The Evaluating Self

3Consciousness Across Three Worldviews

Swami Sarvapriyananda, Blaise Agüera y Arcas and Carlo Rovelli, Noema, 2025

Linked to 19 chapters: supports 16 · contradicts 6 · mechanism 5 · same claim 5 · pairing 4

Consciousness Across Three Worldviews, 2025 — one move in five steps Threevocabularies The sharedword: light The gapvivid vs. cold The bridge:selves, umwelt Nested selves;modest close Pure awareness ·the self-model ·mutual information. Light names theinformational role,not the photons. Awareness is vivid;Shannon informationis cold. Named, notclosed. Evolution draws aself/world line; whatmakes a difference issemantic information. Cell, bee, hive: whichself depends on thecoarse-graining. A single unsectioned essay of about 1,700 words; the steps are the reader's, the order is the authors'.
Figure 3.0 — A short essay with one move: three vocabularies, a shared word, and a bridge built through evolution.

Three authors, a Vedanta monk, a Google researcher in AI and artificial life, and a physicist working on the foundations of quantum mechanics, wrote this note after a gathering on consciousness in Venice. Its aim is modest and stated at the outset: not to reconcile three worldviews, whose differences the authors say they know well, but to look for a partial translation between them. The translation runs through a single word, light, and through an evolutionary account of how selves come to exist. The essay is short and reads in ten minutes; its interest is in how far the analogy can be pushed before it breaks, and the authors themselves mark the point where it strains.

Three vocabularies

In Advaita Vedanta, the individual experiencing self is a manifestation of pure awareness, which is described as self-luminous and universal, and which is not distinct from reality or from knowledge. There is no duality between the experiencer and Brahman, the ground of being.

Among computational neuroscientists and computer scientists willing to speculate, consciousness is characterised as information processing: we evolved to build models of the world, of ourselves and of others in order to regulate our internal state and guide behaviour, and consciousness is what it is like to have such a self-model. This applies to large-brained animals and, perhaps, to sophisticated enough machines; it is a recent and specialised phenomenon.

In the relational interpretation of quantum mechanics, physical systems are described by how they affect other physical systems. Two systems have mutual information, in Shannon's sense, when observing one tells you something about the other. Our own knowledge of the world is a special case of this, and reality as a whole can be read as the ensemble of relevant mutual information among all systems, ourselves included.

Light

The word the authors use to move between these is light. In everyday speech it names a narrow band of the electromagnetic spectrum, but the choice of that band is informational: those are the wavelengths we see by. Radio waves are not called light; an argument can shed it; we say "I see" on the telephone. The Vedanta texts use the word the same way. In the Bṛhadāraṇyaka Upaniṣad, Yājñavalkya answers the question of what serves as a person's light when the sun has set: the moon; when the moon has set, fire; when the fire is out, speech; and when speech has ceased, the Self itself is one's light. In the Bhagavad Gītā (13.18), Krishna calls it the light of lights, beyond darkness, existing in the hearts of all.

What serves as a person's light Bṛhadāraṇyaka Upaniṣad 4.3.2–7, as the essay quotes it the sun the moon fire speech the Self when it has set when it has set when it is out when it has ceased never fails Each item is a source of information about the world; the series is ordered by what fails first. The authors read the last term the same way: light as what one system can know of another. The same word links it to relational physics, where reality is mutual information among systems.
Figure 3.1 — Yājñavalkya's series. Each source of light fails in turn; what remains is the one that cannot be taken away. The word tracks the informational role, not the photons.

Read this way, relational physics is also about light: the information one system can have about another. So the self-luminous awareness of Vedanta and the field of relative information in relational quantum mechanics occupy corresponding places in their respective systems.

Where the analogy strains, and the bridge

The authors then name the gap. Vedanta's awareness has a vivid phenomenal character; Shannon's mutual information is cold and quantitative. Nothing about a number of bits says what it is like. The bridge they propose runs through evolution and computation.

In Vedanta the individual self is built by drawing a distinction, held in the end to be illusory, between subject and object. Evolution, in an analogous way, produces selves: processes that persist and replicate by drawing a distinction between self and world. That interface is physical, in its most basic form a cell membrane, and conceptual, defining what the organism can register and cares about: its umwelt, in von Uexküll's term, which has an inner side as well, hunger and pain and satiety.

Bateson defined a bit of information as a difference that makes a difference. The authors call this semantic information, to distinguish it from Shannon's combinatorial kind. The DNA of a live bacterium and of a dead one with the same sequence carry the same Shannon information; only in the live cell does the sequence have causal consequences. Any causal manipulation of information can be described, after Turing and von Neumann, as computation, and only information that is processed can be behaviourally salient. A far-infrared photon and a visible one both carry Shannon information; only the visible one is semantic information for a human observer, because only it can be perceived and acted on. Semantic information is someone's umwelt.

So everything a living subject experiences, and everything it persistently is, can be described as semantic information, which in the Vedanta sense is light. The authors allow themselves one flourish: in this sense we are all made of light.

Three selves in the same matter the hive years · umwelt: forage, season, threat the bee weeks · umwelt: flowers, sun, hive scent an immune cell days · umwelt: self, non-self Each is a living entity under a different coarse-graining. Each has its own umwelt: what makes a difference to it. Each preserves different information through time, which is what its identity is. Which one is "the self" depends on where the line is drawn.
Figure 3.2 — The same matter, three selves. An immune cell, the bee it lives in, and the hive each count as a living entity under a different coarse-graining, with a different umwelt, identity and lifespan.

The same move also yields the illusoriness of the self–other line. Under different coarse-grainings an immune cell, the bee it lives in, and the hive are each a living entity, each with its own umwelt, its own preserved information, and, as their lifespans show, its own identity. Which one is the self depends on the perspective taken.

What the translation leaves out

The essay is candid that it is a note on resonances, and its authors decline to overplay them. Two things are worth holding onto as a reader.

The bridge is asymmetric. It gets from Shannon information to semantic information, and from semantic information to a self with an umwelt, by way of evolution. That is a real step. It does not get from an umwelt to the self-luminous, phenomenal character of awareness that the authors identified as the sticking point. A live bacterium has semantic information; the essay does not claim it has experience, and the gap it named is still open at the end.

And the correspondence rests on a word used in two senses. Light as information and light as awareness are joined by the Upanishadic usage, which the authors read as informational. A Vedantin might reply that the texts mean awareness first and information only by extension, in which case the translation runs in the other direction. The essay's own closing sentence is the right frame: convergences and approximate translations are more useful than debating differences in first assumptions, provided one remembers they are approximate.

Claims drawn from this chapter · 10

opinionWhere the analogy strains, and the bridge

A self begins wherever something maintains a boundary against the world

Sarvapriyananda's lectures argue that evolution produces individual selves the same way Vedanta says they arise: by drawing a distinction. The boundary it draws is both physical, like a cell membrane, and conceptual, like an animal's sense of its own world.

Becausethe most basic form of the interface is a cell membrane, and von Uexküll's umwelt supplies the conceptual side, with an inner aspect of hunger, pain and satiety; the parallel is an interpretation, not a result.
LimitsVedanta holds the subject-object distinction to be finally illusory; evolution's boundary is as real as the membrane that draws it. The two are said to be analogous rather than the same, and the text this leans on treats the individual self as produced by ignorance rather than by selection.
PracticeProjection log — when you are tempted to ask whether a system has a self, ask instead what distinction it maintains and at what boundary.
After AIIt converts the question about machines from what they are into where their boundary is drawn and by whom. On that framing a model's boundary is drawn by its training and its context window, and by nothing it maintains for itself.
opinionWhere the analogy strains, and the bridge

An immune cell, a bee and a hive are each a self

Which thing counts as the self depends on the level of description you pick. An immune cell, the bee it lives in, and the hive the bee lives in are each a living thing with its own world and its own lifespan.

Becausethe immune cell, the bee and the hive are offered as an illustration of the point, with their differing lifespans as the argument
LimitsThree lifespans make a nice illustration and not an argument that the choice of grain is arbitrary; some coarse-grainings track causal structure better than others, and the essay does not say how to tell.
PracticeBefore arguing about whether something is one agent or many, fix the grain you are describing it at, and notice that the answer changes with it.
After AIThe same question is now live about a running system: the call, the session, the fleet. Which one is the agent is a choice of grain, and the accountability follows the choice rather than the other way round.
opinionLight

In the Upanishads light means what you see by, not brightness

In the Upanishads the word light names whatever you see by, not a band of the spectrum. When the sun, the moon, fire and speech have all gone, the Self is what one sees by.

Becausethe reading is built on Yājñavalkya's series in the Bṛhadāraṇyaka and on Gītā 13.18, offered as an interpretation the authors say is approximate
LimitsThe essay's own closing qualification applies here first. A Vedantin can run the translation the other way, with awareness as the primary sense and information the extension, which would make the correspondence an artefact of the reading.
PracticeRead the word in its informational sense wherever the texts use it, and test the reading against the one place it has to hold: what one sees by when nothing else is left.
After AIThe role the word names — what makes the difference that can be acted on — is the same role information plays in an account of a model's world. That is what makes the analogy tempting and what makes it easy to overdraw.
opinionWhere the analogy strains, and the bridge

Living things become selves by drawing a boundary at the cell membrane

Sarvapriyananda's lectures argue that evolution makes selves in the way Vedanta describes: by drawing a line between self and world. The line is physical at the cell membrane and conceptual in the creature's own world of what it can sense.

Becausethe bridge is built from von Uexküll's umwelt extended to the inner side, hunger and pain and satiety, as an argument rather than a measurement
LimitsThe step from a membrane to an umwelt is a real one; the step from an umwelt to experience is not taken, and the authors say so. The parallel with Vedanta's illusory subject–object distinction is offered as a resonance, not an identity.
PracticeWhen you are asked whether a system has a self, ask instead what it is registering and what it treats as its own boundary, and see whether the question survives.
After AIA model has no membrane and no metabolic stake, which is exactly the difference Seth makes load-bearing. The essay's route to a self runs through staying alive, and that route is closed to anything trained rather than grown.
opinionWhere the analogy strains, and the bridge

Sarvapriyananda argues information matters only when it belongs to someone

Information that is merely present and information that belongs to someone are two different things, and only the second has any effect on the system holding it. The DNA of a live bacterium and of a dead one are identical as information.

Becausethe DNA of a live bacterium and of a dead one with the same sequence carry identical Shannon information and only one has causal consequences, and a far-infrared photon and a visible one both carry information while only the second is semantic for a human; the criterion is Bateson's difference that makes a difference, adopted rather than tested.
LimitsThe criterion is circular at the edges: what counts as a difference that makes a difference depends on which system you have already decided to treat as a subject. Bender's argument runs alongside it, that form without communicative intent is not meaning.
PracticeIf you believed it you would ask, of any output, whose umwelt it belongs to — for whom it makes a difference — before asking whether it is correct.
After AIIt is the cleanest available statement of what a model's output is not. The text on the screen is information present; whether it is anyone's depends entirely on the reader, which is where the causal consequences actually land.
opinionWhat the translation leaves out

The essay's own ending admits it never reaches felt experience

The bridge runs one way only. It carries you from Shannon information to a self with its own world of meaning (an umwelt), and stops before the felt character of awareness that the authors themselves named as the sticking point.

Becausethe essay's own closing says so — a live bacterium has semantic information and the essay does not claim it has experience — and the gap it named is still open at the end.
LimitsAn admission by the authors rather than a demonstration that the bridge cannot be built. Nothing here shows the gap is unbridgeable, only that this essay does not bridge it.
PracticeIf you believed it you would refuse to let an argument about information carry a conclusion about experience, in either direction, including the direction that denies it.
After AISeth refuses the same question as a warning against projecting an interior onto a machine; this essay refuses it as an admission that its own bridge does not reach one. They decline the same question for opposite reasons and give opposite advice about what to do next.
contestedWhat the translation leaves out

The information account explains what a creature registers, not what it feels

The essay builds from Shannon information, through evolution, to a creature's own world of meaning (its umwelt). It never reaches what awareness actually feels like, which it had named as the sticking point.

Becausethe authors concede it themselves: a live bacterium has semantic information and no claim is made about its experience
LimitsContested because the essay's own authors would place the gap differently: one of them takes semantic information as the candidate for closing it, which would admit a bacterium, and the reader is left without a criterion.
PracticeAccept the translation as far as the umwelt and stop there. Do not let a good account of what a system registers stand in for an account of what it is like to be one.
After AIThis is the gap every claim about machine experience has to cross, and neither vocabulary crosses it. The information story reaches an umwelt and stops; the Vedanta story starts on the far side and never has to cross.
contestedWhat the translation leaves out

The match between the texts and information theory turns on one word

The whole correspondence rests on a single word taken in two different senses. A Vedantin can run the same translation backwards, with the texts meaning awareness first and information only by extension.

Becausethe essay's own closing frame concedes that approximate translations are useful only so long as one remembers they are approximate
LimitsContested by construction: which sense is primary is the whole question, and neither side can settle it from inside its own vocabulary. The essay does not claim to settle it.
PracticeWhenever a translation between vocabularies turns on a single shared word, run it in the other direction and see whether it still holds.
After AIThe same one-word bridges are now doing heavy work elsewhere — learning, memory, attention, reasoning — each joining a technical sense to an older one. The test is the same: run it backwards.
contestedLight

Vedanta's light of awareness and physics' shared information play the same role

Sarvapriyananda's lectures put Vedanta's self-luminous awareness — that which shows everything else — in the same position that relational physics gives to a field of shared information between systems. In both, “light” means making something known, not a band of the spectrum.

Becausethe support is Yajnavalkya's series of failing lights ending in the Self, Gita 13.18's light of lights, and ordinary usage — an argument sheds light, we say "I see" on the telephone — and the authors call it a partial translation and decline to defend more.
LimitsThe correspondence rests on one word used in two senses. A Vedantin can reply that the texts mean awareness first and information only by extension, in which case the translation runs the other way; the essay's own closing sentence concedes that convergences are approximate.
PracticeIf you believed it you would stop treating the two vocabularies as rivals and start asking which questions each can answer — while refusing to let the correspondence carry any weight the authors themselves decline to put on it.
After AIIt is the most careful of the essays that put Vedanta beside a physical theory, and its care is the point: it is offered as a note on resonances by authors who decline to overplay them, which is the opposite of how such correspondences are usually deployed.
opinionWhere the analogy strains, and the bridge

Whether something is one self or many depends on the level chosen

Which thing counts as the self depends on how coarsely or finely you choose to describe the system.

Becausean immune cell, the bee it lives in and the hive each count as a living entity with its own umwelt, its own preserved information and its own lifespan; the example is an illustration of a choice of description rather than a finding about selves.
LimitsThe text this sits beside holds that the witness is the same in everyone and not a matter of description at all, which is a direct contradiction rather than a nuance. A choice of coarse-graining explains why the self-other line is unstable, not what is on either side of it.
PracticeProjection log — record the level at which you were describing a system when you attributed intent to it, and check whether the attribution survives a level up or down.
After AIAsking whether a model, an agent loop, or the whole deployment is the thing you are dealing with is the same question at a different level, and most disputes about what an assistant "wants" dissolve once the level is fixed.

All claims · Source map