Seeing Myself through the Machine

How I Came to This Question

My interest in the relationship between people and clothing grew out of professional experience. I graduated from a fashion college, worked in fashion media, and later started my own business, spending time on the garment production floor and working across the clothing business. Eventually, I began exploring generative adversarial networks, or GANs, hoping to make the design process before production faster and more efficient.

These experiences left me navigating a tension between two different logics: the creation of images through art, fashion, and design, and the material realities of making, selling, and holding inventory of clothes. A Japanese course on personal color analysis, together with the body-shape, skin-tone, and style classifications gaining popularity on social media, helped me connect the two. They offered a set of tools for thinking that shifted my attention from design, production, and sales toward the relationship between a person and their clothing. What needs does clothing respond to? What needs might it also shape?

My experience led me to recognize that our understanding of our bodies and personal styles does not arise from introspection alone. It also draws on observations, measurements, and names supplied from outside. As I brought this approach into my research on AI-assisted clothing recommendations, a more compelling question emerged: when a system observes me, generates images of me, and explains what “suits me,” is it also quietly participating in defining what I am?

When I began selling clothing through e-commerce, my attention gradually shifted from trend reports and sales figures toward the people choosing and wearing the clothes. At the time, I genuinely valued methods that classified women’s bodies and identified corresponding styles. Getting dressed can easily become entangled with emotion and uncertainty. Do these colors work together? Is this design fashionable? Does this style feel right? Without a basis for judgment or confidence in our choices, we may hesitate repeatedly, even confusing “Does this look good on me?” with “Am I good enough?”

Classification tools seemed to offer a basis for making decisions. Western four-season color analysis, Japanese approaches to personal color, K’s thirteen-type style framework, and the many styling methods developed by social media creators all seemed to help women break an elusive aesthetic judgment into steps that could be described, compared, and reasoned through. They suggested that understanding ourselves could follow a discernible path, rather than leaving us to follow retailers’ guidance or take our chances with our own judgment.

I trusted these frameworks because they gave me a sense of control. If I understood my type and style well enough, I would spend less time searching for suitable pieces and make fewer choices I later regretted. That practical, efficient help was what I wanted. Yet as these methods began to take shape in my work and offer benefits, I also began to examine them more closely. Where did they come from? How many untested assumptions were embedded in the choices they helped me make?

While reading the preface to The Perception Machine, I paused over a series of images concerning machines, perception, vision, and identity. One posed a question so simple that it felt almost abrupt: “But what am I?” It prompted a further question of my own: when a machine describes my body, is it also helping to define who I am?

That question suddenly made me reconsider the tools I had accumulated. My body can be observed, measured, classified, and described. Once it has been processed in these ways, seemingly objective numbers and labels can become the very materials through which I understand myself.

In the preface and introduction, Zylinska returns to questions of how machine vision organizes and interprets images. Her work led me to see perception as a situated process of selection and construction, rather than a neutral recording. Reading these texts and images, I began to reconsider the relationship between machine observation and self-understanding. Could a machine’s description of me become a basis for how I understand myself?

To explain where technology’s participation in definition might occur, I want to use a concrete example: measuring shoulder width. If a system records my shoulder width as 40 centimeters, that is a measurement—an approximate description of a bodily feature. The issue lies in the sequence of operations that follows.

First comes comparison. To decide what “40 centimeters” means, the measurement must be placed within a distribution. Whether it counts as wide, average, or narrow depends on whom I am being compared with and how that reference group has been defined.

Then comes classification. Once I am placed in the category of “broad shoulders,” the measurement becomes a label that begins to direct what follows. Next comes aesthetic interpretation: the system may explain how broad shoulders should be visually softened, which necklines should be avoided, and which designs should be used to “balance” or “minimize” them. Only then comes recommendation, compressing all these judgments into a list of items that “suit you.”

I want to emphasize that this sequence is my own analysis, not a passage from Zylinska’s The Perception Machine. It expresses my understanding of how technology can participate in definition. Between measurement and recommendation, each intervening step—comparison, classification, and aesthetic interpretation—contains judgments. These may carry assumptions about what is “normal,” preferences about what is beautiful, and expectations about how women should present themselves. Through these apparently reasonable intermediate steps, technology can quietly move from describing a person to directing them.

Technological developments have given us knowledge and improved our lives. Fashion media and social media frequently present “before and after” transformations in which tools and techniques appear to reveal something about a person: “Your skin has warm undertones,” or “Your body falls into the ‘natural’ category.” The resulting transformation can produce a moment of recognition. Previous difficulties in choosing clothes begin to make sense, and the framework used to explain them starts to feel trustworthy.

For me at the time, this knowledge offered explanation and a sense of control. It turned vague self-doubt—“Do I look good?”—into more specific questions that seemed possible to address.

Yet there is another side to these tools. Neither the analysis nor the proposed solution is purely technical. Both can contain socially constructed labels and expectations about women. Why should broad shoulders need to be “softened” rather than accepted? Why is a particular skin tone said to “suit” certain colors while others should be avoided? Which judgments respect a person’s own aims, and which conceal default expectations about how women should look, what qualities they should convey, and what counts as appropriate?

Even the changes shown in before-and-after images may be influenced by lighting, pose, makeup, and image selection. A visual comparison alone cannot establish that a classification method works.

When we use these tools, might we be accommodating someone else’s image of the “ideal woman”? For the sake of convenience, might we then find reasonable explanations for fitting ourselves into it?

This is the tension I find myself confronting: I turn to classification to see myself more clearly, yet I also sense that the framework making me visible is asking me to become a particular kind of person.

These concerns might invite a quick response: if technology contains bias, perhaps it cannot be trusted, and people should simply rely on their own judgment.

I do not find that sufficient. Our understanding of ourselves is also subject to error. I may have an unquestionable right to define myself, but that does not mean my understanding of my body is always accurate. The image I remember of myself often differs from what I see in a photograph. A color I believe “suits me” may simply be one I have worn for years.

Where data collection conditions and measurement methods are reliable, AI may offer information that memory or everyday observation cannot easily provide. But that information still requires scrutiny. It cannot, by itself, establish the correctness of an aesthetic interpretation. Accurate body measurement and a judgment about “what suits me” are two different matters.

Acknowledging AI’s potential does not require surrendering judgment to it. Its information can contain errors, and its explanations have limits. How were the data collected? How was the reference group established? Where were its aesthetic preferences learned? These conditions shape what the system presents as “objective.”

What I want is an ongoing exchange in which both sides can be questioned. AI may supply information where my own understanding has gaps, while I continue to ask where its classifications come from. Using technology, for me, is a way to reconsider my relationship with my body from a wider perspective, rather than to obtain a more “accurate” version of myself.

The discussion so far has centered on knowing myself. For me, discovery also opens up the possibility of change. That possibility must include accepting my present appearance, keeping my existing style, or choosing not to change for now. Becoming myself should not become another demand to keep improving how I look.

Through reading this book and pursuing this research, I have gradually found a way to express what draws me forward: I discover myself in order to change; I change in order to move toward the person I hope to become. These are not stages on a straight path, but parts of a recurring process. A label can be a starting point, but it should not be an endpoint.

For women, this movement from understanding to becoming can carry particular significance. Appearance is part of it, but the question extends far beyond being “beautiful.” It can involve identity, social roles, confidence, self-presentation, and a woman’s changing understanding of her circumstances across different stages of life.

The person a woman hopes to become is fluid, continually revised through relationships, experiences, and the times in which she lives. Any approach that fixes her within a category—even one that once seemed “accurate”—may inadvertently close a door she wanted to open. I would want technology to accompany her in asking, “What else might I be like?” rather than act as a judge declaring, “This is your type, so this is how you should dress.”

My question has moved beyond whether AI can tell a woman which clothes suit her. It now turns toward something more difficult:

I do not yet have an answer. But prediction accuracy alone cannot tell us whether we are moving in a worthwhile direction. I want to explore a continuing process of mutual adaptation: a system that learns about a user while allowing her to depart from any label it assigns, and a user who draws on technology while retaining room to question its classifications and change direction.

Understanding myself should always leave room for becoming myself. This book has given me a starting point for learning that relationship anew: how to use the knowledge machines produce about me without surrendering the right to define myself.

References
Zylinska, J. (2023). The perception machine: Our photographic future between the eye and AI. The MIT Press. https://doi.org/10.7551/mitpress/14471.001.0001

Image from Joanna Zylinska, The Perception Machine: Our Photographic Future between the Eye and AI (MIT, 2023), p.xiv.