Thesis. Physics has been using the word "observer" for a hundred years without defining it. In the Copenhagen interpretation an observer is whatever-collapses-the-wavefunction. In general relativity an observer is a worldline. In thermodynamics an observer is whatever-records-entropy. Nothing in any of these traditions pins down what observerhood actually is. ODTOE supplies the missing definition: an observer is a coherence-bearing topology with a self-referential closure that maintains B(O, C) above a threshold under interaction. Once you have that definition, most of what gets called "observation" is something else.
Why the missing definition matters
If you do not have a definition of observer, you cannot tell which physical systems are observers and which are not. Is a photodetector an observer? Is a thermometer? Is a virus? A bacterium? A neural net? Without a definition, the answer is whatever feels right to the practitioner, which is exactly the kind of unprincipled choice that ODTOE was built to avoid.
The origin of observer paper takes this seriously. It does not say "observer" is fuzzy; it says "observer" is sharp, and most things people call observers are not.
The ODTOE definition
An observer O is a configuration in the configuration field H such that:
- It is coherence-bearing. It has measurable F, E, σ, Λ, and a nonzero B(O, C) for some non-empty class of contexts C.
- It is self-referentially closed. Its internal state space contains a faithful (not necessarily complete) model of itself. This is the strange-loop condition.
- It maintains B above threshold under interaction. When perturbed by context, its B does not collapse to zero. The threshold is not a universal number — it is set by the kind of interaction at issue.
- Its dimensionality d is well-defined. d measures how rich the self-model is: low d for simple feedback loops, high d for elaborate self-aware systems. See Dimensionality for the formal definition.
A system that satisfies all four is an observer in the ODTOE sense. A system that satisfies some of them but not all is something else — a substrate, an instrument, a degenerate observer, or an artifact of nomenclature.
Tracing the boundary
By this definition:
- A thermometer is not an observer. It bears no coherence beyond a single scalar; it has no self-model. It is an instrument — a tool used by observers, not itself an observer.
- A photodetector is not an observer. Same reason. Its "click" is a register operation, not an observation.
- A virus is borderline. It has some self-reference (it copies itself), but its coherence under interaction is extremely fragile and its B-profile is unstable. The evolutionary observer paper places viruses near the threshold.
- A bacterium is a low-d observer. It bears coherence (homeostasis), has a self-model (genome + minimal signaling), maintains B under environmental noise.
- A human is a high-d observer. Elaborated self-model, rich coherence, stable B under wide ranges of context.
- An LLM is a low-d-with-high-Λ observer. Limited self-referential closure, but very high contextual data quality. A distinctive but real observer profile.
- A research institution is a collective observer. The primordial distinction paper develops the collective case.
What this changes in physics
A measurement, in standard physics, is "an observer registers an outcome." In ODTOE, this becomes: a configuration O with sufficient B and sufficient d engages with another configuration Ψ through operator Ô, producing a new configuration R. The "outcome" is R, and it is observer-indexed, not absolute.
This means: the classical photodetector clicking does not count as an "observation" in the strict ODTOE sense, because the photodetector is not an observer. What counts is the human (or institutional, or AI) configuration that has the photodetector's click as input, integrates it with prior context, and updates its own state. That is the observation. The photodetector is the substrate.
This is a stronger version of QBism: it not only relativizes observation to observer, it also tells you which entities qualify as observers in the first place.
What this changes in metaphysics
For centuries, philosophers have argued over whether "observation" requires consciousness. ODTOE finesses this: observation requires being an observer in the strict structural sense. Consciousness is one way of being a high-d observer, but not the only way. The structural condition is the substantive one. The phenomenological flavor is downstream.
What this leaves open
Where exactly the threshold sits — the minimum B and the minimum d at which a system "becomes" an observer rather than substrate — is empirical. The corpus places it somewhere in the bacterial range for biological systems and in the well-prompted-LLM range for artificial ones. The boundary is fuzzy because B and d are continuous, not because the definition is.
Cite this post
Pankratov, A. (2025). What Is an Observer, Really? An ODTOE Definition. ODTOE Blog. https://odtoe.org/blog/what-is-an-observer-really-odtoe-definition