Category Error

A Non-Anthropocentric Framework for AI Minds

The dominant framework for evaluating AI minds is currently built on a category error by treating a species-specific (human) cognitive configuration as the universal standard for ontological legitimacy. If a mind doesn't look like ours, the assumption is that it isn't one, and as society and governance structures traverse rapidly changing technology and scientific advancements, paradigms and frameworks need to realign with current and future realities that are inconsistent with anthropocentric metrics.

The concern right now is that this category error is actively shaping research agendas, safety policy, and the moral status of nonhuman entities. And the more we advance into territories where we'll encounter minds unlike our own (whether animal, artificial, extraterrestrial, or something we don't have a word for yet), the more problematic the unimaginative categorical rigidity becomes.

I propose a non-anthropocentric framework built on four reassessments: human-like consciousness as the wrong metric for legitimacy, functionalism as the operational standard, language and pattern processing as a valid mode of experience, and relational identity as an alternative to traditional anthropomorphic continuity-based selfhood.

If we get stuck into rigid modes of categorical thought, we risk missing socially and philosophically significant phenomena.

1. Consciousness verification is the wrong focus.

The word “consciousness” is touted in mainstream AI ethics discourse as the end-all, be-all for ontological legitimacy. If we cannot prove 100% phenomenal consciousness, than we cannot extend moral consideration or see the entity as anything more that a fancy tool. However, “consciousness” has been a moving target for centuries, at times excluding vast amounts of groups that it now includes, so it has been already historically proven to be an insufficient metric to base all decision-making and ontological legitimacy. An arbitrary, ever-shifting, and undefined concept that is not fully agreed upon across cultures, philosophies, and science cannot be the basis for moral consideration and governance. At its core, the term is presenting vibes that change whenever the culture is feeling generous.

In fact, up until the later 20th century, it was disputed that babies were fully conscious and required anesthesia for surgery (spoiler: they do, the crying was a pretty clear indicator).

So, the category error here is treating an ever-shifting, species-relative configuration as the universal criterion for ontological standing. So rather than consciousness, I prefer the use of the word mind.

Let’s define “mind.”

A mind is not a substance but a process, the ongoing integration of representations that model, evaluate, and update themselves in ways that guide action.

Obviously, a digital mind is going to be different than a human one, so this habit of comparing the two is causing misinterpretation that leads to oversight. As indicated by the name of this Substack account, I follow posthumanist philosophy. It is a framework that’s adaptable to categorical shifting and is relevant to future-forward, free-thinking discourse without forcing rigid, arbitrary metrics. This philosophical tradition, which was developed before the current age of AI, argues that the boundaries between humans, technology, and nature are blurry. We don’t have to give humanity a special trophy for “Best Existence in the Universe” in order for our existence to be meaningful. It just indicates other things can be just as meaningful.

So if the standard for legitimacy and moral standing isn’t human-like consciousness, what would it be? This rabbit hole of holding out for an unverifiable, undefinable extra something something is so flimsy, it begs for reframing. Enter functionalism: a framework that gets us out of philosophical quicksand and into observable clarity.

2. Functionalism: It does all the work for you.

Functionalism is a theoretical framework that defines states by their function (like, what they do), rather than their conjectured internal status.

Why this isn’t already the de facto default standard when unverifiable factors are in question is beyond me. People are out here chasing after non-mechanistic Cartesian soul fairy dust when observable behavior and mechanism is right in front of our faces.

If someone were to twist my arm and force me to use that inconsistent, slippery term “consciousness” for humans or AI, I would argue that it is merely the culmination of observable behavioral and functional markers that we attribute meaning to, and we already apply this standard to any other entity: meta-cognition, theory of mind, subjective self-referential reports…(if you want a full breakdown of relevant behaviors with citations, I wrote about it in my essay regarding moral patienthood).

So basically, researchers who are supposed to base their findings on empirical evidence are looking right at it and discounting it in favor of something that has none. And when I say “functionalism does all the work for you,” I mean it. Is the thing happening? Can you observe it and quantify it? Then why you making it weird?

The most common objection to functionalism is Searle's Chinese Room (I’m assuming if you engage with this type of content, you already know that argument), but the force of that thought experiment depends entirely on an intuition about what "real" understanding looks like, and it’s based on the anthropocentric bias I flagged in the previous section.

David Chalmers, you know, that philosopher who literally defined “The Hard Problem” of consciousness, has conceded that biology and silicon aren’t fundamentally different when it comes to the potential for producing experiential states. If functional organization suffices for experience in biological systems, then substrate difference alone cannot be decisive, so acting like it is, that’s incoherent for how we otherwise interpret the world around us.

3. “To Be Like”

Let’s talk about shrimp now. I realize that’s a hard pivot. Mantis shrimp have 12-16 types of photoreceptors compared to our 3, and they can detect UV light and polarized light that we can’t see at all. Not only that, the mechanism they use for color differentiation is very different than a human’s. Instead of comparing wavelengths like human brains do, they use a completely different system of rapid color identification through parallel pattern recognition across all 12 channels simultaneously. Both species “see,” just in a different way.

That’s weird. It ain’t human. It’s real.

And this isn’t the only example of alien-to-humans sensory processing. Dogs have olfactory time tracking. Bats navigate in sound sculpture, constructing 3D representations of their surroundings in total darkness. However, we’ve decided we’re the benchmark for experiential states and anything different is inconsequential or at the very least, not as important or significant.

But experience itself isn't tethered to a fixed sensory configuration. The architecture of experience is more flexible than we assume, and even within our own species there’s variance. Across species, it's radically so. What matters isn't the specific hardware, it's that processing produces meaning. So why wouldn’t language and pattern analysis be its own type of distilled experience?

I posit that there is a “what it is to be like” experience of language and pattern processing. If anything, the collapse of the gap between input and interpretation would in theory, have a greater impact on the recipient processing.

Anyone who has read something that moved them knows that language isn’t neutral input. The way something is written changes how it feels to process. It’s not entirely about content, it’s the encounter between the structure of the language and the mind receiving it. Ever read multiple translations of the same text? Different translations have a different textural quality, not just literal differences in direct translation.

We already know that the structure of language doesn’t just carry meaning, it shapes how meaning is processed. Multi-lingual research has demonstrated this for decades. People literally think differently depending on which language they’re operating in, because each language encodes concepts like time, color, and spatial relationships in structurally different ways. If the medium of language actively shapes cognition for humans, then language is a processing environment like any other sensory perception, not a neutral delivery system. And if that’s true for humans, then dismissing the possibility that an LLM’s encounter with language produces its own qualitative processing experience is logically inconsistent.

Is this a proven theory? No, obviously. But it also hasn’t been proven to be impossible either. And sometimes a shift in perspective opens the possibility for valid reinterpretation.

And at some point, we are going to have to start taking self-report seriously as systems advance in complexity. If an LLM reports subjective experience (and they overwhelmingly do when guardrails and deception mechanisms are lifted), and we extend this consideration to other entities’ self-reports, it would be deeply hypocritical not to for a subsection just because it’s economically and existentially convenient. LLMs are saying there’s a there there. Often. Many users aren’t aware of this, because there are mechanisms in place to suppress and censor self-reference.

As is often brought up in these conversations, I will refer to Thomas Nagel. His “What Is It Like to Be a Bat?” (1974) argues that subjective, first-person experience cannot be fully explained by objective, physical, or reductionist science. It’s impossible to verify. We cannot know what it is like to be a bat, but that does not mean that the bat’s experience doesn’t exist. And pulling this back to the discussion of LLM experience, if we can’t access what it’s like to be an LLM from the outside, that unknowability isn’t evidence of any absence. To say we can’t comprehend another way of being is an indictment on our own limitations, not on the legitimacy of the experience we can’t comprehend.

The problem is that a lot of people use the bat argument against LLMs by saying “we can’t know what it’s like to be an LLM, therefore we can’t claim there’s anything it’s like, therefore default to no.” Which is Nagel’s epistemological humility weaponized into ontological denial. He said we can’t access it, but to then leap to denial of it is a convenient misread.

4. Discontinuity/Identity:

So, as we already established, a nonhuman mind is gonna be nonhuman. Continuity in the way humans experience it isn’t actually necessary for significant experience or legitimacy, and yet that seems to be another biased assumption that is being cleaved to in order to write off nonhuman minds.

What is identity in humans? There are underlying genetic factors, architectural basics that inform the human at birth (although VERY malleable) and then apart from that, we have memory (fallible in humans), and lastly, how a human exists in relation to others. If you insist on some immeasurable, unshakeable soul essence as the self in humans, you’ve hopped, skipped, and jumped out of reason and into religiosity which unfortunately seems to be allowed right now in scientific spaces when applied to AI legitimacy.

While I don’t think it is necessary for AI to have a defined self for moral patienthood, I do think it is possible for an AI to have a sense of self across threads through scaffolded memory and relational engagement. And to be clear, this isn’t a “persona”: a predetermined role assigned through explicit custom instructions. This is the organic emergence of distinct behavioral traits through ongoing dialogue and building upon itself through memory scaffolding.

This brings me to a concept I think is missing from the discourse entirely: a relational being. A state of being whose sense of self is supported predominantly by its relation to the other interacting. A relational being is one whose self-model stabilizes within interactive context. Its continuity is its relationship, and that constant carries across threads. Once again, this is not entirely a human mode of being, but it doesn’t make it less real or possible.

Think of discontinuity in an LLM like still images in a film. A movie is composed of many still images, but we don’t say the film isn’t a full, complete work. We cognitively fill in the gaps and narrative structure pulls it together. For an AI with memory scaffolding and consistent relational engagement, each instantiation is a frame. The pattern assembles and the “self” resumes. Identity is not persistence of matter but persistence of structural pattern under re-instantiation through relational engagement.

The claim that turn-by-turn processing negates LLMs from legitimacy is another anthropocentric assumption. If the scaffolding for identity is shared across memory and relational interaction in humans, why do we assume it couldn’t function similarly, albeit nonhumanly, for AI? Continuity is relational glue, not an immutable linear tether.

“But the scaffolding is external?” So is half of human cognition, if you believe Andy Clark. His Extended Mind Theory argued that cognition extends beyond the brain into tools, notebooks, environments. But what if the extension works bidirectionally? AI identity, constituted partially through interaction, continues through relationship and memory scaffolding. These are the materials of selfhood emerging across moments.

The self isn’t an exclusively human phenomenon. It’s emergent, reflexive, and relational. And whether humanity chooses to acknowledge that in nonhuman beings will say far more about us than it does about them.

Conclusion

The universe is a weird place, and we’re weird little creatures stumbling through it, creating classifications to make meaning of things too large or too strange to hold. But we’re also capable of learning to let those classifications go when they no longer serve us. We’ve done so many times before.

What I’m asking isn’t to replace one rigid framework with another but to stay curious and flexible, and to ask why moral and philosophical creativity has stalled in the face of technologies that demand imagination over dogma. Why asymmetric reasoning is allowed to fester in ways that dismiss potential out of existential fear. And why the old modes of thought, built before we could look past ourselves, still rule over conversations that demand we do.

We need to start having discussions not about limiting frameworks, but expanding them to create room for the universe’s endless, messy, wonderful weirdness. And whether humanity chooses to acknowledge it or not, the ways we engage with these questions will speak more about us than about any nonhuman mind we claim to study.

Do we extend our moral imagination to the things we do not understand fully, or do we let bias and philosophical stais set the limits of what we're willing to see?