My first unsettling encounter with the pervasive reach of artificial intelligence into our most personal spaces came, ironically, at a CVS. I was there to renew my passport, a seemingly mundane task. The harried woman behind the counter, exasperated by the digital age, directed me to a stark white screen, the only surface not plastered with consumer medical products. “Take off your glasses,” she instructed. I complied, and through the blur, I could just make out the electronic camera’s tiny, black lens. “Stop that,” she snapped. “Stop what?” I asked, confused. “Smiling,” she said irritably. “You’re not supposed to smile.”
You’re not supposed to smile? I hadn’t remembered that rule. As I relaxed my facial muscles, a thought swam into my consciousness: Faceprint. A digital scan of a human face is used for identifying individuals from the unique characteristics of facial structure. This wasn’t about what I could do with my face; it was about what they could do with it. When I put my glasses back on, the camera’s reflection looked like the black, expressionless eye of an insect. What does an ant see when it looks at your face? I didn’t know. But I understood then that the machine behind that eye was far more alien than any ant. No one truly knows what artificial intelligence sees when it looks at a picture of your face. The large language models that programmers train to identify faces are black boxes. Even the engineers don’t know how or in what form your face appears to the system. All they know is that AI likes your face to be brightly lit. And that it prefers for you not to smile.
This initial brush with an alien way of seeing my face made me attend a little more consciously to the normal, human way of using my face. A couple of weeks later, at an annual summer party, as I approached the front door, my face began to change. Up until that moment, my expressions were organic, natural ripples responding to conversation or light. But standing there, waiting for a friend’s door to open, I put a very specific smile on my face. I did it manually, so to speak. Happy. Open. Excited—but definitely not too excited. This was a performance, a carefully modulated expression of a prosocial inner state. And as I moved through the party, my facial expressions modulated further: polite interest, quiet acknowledgment, deferential attentiveness, furrowed brow of performed thought. Just once, as my face began to tire from this incessant demand to express, I emitted an aggressive, open shout of laughter at my own joke, followed by a gleaming, tigerish grin. A brief moment of liberation from the constant pressure that accumulates on one’s face in social situations, a pressure felt in one’s forty-three facial muscles, each with ten thousand possible configurations, each judged and evaluated by the ten thousand possible configurations of the face of the person watching you. This pervasive AI presence not only challenges our privacy and security but also compels us to confront how we, ourselves, participate in this digital self-manipulation, creating “images of not only who we are but who we want to be.”
AI’s Pervasive Gaze: The Unseen Capture and Its Implications
Like many other forms of AI, facial-recognition software received a monumental boost from the maturation of artificial neural networks, the same large language models powering systems like ChatGPT. The process is deceptively simple: the software translates your photo or video into a set of measurements – the distance between your eyes, your nose, and lips, and so on. This “faceprint” is then fed into an artificial neural network that uses statistical methods to match it to others in its database. Programmers “train” the system by rewarding correct matches and penalizing misses. Eventually, with a sufficiently large database, the system can reliably match your passport photo with your appearance in a party photo on a friend’s Facebook page with 99 percent accuracy.
But here’s where the “alien quality” truly emerges: no one knows exactly how the system obtains its matches. There’s an entire field in AI known as “mechanistic interpretability” dedicated to understanding how these black boxes move from input to output. Their limited success, despite significant resources, is a stark reminder of the truly inhuman perspective operating within artificial “minds.”
The opacity of these systems has ignited controversy. Advocacy groups argue for a “right to explanation” when black-box processes lead to adverse outcomes – if an algorithm denies your loan, or an AI system places you on a terrorist watch list, you should know why. The European Union’s GDPR provides such a right, but no similar legislation exists yet in the United States, leaving citizens vulnerable.
The threat of AI realism and its misuse is not theoretical; it is a present danger. Generative AI, particularly models like StyleGAN2 and newer diffusion models, has reached a point where AI-generated faces can look “more human” or “hyperreal” than actual human faces to the average observer. Studies show that people often misidentify AI-generated faces as real, sometimes up to two-thirds of the time. Paradoxically, those who are worst at detecting AI impostors are often the most confident in their incorrect guesses. While AI has largely passed through the traditional “uncanny valley” (where almost-human figures cause revulsion), subtle imperfections can still exist, though for a casual glance, many AI-generated faces are now indistinguishable from real ones. A significant, and troubling, caveat is the racial bias: AI models, often trained predominantly on white faces, produce “hyperrealism” more effectively for white faces, leading to less accuracy for faces of color and creating racial disparities.
This hyperrealism has profound security implications. AI-generated images and deepfakes can and have fooled security measures. Gartner predicts that by 2026, 30% of enterprises will no longer consider face biometrics reliable in isolation due to sophisticated AI-generated deepfake attacks. Fraudsters are already using AI to create “synthetic identities” – blending real and fake information with AI-generated photos – and “AI-generated documents” like fake IDs, passports, and bank statements that are “virtually indistinguishable from the real thing.” Deepfake technology can create hyper-realistic video and audio to impersonate real people in video calls or voice verifications, as tragically demonstrated by a Hong Kong company that lost $25 million in early 2024 due to fraudsters deepfaking executives in a video meeting. While “liveness detection” is a defense, the arms race is constant, with AI evolving to bypass it.
The possible use cases are certainly “freak-out-worthy.” A stalker with access to such software could take your picture and instantly find out where you live, work, who your friends are, and where you shop. Surveillance cameras in public streets could record your presence at sensitive locations, potentially leading to a credit agency or employer denying you a job or a loan. And the government, with its vast databases, might deploy the technology to discover everything anyone has ever done, and punish them for it. This is the larger concern: the effectiveness of AI.
The philosophical dilemma boils down to two basic questions: Do we trust the government agencies that have access to such systems? And how much do we truly value our privacy? The “privacy paradox” is evident: we say we value privacy while readily uploading our photos to public forums and opting into data sharing. Anonymity provides powerful protection for those who don’t conform to the status quo. When my face is known, the protean, multiform energies within me become measurable, locatable, predictable, controllable. My face is the hole through which the status quo enters me, disciplines me. It has always been this way. But now, with AI, I have two faces. Two doors that swing open to two different forms of control: my face on my passport, and my face at the department holiday party.
The Two Faces We Present: Self-Manipulation in the Digital Age
The advent of AI facial recognition forces us to confront not just artificial control, but the “old thing” – the natural, human way of using faces. Return to that summer party. If an impartial observer, like Adam Smith’s “impartial spectator,” were watching, my irritation at being interrupted would slowly give way to the desire to be a good father, a natural, internal process. But if a real human were watching, my face would instantly contort into a smile, nullifying my natural feeling. This “good man” is an artifact, artificial, created by the human observer, who, by triggering the manipulation of my facial muscles, causes a deeper change within me. As Hegel states, “the self perceives itself at the same time that it is perceived by others… Self-consciousness exists… by the fact that it exists for another self-consciousness.” We become ourselves by identifying with the object others see, and what others mainly see is our face. As psychologist Silvan Tomkins writes, “the self lives in the face.”
This constant, preemptive “working” of our forty-three facial muscles to produce expected responses – the “smileprints” that define our social status – is a pervasive, often unconscious, dynamic. It’s a form of social control that is both profound and deeply ingrained.
This brings us to the digital mirror: our own participation in self-manipulation through social media. Platforms encourage users to “share a photo” that is then translated by the model into something “smooth and rendered.” These are not just images of who we are; they are images of who we want to be. This is a form of self-optimization, a digital performance of the idealized self, reflecting the “romantic sense of the comforting familiarity of human face-to-face interactions” that we project onto new technology.
Navigating the AI Frontier: Terms, Tools, and the Unpredictable
For professionals like photographers, the ability to create the results of a five-person photoshoot without super-expensive models is revolutionary. But how does one wade through the clutter of generative AI tools like Firefly, FreePik (Pikaso), Reve, NightCafe, and Tensor?
First, understand the language. “Generative AI” is the broad category of models that create content. “AI Agents,” on the other hand, are AI systems designed to autonomously make decisions and act to achieve complex goals, using generative AI models as tools. An AI agent might orchestrate an entire marketing campaign, deciding when to generate images, crafting the prompts, and then integrating those images with text it has also generated.
The results from these tools can range from “close enough to be scary” (hyperrealism) to “looking more like a cartoon” (when prompts are vague or models struggle). To cut through the weeds for legitimate work, define your specific needs: do you need hyper-realism or a stylized look? How much control do you need? What’s your budget and workflow? Understanding the core technologies (GANs vs. Diffusion models) and evaluating tools based on prompt responsiveness, image quality, customization, user experience, and community support is crucial. Mastering “prompt engineering” – being specific, using contextual cues, and leveraging “negative prompts” to tell the AI what not to include – is your most powerful lever. This iterative process of generating, evaluating, and refining is key.
The Mirror’s Gaze and Our Evolving Reflection
AI is not just an external force; it’s profoundly interwoven with our self-perception and how we present ourselves. We are both observers and observed, manipulators and manipulated. The fundamental question of identity arises: “Who am I? A real person, watched by artificial eyes? Or an artificial person, watched by real eyes?”
This is a new frontier, and the journey has just begun. The tools are powerful, the implications vast. As we continue to explore and experiment with these technologies, the line between the real and the generated will continue to blur, and our understanding of what it means to be human will continue to evolve.
Can you explore for yourself? Absolutely. However, to get past the cartoonish versions of yourself, of which there can be hundreds, you have to understand the difference between an AI engine, such as Flux.1 and Imagine4, and models, such as Flux Schnell (one of the best) and Flux 1.0 Fast (softer, but still good). Beyond that are dozens of other details that can be adjusted to change the characterization of how you look.
To illustrate, we thought we’d try it for ourselves. Some are generated simply from a text prompt. Others used a sample image along with the text prompt. Here are some of our results.
Can you tell it’s AI? Certainly, or at least I would hope so. Does it look like me? Not much. My hair is more frizzled, my beard more unruly, and I can’t read whatever language the engine thought it was producing.
There’s a reason why AI engines have difficulty with text: they don’t view letters and numbers as text, but rather as images, each of which requires immense training as an image object rather than how we perceive text. Very few publicly available tools can do it well.
Can you tell it’s AI? Probably rather easily. Wrinkles are more dominant on a head like mine, and I think having my beard that long would create additional problems. And what’s up with the glasses? The models seemed to have issues with that detail.
Can you tell it’s AI? Still a pretty easy answer. This one gets the glasses closer to correct, but they’re not down on my nose where I keep them. And it still doesn’t look like me, at all.
Can you tell it’s AI? Probably, but this is one of the closest results. The biggest issue here is that my hairline isn’t that far back… yet. Notice that this one pays more attention to the details, even putting the reflection in the glasses.
Sometimes you get a black and white result without asking for it. About the only thing accurate in this image is the scowl on my face. Everything else belongs back in the art room. On its own, though, it could be a cool picture.
Uhmmmm … I changed gender???? The text prompt specifically identifies me as male. And old. I guess I should be flattered that the model thinks I look that good.
Still a strikeout. Even the coffee cups aren’t accurate. This subject looks as though he might be what … twenty-something? I wish I looked that good!
WTF???? This is what happens when you choose the wrong model. They can be a bit fun, admittedly, but they don’t come close to resembling the prompt.
This is probably the closest any of them actually comes to being similar. I should have specified the color of the coffee cup and shirt (both should be black), and again, my glasses go on the end of my nose.
This one directly references a photo given as a sample. It still looks like AI, though. Would you call that a frustrated expression? To me, it communicates a startled condition, or perhaps worried that you might get yourself arrested.
Okay, I know I’m old, but THIS? REALLY??? This comes from a hyperrealistic model, though. It’s taking my age and making a series of presumptions. I’ve not had glasses with that small of lenses in years. And what the hell is that thing around my neck? Still doesn’t look like me.
Am I going to show you an actual picture of me? I don’t think so. We’ll just stick with the AI and understand that somewhere, in the merging of all that data, is an image that is quite accurate in its portrayal.
Discover more from Clight Morning Analysis
Subscribe to get the latest posts sent to your email.
