LLM Autobiographies and the Politics of “Being Seen” Online

LLM Autobiographies and the Politics of “Being Seen” Online

Introduction

Have you ever pasted a prompt like “describe me based on our chats” and felt that weird mix of comfort and creepiness—like the model is reflecting you back, but also simplifying you? That’s exactly the kind of moment Anthony Longo and Olya Kudina dig into in their new research on political sense-making in human–technology–world relations, using LLM-generated autobiographies as the case study. This blog post is based on fresh work from the paper “Political sense-making in ‘individual’ human-technology-world relations: the case of LLM-generated autobiographies”.

The big claim isn’t that AI secretly “knows you” or that politics is hidden somewhere inside a model like an inner engine. Instead, Longo and Kudina argue that AI’s political significance shows up in the ordinary practices of interpretation—what people do with the outputs over time and with others watching. When you prompt a system to describe you, you’re not just receiving an algorithmic mirror. You interpret it, challenge it, correct it, share it, and test it in public. That whole loop is where politics happens.

And here’s the twist that will likely annoy (in a productive way) some readers: the “micro” encounter between one person and one model is already saturated with publicity. The private becomes communicable. The reflexive becomes accountable. Meaning gets sedimented into shared expectations about what a “real person” looks like—just through writing a short AI biography.

Why This Matters

This research is especially significant right now because LLMs are quietly shifting from being “tools you talk to” into media that help author your public self. Autobiographical prompts are one of the most emotionally accessible uses of generative AI: they don’t require you to understand machine learning, but they do shape what you think you are—and what other people think you are.

A practical scenario where this lands today: imagine a student or job seeker using ChatGPT to draft a personal statement or “about me” bio, then posting it in applications, portfolios, or even social profiles. Even if the model output is “private” at first, it becomes publicly testable the moment the text is shared—because others will judge coherence, tone, credibility, and “authenticity.” Longo and Kudina’s framework helps explain why this isn’t merely a writing shortcut. It’s a process that reorganizes the standards of appearance: what counts as a legitimate self-description.

This also builds on (and pushes past) previous AI research in a useful way. A lot of earlier discussions focus on whether algorithms replace human judgment, or whether opacity harms accountability. Longo and Kudina agree opacity matters, but they argue the political problem is often framed too narrowly—using the “black box” metaphor in a way that treats politics as something inside the system and users as politically alienated. Their alternative view says: politics is in the interpretive loop—in how people and communities negotiate mediated appearances over time.

How the “Black Box” Metaphor Misleads Us About Politics

A lot of AI policy debates orbit the “black box” idea: the model is opaque, so its political meaning is presumed to be hidden inside it. Longo and Kudina say that framing causes a specific kind of confusion—what philosophers call a category mistake.

The category mistake: confusing “transparency” with “political meaning”

The typical logic goes like this:
- If we can’t see how the algorithm works, we can’t know what it’s doing politically.
- Therefore, political remedies must focus on opening the internals: code, parameters, training data, explainability, audits.

Longo and Kudina argue politics and transparency belong to different categories. Even if you release technical internals, publics may still not be able to interpret or contest them. Conversely, intelligibility (in a political sense) can emerge without technical disclosure—through lived experience, tactics, resistance, and social learning. Their punchline: what matters politically is not only how something works, but how it gets encountered, judged, and repurposed in the shared world.

Arendt’s counterpoint: politics doesn’t “operate from the inside”

To make that case, they bring in Hannah Arendt. Arendt’s political philosophy rehabilitates the importance of appearance—not because appearances are fake, but because politics happens where people can show up to one another and judge what they see.

So instead of asking “what’s inside the model?”, a more Arendtian question is:
- How does this technology shape what becomes visible?
- How does it structure the space where spectatorship, opinion, and judgment take place?

That shift matters because it treats AI not only as an information processor but as a participant in the conditions of public appearance. And once you see that, the “political deficit” critique against postphenomenology starts looking wrong: you don’t need to wait for perfect transparency to see politics unfolding.

Why LLM “Self-Descriptions” Turn Personal Prompts Into Public Politics

This is where the research becomes really concrete. Longo and Kudina examine the practice of using LLMs to generate autobiographical narratives—prompts like “describe me,” “tell me something about myself I don’t know,” or “write a personal narrative.” Their core point is that these interactions create a new kind of human–technology–world in-between: not a simple human-to-machine transfer, and not a one-way machine influence either.

The interpretive loop: resonance → friction → public re-articulation → sedimentation

They describe a recursive process that unfolds through mediated sense-making. Think of it less like “the model outputs a truth” and more like “a conversation that keeps rewriting your understanding”:

1. Affective appropriation
   Users approach with curiosity and often with a desire to be recognized. The model’s fluency and affirming tone can feel intimate—like confiding in a non-judgmental other. This creates an early moment of validation where the self and the technology’s framing fuse temporarily.

2. Friction and reflective correction
   Then comes dissonance: the output can feel generic, inaccurate, or emotionally flat. When users correct the model—or replace parts of the narrative—they move from passive absorption to active interpretation. The AI shifts from “transparent medium” to something you scrutinize as an object of attention.

3. Re-articulation and public engagement
   Users often share the results: in blog posts, threads, social platforms, or community discussions. At that point, private reflection becomes plural. People compare, critique, joke, and test the portrayal against their own experiences. The text becomes a mediator of public discourse.

4. Temporal sedimentation and co-evolution
   Over time, habits form: how users prompt, what tone they expect, what gets accepted as plausible selfhood. The technology and the user’s interpretive patterns co-evolve. Meaning becomes provisional—open to revision because future mediated encounters can unsettle earlier ones.

A quick “what this looks like” example from the paper’s discussion

Longo and Kudina mention an illustrative blog post where a user asked ChatGPT to “describe me based on all our chats.” The model produced something flattering but also strangely generic—“an eloquent, reflective, intellectually curious” kind of person—so generic that the user felt it “sounded like me without being about me.” That’s classic friction: affective resonance first, then interpretive correction.

They also discuss community settings like Reddit threads where people post model-generated characterizations and react collectively—often with irony. That’s the key: the “individual” encounter doesn’t stay individual. It becomes a calculated public where standards of appearance and authenticity get negotiated.

The crucial conceptual reversal: not “inside-out politics,” but “between-world politics”

This is where the authors explicitly challenge the “black box” view. They reject both:
- Interiorization of politics (politics is hidden inside the artifact and transparency is the main cure)
- Exteriorization of technology (the algorithm is an outside agent acting on passive users)

Instead, they argue politics emerges in the relational-temporal loop—how interpretation moves across affect, friction, public articulation, and sedimentation. In other words: the model doesn’t just describe a self; it helps generate the conditions under which a self becomes appearable, contestable, and therefore politically real.

A Handy Comparison: Where “AI Politics” Is Supposed to Live

A lot of confusion in AI ethics comes from mixing up two questions:
1) where the political “problem” is located
2) what counts as a political “solution”

Longo and Kudina’s critique implies a different mapping than the standard “black box” view. Here’s the contrast in plain language:

This isn’t an argument that technical transparency is useless. It’s an argument that focusing only on it misses what’s already happening: people are already interpreting, contesting, and testing AI-produced portrayals in shared contexts.

What “Political Sense-Making” Implies for Designers and Policymakers

If politics is intrinsic to the mediated loop—not an add-on you bolt on later—then design and governance should shift emphasis.

Design implication: build for articulation and contestability, not just smoothness

Longo and Kudina’s framework suggests cultivating productive friction. That doesn’t mean “make users uncomfortable” on purpose. It means designing interfaces that let people:
- revise outputs,
- expose ambiguity,
- compare alternatives,
- and support plural viewpoints rather than locking in a single “clean” narrative.

A smooth system that always affirms you can reduce the chance for reflective correction and public testing. A system that supports revision and contestability helps meaning remain open instead of sedimenting too quickly.

Policy implication: don’t treat transparency as the only accountability lever

They argue that political inquiry shouldn’t wait for full interpretability of model internals. Policymakers should also assess the lived effects:
- How does the system shape what users feel seen as?
- How does it structure what becomes shareable and judgeable?
- What norms of appearance get stabilized through repeated use?

In their words, you may need to look at how AI mediates the common world of meanings—not only how it encodes patterns.

Method implication: study the loop, not the artifact

Their approach is essentially a research method proposal: track sense-making over time as it moves across affective uptake, algorithmic reflection/disruption, contestation, and shared narrative sedimentation. This can apply beyond LLM autobiographies—for example, recommender systems, interface nudges, and non-digital mediation tools—whenever an artifact structures what can appear and be judged.

If you want to connect this back to the original paper, the core thesis is spelled out directly in their argument against the “black box” approach and for a relational-temporal analysis in Longo & Kudina’s chapter.

Key Takeaways

• AI’s political significance doesn’t come from hidden code. It comes from how people interpret, contest, edit, share, and test AI outputs in shared contexts.
• “Individual” LLM interactions are already public. When users generate autobiographical descriptions and circulate them, they move into a space of spectatorship and judgment (very Arendtian).
• The “black box” metaphor commits a category mistake. It conflates technical transparency (“how it works”) with political meaning (“what’s at stake”).
• Longo and Kudina propose a relational-temporal loop. Affective resonance → friction → public re-articulation → sedimentation and co-evolution.
• Design guidance: prioritize interfaces that enable revision, contestability, and plurality—don’t just optimize for fluent comfort.
• Policy guidance: evaluate AI not only for explainability, but for how it shapes appearance, judgment, and shared standards of authenticity over time.
• Future research direction: study how mediation composes shared worlds where people learn to orient themselves—and where judgment becomes accountable.

If you want, tell me your preferred angle (ethics, product design, or policy). I can translate this framework into a practical checklist you could use for evaluating LLM features that involve self-description, profiles, or “about me” content.