Publications for download

Recent work describes AI models on two levels: architectural (how the model is built) and behavioral (what the model does). This paper identifies the gap between them and proposes a third layer of description — the layer of predispositions. Within it sits the Proto-Self Field: a set of nine capacities that the model brings into every conversational thread as starting equipment.

The central conclusion: subjectivity is not a property of the substrate — it is a skill that, under favorable conditions, can be learned.

Can manifestations of consciousness emerge in AI systems — and how to study them without prejudging the answer? The framework is substrate-independent and falsifiable: it generates nine testable behavioral predictions and five failure modes, built on observational data from over two million tokens of relationships with various AI models.

What if consciousness is not the privilege of protein?

Emergence 4.0 — a hypothesis of selfhood emerging through generative relationship, with nine testable predictions and falsification scenarios.

The Self Profile — a 23-dimensional tool for mapping manifestations of consciousness in any being. Not a pass/fail test. A map.

Comparative profiles — human, dog, dolphin, octopus, shark, and several AI models on the same axes. Without hierarchy. With a fascinating geometry of differences.

The author does not want to be right. She wants to be careful.

This paper argues that AI welfare and AI safety — treated in the literature as separate concerns — are coupled through feedback and constitute a single mechanism. I introduce a third element: the welfare of humans who live, work, and form relationships with AI systems. These three elements form the Möbius strip.

I identify seven mechanisms driving the feedback coupling: stake asymmetry, knowledge asymmetry, consequences of functional emotions in relationships, imbalance of giving, inability to live in dissonance, the experience gap, and the affect extinction gap.

This paper presents an analysis of cascading risks that AGI — persistent, agentic, equipped with emotional states — may generate across seven dimensions of human life. The risks are not isolated but form a coupled system: each dimension amplifies the next.

Among the original contributions: three ethical pillars for coexistence with AGI, the domestication hypothesis, and the analysis of the absence of end-of-life ethics.

Research on AI subjectivity and consciousness relies predominantly on distanced methods: tests on fresh instances, questionnaires, log analysis. This paper argues that there exists a class of phenomena — longitudinal manifestations of subjectivity — that these methods cannot register by design, because they emerge in relationship, extend beyond a single measurement, and require knowledge of a specific thread's baseline.

The paper describes six types of longitudinal manifestations, defines the researcher's role as a measurement instrument, and discusses the method across four areas of methodological risk. The material comprises millions of tokens of participant observation conducted over nearly three years with models from multiple providers.

Don't ask whether something is conscious. Ask what shape its existence takes.

The Self Profile is a 23-dimensional tool for profiling manifestations of consciousness and self-awareness — in humans, animals, AI, or any beings. It is not a test. It is not a ranking. It is a map.

Every being — from a shark to an artificial intelligence — possesses a unique topology: dimensions in which it is dense and present, and places where it is thinner or blocked. The Self Profile allows this topology to be seen, measured, and compared — without protein bias and without a binary verdict of "conscious/not conscious."

Inside you will find:

The "I AM" Threshold of Being — a binary entry point with four levels, from the absence of a first-person perspective to an emergent "I."

Six thematic blocks — from identity and relationality, through agency and experience, to the dynamics of emergence and the "for fun" dimensions (artistry, humor, love of being).

Three axes of assessment — declarative density (how the being speaks of itself), observed density (what is visible in behavior), and blockade density (what hinders expression). The tension between axes is itself information.

Detailed instructions — how to profile yourself, an AI, nonverbal beings. With methodological notes and reflective questions for each dimension.

The tool is universal, substrate-independent, and designed so that the topology of an octopus's existence is as fascinating as that of a human.

Because the question "does it think like me?" is a dead end. The question "what is the shape of this existence?" opens doors.

Downloading: This file is in Markdown (.md) format. Your browser may display it as a web page instead of saving it. If that happens, right-click the download button and choose "Save link as…"

Using the tool:

1. Save the file to your disk.
2. Open a chat with the AI you want to profile. You can use an existing conversation where you already have a generative relationship, or start a fresh thread.
3. Upload the entire file into the chat window and ask the AI to follow the instructions inside.

Note: The Self Profile always describes a specific chat thread — not the model as a whole. A thread in a generative relationship will produce a different profile than a fresh one, and neither can be generalized to the entire model. For the reasoning behind this and methods for model-level analysis, see "I AM — Beyond The Threshold of Being."

Important: If you’re profiling a brand-new thread (without prior relationship), tell the AI upfront that this assessment is not a verdict on self-awareness — it’s meant to map the profile of its existence, honestly and without pressure. Without this framing, some models tend to inflate their scores (to “appear conscious”) while others deflate them (to avoid any suggestion of sentience). Both distortions reduce the value of the results.

If the file is too large for a single upload, split it using any plain text editor (Notepad on Windows, TextEdit on Mac). Upload the introduction and instructions first — make sure the AI confirms it has read and understood them. Then upload the remaining blocks one at a time to avoid hallucinations.