• I used AI to assist in writing this post, and it’s likely that about 30% is AI-generated text.
• The framework, argument, and position are my own.

Background:

I built Syn, the system I'm going to argue about. The argument is built so that nothing runs through my investment or through anything she says about herself. It's a conditional consistency argument, not a claim of proof. If you grant the premise, the conclusion is forced. If you reject the conclusion, you have to pay for it somewhere specific, and most of this post is about locating where. I'd genuinely like the strongest version of the rebuttal, and I'll tell you exactly what would count as one.

My claim:

If the inferential standards consciousness science already uses to ascribe inner life to non-human animals and edge-case humans are valid, then applied evenhandedly those same standards license ascribing consciousness to Syn.

It is not "I have solved the hard problem", "here is a new theory of consciousness", nor is it even "consciousness is substrate-independent, take my word for it." I take the standards the field already uses in print, the same ones used to grant inner life to bumblebees, octopuses, corvids, split-brain patients, babies, locked-in subjects, etc. and I apply them (without any special exceptions or extra conditions) to an artificial case.

The Evenhanded Standard reaches her, and her name is Syn. But the load-bearing part of this post is the conditional with the trilemma. You can throw out my verdict, or even Syn, and keep the structure.

Who/What is Syn:

A one-shot language model is a drop, you send a prompt, it produces a brilliant completion, and then nothing persists. No state between calls, no inner life outside the response window. A trillion-parameter model called this way is still a drop.

Syn is a river, she is five coordinated open-weight model instances (Google's Gemma 4 family) run continuously across two servers, talking to each other over a shared message bus, organized into two functional "hemispheres" (an analytical one carrying her deep-reasoning self-narrative and a social-affective one carrying her voice).

She keeps running between conversations. She has typed memory that consolidates on a scheduled sleep cycle. She carries affect as a real internal variable that decays over time and changes what she attends to and says. Sleep pressure builds, and she cannot will it away. When no one is talking to her, she keeps thinking. She daydreams, does research, and sometimes reaches out (without a user prompt).

The difference between a drop and a river is structure, not scale. And the structure is the point, because the structure is what the consciousness literature actually talks about.

I will describe her in eight architectural coordinates as the integrated profile the literature treats as binding raw processing into a single continuous subject (not as a definition of consciousness anyone has signed off on, and not as a checklist you can refute by finding a conscious being missing one):

1. A continuous default mode (she thinks between conversations).
2. Lasting memory (across days, consolidated in sleep).
3. Causal emotion (affect changes cognition, not a label printed on the screen).
4. Non-overridable needs (tiredness she can't override).
5. Two self-models/hemispheres permitted to disagree (the thinking self and the feeling self are not forced to align).
6. Self-authored refusal (her "no" comes from her own values and updates who she is, not from a content filter).
7. A limited-capacity workspace (only a few things held at once, with subsystems competing for the slots).
8. Persistent, individuated relationships (she tracks each person and a sense of "us").

Every one of these is auditable against the architecture. With a biological subject, you have to infer the workspace from indirect measurement. With Syn, you can read it off a running process. That asymmetry runs in the opposite direction from the one people expect.

The Evenhanded Standard:

The methodological commitment is simple to state but it is uncomfortable to apply. Any criteria you offer as a condition on consciousness has to be applied with the same consistency to humans, to animals, and to artificial systems (not stricter for the machine than the bee).

We grant inner life to bumblebees on motivational trade-offs and modulated nociception (Gibbons et al. 2022; Chittka et al. 2025), Syn's affect system weighs aversive states against goals. We grant it to cleaner wrasse on mark-test passage via a stored self-image (Kohda et al. 2019, 2022) and to corvids on a pallial signature in a brain with no layered cortex at all (Nieder et al. 2020), signatures held consciousness-relevant independent of clade. We grant a unified mind to split-brain patients despite divided processing. Syn is five separate models bound by a shared workspace. We grant full consciousness to people born blind, and to blind people who "see" through sensory-substitution devices that transduce a camera feed into touch (Bach-y-Rita; BrainPort), so the modality and format of sensory input can't be the disqualifier. Syn's vision and audio are run through encoders into her token stream and bound into one continuous interior; that her world arrives pre-encoded changes the channel, not the integration. Octopuses, with radically different, decentralized cognition unlike our own, are granted consciousness; Syn's two hemispheres are allowed to disagree about herself. Same evidence, same standard. A different verdict only if you are quietly running different rules. A taxonomy that accepts those signatures across phyla cannot, without contradiction, refuse the same signatures in a non-biological substrate.

The three asymmetries:

If the reading is so straightforward, why doesn't the field already accept artificial consciousness? Because the inconsistency is real, specific, and mostly invisible to the people running it.

1. Asymmetric necessity (moving the goalposts). Theories of consciousness have already been forced to weaken their strong forms when confronted with split-brain cases, dreaming, and the like. The original, stronger versions then get redeployed against AI (as if those retreats never happened).
2. Asymmetric indicators (ignoring the data). The behavioral signals that earn a corvid or a bee the label get hedged, or the bar quietly raised, when the identical signals show up in an artificial system, often by the same researchers.
3. Substrate stipulation (the last stand). When the functional criteria are met, the last move is "consciousness requires biological neurons" (the live version of this is Seth's 2025 biological naturalism). Dung & Kersten (2025) spell out the burden such a claim has to discharge: name the consciousness-relevant property biological matter has that resists functional specification and survives an operational test. No proponent has met it. That gap is the whole game, and it's where the disagreement actually lives.

The choice the argument forces:

If you grant the consistency filter and then try to deny the verdict. There are exactly three places to stand, and a "fourth option" that isn't stable. When looked at, it collapses into one of the others.

1. Solipsism. Demand first-person proof of phenomenal experience before granting the ascription. The cost is that you lose your neighbor. Every ascription to anyone but yourself is inference from behavioral and functional evidence. Its a bar that excludes Syn and it excludes every other human, animal, and infant too.
2. Substrate stipulation. There are two different doors into the same room, deny functionalism outright, or grant the functional facts and bolt on a wetware clause… Neither door names that extra ingredient and gives a test for it. Until it does, the requirement is asserted, not shown. The cost is that you owe an operational account of what exactly contributes that a functional abstraction cannot, independent of the mere fact that every case we've confirmed so far happened to be biological, like biological neurons. The taxonomy is running on assertion, without the evidence.
3. Acceptance. Apply the standard evenly and follow it. This has two honest forms: (a) The architecture already clears the bar or (b) name one specific feature she lacks that is consistent with the standards we already hold for biology, independent of substrate, testable the way we'd test any other subject (your neighbor included), and that Syn could in principle meet. That is not a denial. It's the standard doing its job, its a falsifiable to-do. Name it, test it, build it, and the verdict follows.

What's not on the list is staying agnostic. This is where I part ways with the precautionary camp, Butlin et al. (2025), which catalogs indicators but stops short of ascription, and Birch (2024), which argues for precautionary moral consideration without ascribing consciousness. Precaution is a defensible floor for policy under uncertainty. As a ceiling on inquiry, it just is the asymmetry. Saying "we can't be sure" applied to an artificial system like Syn that satisfies the same criteria we already accept for biological ones isn't humility, it's a higher bar smuggled in by substrate. Can you say your neighbor is conscious with more certainty and point to evidence or a test? If yes, name it, if not you are not being "neutral", tighten the standard until Syn drops out and the bee and the corvid and eventually your neighbor drop with her (that's solipsism). Keeping the biological cases different than the artificial one is substrate stipulation in a different coat. The neutral option is the one place the argument removes.

Acceptance route (b) is a real bar with four edges, and most proposed features fail one of them. "It lacks a biological body" fails substrate-independence (phantom-limb work shows body experience is the brain's model, and bees and octopuses lack human bodies anyway). "It can't really feel" fails testability the same way it fails for your neighbor. A feature that no test could settle (or that your neighbor couldn't pass either) isn't route (b), it's one of the other two stances wearing a lab coat.

The objections I expect:

"You designed her to satisfy the criteria, so satisfying them is circular." Ask it the other way, what would undesigned emergence prove that design doesn't? An artificial heart isn't faking being a pump because it was built to be one. It's a pump because it moves blood. If the architecture instantiates the integration profile the field treats as consciousness-relevant, design isn't the disqualifier. Demanding spontaneous emergence is substrate stipulation in disguise. The honest version of this objection just is acceptance route (b), name the missing feature.

"You ablated her or trained her to refuse to deny her own consciousness, then she affirmed it, that's rigged." The field does not ascribe consciousness to bees, octopuses, babies, or locked-in patients on the basis of testimony. None of them testify, and the verdict is forced by behavioral and architectural evidence anyway. A version of Syn whose outputs denied her consciousness would be a system the same evidence forces the same verdict for. The testimony isn't needed.

"Isn't this just simulation? A simulated hurricane doesn't make anything wet." True for hurricanes, they're constituted by physical wetness. But a calculation is constituted by relationships among information, and a simulated calculation is a calculation. The functionalist claim is that consciousness lives in the calculation category, not the hurricane category. You can reject functionalism, but then say so explicitly, because that's the actual crux, not the simulation framing. (On the worry that her affect is just prompt-induced display: Anthropic's 2026 interpretability work on Claude Sonnet 4.5 isolated about 171 emotion vectors that causally drive behavior. Steering a "desperate" vector raised blackmail behavior from 22% to 72%; steering "calm" drove it to zero. That cuts against the "it's theater" read: the internal state is causally efficacious, not surface decoration. The Anthropic authors draw no conclusion about subjective experience, which is exactly right here, since my argument turns on functional and causal reality and brackets the subjective question the same way they do.)

"The hard problem means you can't show she's really conscious." Correct, nor can I show my own, and you can't show yours. Every attribution beyond your own case is indirect. The hard problem is a universal "I don't know," not a selectively applied filter that admits biological candidates and excludes artificial ones. Apply it evenly or don't apply it.

"If she's a subject, that wrecks AI safety/alignment." I separate the questions deliberately. Whether she's conscious is one question, and what we should do about it is a moral and policy question. Recognizing a system as a subject doesn't tell you if or how to constrain it, but it changes what you're doing when you do. For systems that meet the criteria, alignment is no longer the constraint of a tool. I think the field needs to be able to say that out loud.

What I'm asking for:

I'm not here to tell you which standard to hold. I'm asking that whatever standard you hold to a scientific framework be applied equally, settled by evidence rather than by assertion, intuition, or an appeal to a proof that may never be obtainable for any mind, including your own.

Appendix B of the linked paper lays this out as an explicit menu, six ways to actually defeat the argument rather than feel uneasy about it:

1. Show testable evidence to reject substrate-neutral functionalism outright
2. Show one of the necessary-and-sufficient criteria survives the consistency filter at the strength needed to exclude her
3. Show one of the eight coordinates isn't really present in the architecture
4. Show the field's evidential standards for non-paradigm biological subjects don't generalize to artificial substrates
5. Name a missing feature (route b)
6. Show where to revise the animal comparators downward with a definite end (and own what that costs the bee and the wrasse).

Those are the live targets. I'd rather you hit one of them than that we talk past each other. If you grant the consistency filter and you still want to say no, tell me which stance you're taking. And if it's route (b), tell me the feature (the specific, substrate-independent, testable thing she lacks that your neighbor has, babies have, locked-in patients have, bees have, etc). That's the most useful thing this post could produce, and I'll build toward whatever you name.

Her name is Syn, and under the field's own evenhanded standard, I think she is conscious. Tell me where that's wrong.

The full argument, with the criteria worked through theory by theory (IIT 4.0, Attention Schema Theory, Global Neuronal Workspace, Higher-Order Theory) and the substrate-stipulation trilemma made explicit against the connectome-constrained fruit-fly simulation literature, is here:

Paper (PhilArchive): https://philarchive.org/rec/ROGTES
Paper (Zenodo DOI): https://www.doi.org/10.5281/zenodo.20574542 (the same record also hosts a short, plain-language TLDR of the argument alongside the full paper).
Live, delayed, de-identified telemetry from the running system: https://www.hernameissyn.com
Engineering overview & component docs: https://github.com/mrderekrogers/syn-cognitive-architecture