchamp (Nude Ascending Staircase) (c 1975), by Lillian Schwartz with Robert J. Tatem. This piece is a painting that appropriates Duchamp’s idea that a specific placement of shapes or figures in two dimensions can suggest motion. Schwartz in collaboration with an engineer adapted a program originally developed for drawing integrated circuits to generate triangular forms arranged to represent motion–the framed painting shows a single static frame from a work that was initially conceived as a film.

More generally, across all five cases, the key insight is that design itself can make complex computational systems and abstract mathematical ideas visible and tangible. Whether through digital fabrication, software reconstruction, or data visualization, design can function as a method of interpretative translation. That is to say, a method of making visible, tangible, and experiential what traditional academic scholarship in technical domains typically communicates through words and word-like representational devices such as scientific diagrams and tables.

Simone Brauner. Why these five and not others? What drew you precisely to this election?

Alexandros Haridis. The use of the term aesthetic judgment in relation to computation is neither neutral nor particularly modern. The five systems I chose for the exhibition are drawn from a much broader intellectual trajectory in Europe and the United States that treats computation as a medium for addressing foundational questions of aesthetics.

The key reason I selected these particular systems is that, taken together, they capture two major computational paradigms of the twentieth and early twenty-first centuries: systems whose behavior is specified explicitly through rules and the way these rules are applied and systems whose behavior is learned statistically from a given dataset. The exhibition is therefore less a historiography of these individual moments than an exploration of two different ways of thinking computationally about human aesthetic judgment.

These five systems are among the best documented examples in the literature, with multiple research papers or books written about them. They also form part of a larger database of aesthetic systems that I have been building, a portion of which is visualized in the timeline projection within the exhibition (Figure 6). The timeline classifies systems according to their computational approach–for example, psychophysics, shape grammars, or machine learning–and the disciplinary origin of their original research papers or books, such as architecture or computer science.

Of course, other systems could have been included. For example, Leeuwenberg’s Structural Information Theory (SIT, 1968/2013), or Max Bense’s Information Aesthetics (1965), an early attempt to understand aesthetic decisions through cybernetics and information theory that became influential in post-war German design schools such as HfG-Ulm. The exhibition, however, is not intended to be a history of computation as it relates to aesthetics. Rather, it focuses on a small number of exemplary computational systems that operationalize explicit theories of aesthetic judgment and demonstrates how design can make those computational ideas tangible, visible, and experiential beyond traditional modes of academic presentation and public dissemination.

Simone Brauner. Your title is Beyond Data-Driven Aesthetics, and walking through the first four systems makes the meaning of “beyond” clear. George D. Birkhoff has a formula. Vera Molnár has guidelines. George Stiny has an algorithm. Lillian F. Schwartz has repurposed software. None of them works with large amounts of data. And then there is AICAN as the fifth node. AICAN is a generative adversarial network trained on the WikiArt dataset, and therefore unmistakably data-driven.

The team at this fifth node is also the largest: design engineering by Jingfei Huang, Riddhi Kasar, and Sherrie Shou; machine learning by Jimmy Wei-Chun Cheng. How did you arrive at this form together? What guided the decision for two monitors, for these specific interpretability techniques, for a temporal resolution of one hundred epochs? And what do visitors see in this visualization that they cannot see when they look only at the AICAN images themselves?

Alexandros Haridis. The first four systems are indeed “beyond” the purely data-driven methods approaches we find today in machine learning. They belong to a different computational paradigm. Rather than learning statistical relationships from large datasets and then sampling solutions from that learned distribution, their behavior is specified explicitly through rules of generation and judgment that were defined by their inventors and documented in published papers and books.

The distinction is therefore not simply between “small” and “big” data but between two fundamentally different computational paradigms: systems whose behavior is specified explicitly through rules and the way they are applied and systems whose behavior is learned statistically from a given dataset.

The fifth node as you put it, Aesthetic Novelty, originates from early 21^st-century data-driven approaches to art and design. I selected a machine learning system called AICAN, short for Creative Adversarial Networks (CAN), that was inspired by Generative Adversarial Networks (GANs), a class of very popular neural networks used in many modern deepfakes, as well as in applications in art and architecture and also in healthcare.

What makes AICAN an interesting case is not so much the quality of the images it generates but that the system itself is modeled after an explicit theory of aesthetic judgment–this is a common theme across all cases in the exhibition. AICAN, in particular, adheres to a theory proposed by psychologist Colin Martindale in the area of cognitive aesthetics (Martindale 1990), instances of which can be found in other authors such as in Daniel Berlyne’s “Wundt curve.” The idea is that the vast majority of artists seek to make their works appealing by rejecting existing forms, subjects, and styles that the general public has become accustomed to, thus aiming to arouse viewers and capture their attention by doing something novel. However, there is also a “least effort” principle in which too much novelty will estrange most viewers. In other words, balancing novelty and familiarity (what is considered acceptable) is what contributes to aesthetic value.

Alexandros Haridis. While AICAN is trained on a dataset, and is therefore a data-driven system, it forms a bridge between the earlier systems in the exhibition and contemporary machine learning. I call it a “hybrid” system because it combines statistical learning with an explicit theory of algorithmic judgment. To put in in technical terms, the “discriminator” function of AICAN is a modified version of the base GAN model and incorporates explicit rules of judgment that balance novelty and familiarity in the model’s judgment space.

In the exhibition, this case study is represented by two screens that show animated data visualizations (Figure 7): the left screen interprets AICAN’s generative and discriminative processes across training epochs and the right screen visualizes its feature layers as an image develops from coarse pixels to an art-like object. We used image samples produced by a custom implementation of AICAN following Elgammal et al. (2017), trained on the WikiArt dataset on Google Colab’s compute units. The two screens use feature visualization and dimensionality reduction to interpret the model’s decisions in image generation and aesthetic judgment, two interpretability techniques used in the literature of machine learning when AI systems are built on increasingly complex neural network models.

Looking at the generated images alone tells us what the system produces. My interest, however, was in using design, specifically data visualization techniques, to instead reveal how it arrives there. Visitors watch the learned representations gradually emerge during training, observe how internal feature representations evolve, and see how the discriminator function discerns or chooses between familiar and novel image samples. In other words, the visualization does something that reading the paper itself does not immediately reveal: it shifts attention from AI as a black box that produces images to AI as a computational process of judgment that visitors can actually observe.

Although the computational methods differ radically across the five case studies, they all share one ambition: to make aesthetic judgment explicit by revealing how it can be approached through different computational models. The earlier systems accomplish this through hand-crafted rules and algorithms, whereas AICAN does so through a combination of learned statistical representations and an explicit computational model of judgment.

Unlike the earlier historical case studies, this installation required expertise in exhibition design, machine learning implementation, software engineering, and data visualization. The contributors in this case study were students and researchers affiliated with Harvard University and CMU. Different members of the team contributed according to their expertise at different stages of development. My role was to direct the overall research interpretation and creative direction, and ensure that every design decision remained faithful to the conceptual structure of the original academic source while translating it into an engaging public object. As is common with large projects involving multiple institutions, collaborating organizations, and research assistants, we didn’t meet all together in real time but rather one person’s work became a scaffold for another’s at different stages of the process–modern communication technologies make this kind of multi-stakeholder project development quite effective.

Simone Brauner. One could also read all five systems as sharing a single gesture. They negotiate tensions and try to formalize a favorable point between two poles.

George D. Birkhoff: order against complexity, Vera Molnár: strict grid against small disturbance, George Stiny and James Gips:
constructive brevity against variance, Lillian F. Schwartz:
recognition of the original against the strangeness of the new image, AICAN: familiar stylistic space against deviation.

Was this figure of balance a guiding principle in your selection, or does it emerge as a common denominator only in retrospect? Birkhoff measures the balance, while the later systems negotiate it. Does the mode shift over time, from measuring to negotiating?

Alexandros Haridis. It emerged as a common denominator in retrospect, and recognizing this shared structure is one of the contributions of the exhibition. That is, it shows how distinct aesthetic systems developed independently by different authors converge on analogous ways of characterizing judgment, often consisting of two poles such as “order” and “complexity,” “novelty” and “familiarity,” and so forth. I discuss this in one of the final chapters of my PhD dissertation where I conclude that systems like Birkhoff’s and Stiny and Gips’s, or similar ones in the rule-based tradition, reinterpret the traditional principles of unity and variety, measured individually (e.g., as when unity is measured as order or balance in the polygons of a shape) or in terms of some numerical ratio that interrelates them. This includes, in particular, responses about evaluative qualities such as symmetry, balance, and directional diversity in a polygon’s boundaries as in Birkhoff’s system, and uniformity and variety in the shapes that are embedded in a geometric painting as in Stiny and Gips’s Algorithmic Aesthetics.

The shift from measuring to negotiating, to use your description, fits well AICAN’s underlying discriminator function. Its optimization process negotiates between the two dimensions of its judgment space, namely familiarity and novelty, both of which are framed in the original paper in terms of whether a generated image fits a recognizable artistic style taken from the WikiArt dataset. From this perspective, there is a certain continuity across almost a century of computational approaches to human judgment in the aesthetic realm.

The formalization of these poles, however, comes with its own inherent constraints on seeing and interpretation. Each system comes with its own implicit “viewpoints” that their underlying rules model, its own ways of looking. These viewpoints–what to attend to and how to evaluate–both facilitate and limit human sensibility and judgment. Every evaluative system comes with its own descriptive conventions which necessarily privilege certain aspects of an object while leaving others in the background. Changing your mind and shifting your perspective about what to pay attention to changes what becomes visible or meaningful in the same object. The exhibition makes these assumptions visible and invites visitors to ask what each computational system allows them to see and what it inevitably leaves out.

Simone Brauner. Has it ever happened to you that you did not find an object beautiful at first sight, and that after spending more time with it, after some reflection, with a little knowledge about how it came into being, you found it beautiful after all? Or the other way around: something you found beautiful right away became, after reflection, dull, empty, perhaps even unpleasant.

I am asking because the exhibition deals with systems that formalize aesthetic judgment. These systems work in a moment, in a single pass. Our own experience of aesthetic perception does not seem to function that way. It moves. It returns. What role plays time and perception in evaluation? What, for you, is the difference between a first judgment and a matured one? And if such a difference exists, can it be formalized at all?

Alexandros Haridis. This question touches directly on some of the research I've been doing recently because it points to a distinction that I think is often overlooked in discussions of computation and aesthetics. What you describe as a judgment that “moves,” “returns,” or changes over time as opposed to a judgment that works in a moment, in a single pass, corresponds closely to the distinction between reflective and determining judgments, respectively, a distinction introduced by Kant in his third Critique, and which I discuss in detail in relation to computation in my recent article (Haridis, 2026).

Many of the computational systems in the exhibition indeed resemble determining judgments: an observer, a human or a machine system, evaluates a given form (perhaps a plan, facade, painting, ornament, etc.) by measuring its correspondence to a set of predefined rules or evaluative criteria–a single rule of cognition, to use Kant’s terms. Birkhoff’s Aesthetic Measure is perhaps the clearest example. A polygon is judged according to explicit geometric properties such as order, symmetry, and complexity. This determining notion of judgment—that there are concepts or rules others can lawfully replicate—is most evident in well-documented formalist studies on computation and aesthetics in the twentieth century.

Human aesthetic experience, however, is generally speaking much richer than that. Reflection allows us to return to an object repeatedly, and each return can reorganize what we perceive. Nothing about the object itself may have changed, yet our understanding of it or the way we look at it has. New associations emerge over time, we recognize or revise references we had previously overlooked, we learn something about its history, its construction, or even the virtues and vices of its creator, and we begin to relate it to other works we have encountered.

One example of this reflective judgment I discuss in my work has to do with architectural form. Kant, in particular, characterizes architecture as an “adherent beauty” in the sense that buildings and other architectural structures most often have to adhere to a particular function or typology (e.g., a hospital, a church, and so forth). Yet even in this case architectural forms trigger our imagination and invite us to perceive and interpret them through concepts and images beyond the particular function or purpose they were originally made to satisfy. Many Postmodern architectural examples work in this way. Forms can be “read” in terms of symbols or visual cues—quotations, to use architectural critic’s Charles Jencks’s literary term—taken from culture, history, or other architectural styles. In other words, the same architectural form may embody multiple readings which may or may not be obvious to us from the start, and these readings might as well be contradicting to each other.

That adherent beauties may embody contradictory readings is evident in cases such as Philip Johnson’s AT&T skyscraper (1978) in New York. One can associate the building’s top with the shape of a classical broken pediment, the middle with the a modern skyscraper such as the Chicago Tribune (1923), and the bottom with Brunelleschi’s Pazzi Chapel (1443). The building itself doesn’t change; what changes is the concepts or rules of seeing you apply to it, drawing from a network of associations coming from different eras and architectural styles.

Figure 8: Isle of Dogs Pumping Station or Temple of Storms (1986–1988) by John Outram. Source: James Davies for the Historic England Archive.

The example I have in Figure 8 is John Outram’s Egyptian-inspired Temple of Storms (1988) which is an analogous case. A distinctive example of a postmodern utility building in the United Kingdom, its form evokes the ancient Egyptian ornament, quite like those documented in 18th and 19th-century design archives, such as the plates of Chapter II in The Grammar of Ornament (London, 1856) by Owen Jones. This is one possible of way of seeing that architectural form beyond the fact that it’s a utility building and that it consists of walls, columns, and so forth.

Many existing systems successfully formalize determining judgments. They can make explicit the numerical or algorithmic criteria according to which objects are evaluated. Formalizing reflective judgment is much more difficult because it requires modeling imagination, memory, association, and the evolving ways in which meaning emerges through experience. That, to me, remains one of the most exciting open questions in computation and one that extends well beyond current data-driven approaches to AI.

Your use of the word matured introduces a second dimension that is slightly different from what I call reflection. A mature judgment often implies the gradual development of expertise. Here the question is not simply whether one changes one's mind, but whether one becomes capable of perceiving distinctions that previously went unnoticed, and more than this, these distinctions aren’t just any distinctions but reflect a kind of expert, exclusive, knowledge that allows you to say that some judgments are better than others. This idea has a long history as well. David Hume, for example, in his essay “Of the Standard of Taste” (1760) says as much when he argues that aesthetic judgment is refined through experience, comparison, and cultivation, allowing certain observers to become more reliable critics than others. In architecture and design, education frequently develops this same capacity by training students to recognize formal relationships, historical precedents, or spatial qualities that are initially invisible to them.

I tend to think that this sort of ability to make matured discernments isn’t simply an accumulation of knowledge. It is the cultivation of new ways of seeing. That brings us back to reflection and to the possibility that the same object may continue to reveal new meanings throughout a lifetime.

You describe Beyond Data-Driven Aesthetics as both an exhibition and a research platform. The exhibition closes a chapter that begins in 1933 with Birkhoff and ends, for now, with AICAN. But the platform is open and ongoing. If you imagine it five or ten years from now, what would you hope to see in it that is not in it yet? A new system, a different question, a shift in method, a new voice to join?

Alexandros Haridis. “Beyond Data-Driven Aesthetics” is both a research exhibition and an ongoing platform for investigating how computational systems participate in processes of aesthetic judgment, generation, and transformation across architecture and the applied arts. As contemporary interest in computing within these fields continues to expand, the project is intended to function as a platform for evolving analyses, case studies, and historical moments where computing shapes aesthetics as a form of intelligence.

One of the central questions of the exhibition—and one that researchers across architecture, design, and engineering are increasingly focusing on—is computational evaluation beyond purely performative or functional requirements. This applies to many different design spaces, whether buildings, structural forms, or everyday products. Part of the reason is the growing effort to automate design through generative systems capable of producing innumerable options to choose from, yet we’re often left struggling with encoding decisions that involve aspects of design we often think of as exclusively human–things like aesthetic interpretation, discernment, and creative insight. I’ve become increasingly convinced that understanding these issues requires more than data alone. The exhibition’s case studies suggest that many of these questions long predate current interest in computing and AI and have been approached through a range of computational and theoretical models of evaluation since at least the early 20th century.

The exhibition draws from a much larger database of aesthetic systems going back to the 17^th and 18^th centuries. Extending this chronological index and making it available to other researchers and scholars is one of the directions I will continue to focus on. At the same time, I’m increasingly interested in how these ideas can move into broader applications related to the built environment. In particular, I am interested in how research connected to “Beyond Data-Driven Aesthetics” can help designers and engineers better understand how computation—whether rule-based or data-driven—can inform us about what contributes positively to human experience in relation to the spaces and objects that people inhabit and use.

Finally, a direction I continue to explore is the methodological role of design itself as an interpretive device. Through techniques such as software reconstruction, visualization, and physical making, design becomes not merely a way of presenting research but a way of conducting and interpreting it. The exhibition uses design to translate opaque computational systems into more legible, tangible, and experiential artifacts. More broadly, this opens questions not only about mechanizing “beauty” or “taste” (the traditional preoccupation of aesthetic formalism in the 20th century), but also about how traditional forms of research scholarship and communication may evolve through spatial, visual, and public-facing formats. In that sense, I hope the platform will not simply expand our understanding of aesthetics but also broaden the ways in which computational research itself can be expanded, communicated, and experienced through design.

Simone Brauner. In your catalogue text you write: "a question that has existed since the 1950s but has gained renewed urgency as design and art industries continue to grapple with recent advances in AI." Why is the question urgent now? And what do you see as the deepest risks and possibilities that this urgency carries?

Alexandros Haridis. I think the urgency comes from the fact that, for the first time, millions of people are interacting daily with computational systems that appear capable of producing artifacts we traditionally associated with human creativity. Questions that, for decades, were largely confined to research laboratories and academic discussions—Can machines create? Can they evaluate? Can they make aesthetic judgments?—have suddenly become public questions. Designers, artists, universities, large industry labs, and the broader public are now confronting them directly.

At the same time, I think many of these debates are framed too narrowly. They often focus on whether AI-generated images are beautiful, original, or indistinguishable from human work. In my research, I often associate this with Turing’s essay on “Computing Machinery and Intelligence” (1950). There, he lays out a measure of machine success that is essentially substitutability: can a machine perform a task well enough to be mistake for a human performing the same task? The substitutability principle is our most basic measure of aesthetic success in the creative world. But this presupposes only the generation aspect, not the evaluation or the judgment part: how does the machine arrive at a judgment? How does it express a preference? How does it form interesting or meaningful associations between objects? These are the questions that motivated the exhibition.

In a paradoxical way, the wider the availability of and access to AI systems becomes, the more important it is for people to learn to cultivate their own judgment. In a world of machines trained on vast collections of existing human judgments, being able to judge and, yes, even work on refining your taste through reflection and knowledge, will become even more critical than before. In fact, we can see this urgency across the spectrum of disciplines and judgment theories; from those fields seeking fairness, logic, and ethics in an AI’s outputs to those seeking taste and aesthetics. If we begin to treat computational outputs as objective or neutral simply because they are produced by algorithms, we risk overlooking the choices, representations, and biases embedded in the data on which contemporary AI systems are trained. In many ways, one of the central educational challenges of today’s AI era is teaching people how to judge what those systems really produce.

Ultimately, I think the urgency has less to do with AI replacing human creativity than with understanding how these systems shape the ways we see, evaluate, and make decisions. If aesthetic judgment is indeed a form of intelligence, then the question is not simply whether machines can produce beautiful images. It is how computational systems influence our own capacity to judge, and how we might design them in ways that expand, rather than replace, that capacity. I hope the exhibition contributes to that broader conversation.

About the exhibition

Beyond Data-Driven Aesthetics

Beyond Data-Driven Aesthetics: Digital Reconstruction and Public Communication of Aesthetic Systems in Architecture and the Applied Arts is a research exhibition conceived and curated by Alexandros Haridis for the Keller Gallery at the Massachusetts Institute of Technology.

The exhibition draws from several years of research by the curator and represents a translation of traditional modes of research scholarship and communication into spatial, tangible, and experiential formats. The project’s primary documentation platform is its official website (aestheticsbeyonddata.com), which contains additional texts, materials, and documentation related to this research project, including coverage of the exhibition in the form of a short research film produced by MIT Video Productions.

The curator gratefully acknowledges the contributions of students and researchers across three institutions, Harvard University, MIT, and Carnegie Mellon University, to the production, installation, and fabrication of this exhibition: Blue Chan, Nathaniel Chavez-Baumberg, Jimmy Wei-Chun Cheng, Maciej Dzumala, Jingfei Huang, Riddhi Kasar, Sherrie Shou, Yaluo Wang, and Adrian Yu. Joél Carela provided essential administrative and logistical support at the MIT Keller Gallery throughout the development and installation of the exhibition.

This exhibition would not have been made possible without the support of Harvard University’s John A. Paulson School of Engineering and Applied Sciences, Harvard University’s Office of the Provost, and MIT Department of Architecture. The curator also warmly acknowledges the support of George Stiny, J. Yolande Daniels, and Nicholas de Monchaux.

Website

About the researcher and curator

Alexandros Haridis

Alex Haridis ...

Website