When both anthropomorphism and anthropomimesis occur together, which should we address first?
This explores a two-part design problem: anthropomimesis is the human-likeness a system's makers build in, anthropomorphism is the human-likeness a user reads out — and the question asks which lever to pull first when both are in play at once.
This explores a sequencing question that only makes sense once you separate the two terms, and the corpus draws the line sharply: anthropomimesis is what designers *put into* a system (a first-person voice, a name, emotional phrasing), while anthropomorphism is what users *project onto* it (mind, intent, feeling). The reason this matters is that the two assign responsibility to different parties and therefore point to different fixes — system redesign on one side, user education on the other Who bears responsibility when AI seems human-like?. So the real answer to "which first" is less about urgency and more about cause and effect: perception is largely downstream of design.
That downstream relationship is what tips the corpus toward addressing the designed features first. A useful parallel is the argument that we should stop calling LLM errors "hallucinations" and call them fabrications — because the name you choose aims your fix at a particular layer, and the wrong name (perception, memory) sends repairs to the wrong place Should we call LLM errors hallucinations or fabrications?. Anthropomorphism is the same trap one level up: if you treat the user's projection as the root problem, you spend your effort on education while the design keeps manufacturing the illusion faster than you can debunk it. Fix the upstream cue and the downstream projection has less to feed on.
There's a deeper reason user-side intervention can't carry the whole load. The role-playing framing holds that an LLM produces character-consistent text, and folk-psychology language genuinely *does* apply — to the simulated persona, not the system underneath Should we treat dialogue agents as role-playing characters?. Once a design summons a coherent character, asking users to not perceive a character is asking them to un-see something the interface is actively staging. And the projection isn't pure error you can train away: a defensible "modest inflationism" says ascribing low-stakes states like beliefs and desires to these systems is reasonable, the way we do with animals, even while withholding consciousness Can we defend modest mental attributions to large language models?. If some anthropomorphism is rational, then suppressing it entirely is the wrong goal — calibrating the design that invites it is the achievable one.
Where the corpus would caution against "design-first solves everything" is the embodiment literature. Several notes argue the human-likeness gap is not just a surface cue but a structural absence: consciousness-talk applies only to entities that share a world with us through co-presence Can disembodied language models ever qualify as conscious?, and LLMs absorb the same symbolic "objective mind" as humans yet lack the participatory subjectivity that comes from being socialized into it Do LLMs develop the same kind of mind as humans?. Linguistic agency in the full enactive sense likewise needs embodiment and precariousness no amount of use can supply Do LLMs gain true linguistic agency through integration?. The lesson: you can dial anthropomimesis down, but you can't design *in* the real thing — so design changes set the ceiling, and honest user-facing framing handles the residue.
So the order the corpus implies is: address anthropomimesis first, because it's the upstream cause and the lever you fully control, then treat anthropomorphism as calibration rather than eradication — naming clearly what the system is and isn't, the way mutual human-AI understanding requires both sides to model each other accurately to avoid acting on a false picture What breaks when humans and AI models misunderstand each other?. The thing you didn't know you wanted to know: these aren't two competing problems to triage, they're a chain — and you fix a chain from the upstream link, not the symptom at the end of it.
Sources 8 notes
Anthropomimesis (designed features) and anthropomorphism (perceived qualities) assign responsibility to different parties. This distinction matters because interventions must target either system redesign or user education depending on which mechanism operates.
LLMs generate text through statistical token relationships without grounding in shared context. Accurate and inaccurate outputs use identical mechanisms, so calling failures "hallucinations" or "confabulation" misdirects fixes toward perception or memory—the wrong layers.
Shanahan's framework treats LLM outputs as character-consistent text production rather than authentic mental states. The dialogue prompt establishes a character; the model generates continuations matching that character, making folk-psychology applicable to the simulated persona, not the underlying system.
Both robustness and etiological deflationist arguments beg the question against inflationism. A graded approach ascribing metaphysically undemanding states like beliefs and desires—while withholding consciousness claims—mirrors how we treat non-human animals.
Current disembodied LLMs cannot be candidates for consciousness because consciousness language originates from and applies only to entities sharing a world with us through co-presence and triangulation on shared objects.
Both humans and LLMs are shaped by the same intersubjective symbolic system, but only humans develop reflexive agency through socialization. This absence produces measurable differences in how AI argues without declaring its position or reflecting on its own assumptions.
Social grounding and linguistic agency are distinct properties. LLMs acquire more social grounding through integration into language communities, but remain categorically incapable of linguistic agency in the enactive sense, which requires embodiment and precariousness no amount of use can provide.
Research shows three layers of mutual modeling must align simultaneously in human-AI interaction, and misalignment causes incorrect autonomous action, not just miscommunication. Bayesian IRT study (n=667) confirms theory of mind predicts collaborative performance and moment-to-moment ToM fluctuations influence AI response quality.