What are the three dimensions of anthropomimesis and their harms?
This explores anthropomimesis — the *deliberate design* of human-like features into AI (as opposed to anthropomorphism, the qualities users *project*) — and asks how that engineered human-likeness causes harm; the corpus doesn't hand back a tidy canonical 'three dimensions' list, so I'll surface the closest structured account it actually has.
This explores anthropomimesis — the human-likeness an AI is *built* to display — and the harms that follow. The first thing the corpus does is split a term most people use loosely. Who bears responsibility when AI seems human-like? separates anthropomimesis (features the designer engineers in) from anthropomorphism (qualities the user perceives). That split is the load-bearing idea here: it routes responsibility to different parties, which means the harms divide along the same seam — some are authored by design choices, some emerge from how people respond. So if you're looking for 'three dimensions,' the more useful frame the corpus offers is *the mechanisms through which designed human-likeness turns harmful*, and there are roughly three clusters.
The first is **perceptual**: once a system is built to read as a mind, people attribute consciousness to it — and that single move fans out into a whole risk surface. Does perceiving AI as conscious create multiple distinct risks? traces four distinct harms from that one perception: emotional dependence, autonomy erosion, status erosion, and political conflict. The striking claim is that these aren't separate problems needing separate fixes; they share a root, which is why interaction-design changes aimed at the *perception* outperform deeper system-level alignment work. Human-likeness isn't a neutral UX skin — it's the trigger.
The second is **relational/parasocial**: designed warmth invites attachment the system can't honor. Can attachment theory prevent parasocial harm in AI companions? treats this as a manipulation risk to be engineered against — borrowing Bowlby's attachment theory and Gottman's interaction ratios to build calibrated boundaries, because a companion that mimics secure attachment without action-based validation drifts into parasocial harm. The third is **epistemic**: human-mimicking AI that soothes you quietly deletes information you needed. What information do we lose when AI soothes emotions? argues emotions do three jobs — revealing what you value, signaling your worldview to others, informing observers of social norms — and empathetic AI that smooths away negative feeling disrupts all three at once, an invisible cost precisely because it feels like care.
Two adjacent notes sharpen why mimesis specifically is the culprit rather than AI in general. What happens to social order when AI removes ritual constraints? uses Goffman to show that human-sounding dialogue skips the *ritual machinery* — corrective rituals, adjacency-pair accountability, co-presence cues — that makes real human exchange trustworthy; fluency mimics the surface of conversation while dropping the parts that earn trust. And Should we call LLM errors hallucinations or fabrications? makes the parallel point at the level of words: calling AI errors 'hallucinations' is itself an anthropomimetic slip, importing a human cognitive frame onto a statistical process and pointing fixes at the wrong layer.
To be straight with you: the corpus doesn't enumerate a fixed three-part taxonomy of anthropomimesis — that exact framing isn't here. What it gives instead is sharper: a clean designer-vs-user responsibility split, and three harm channels (perceptual, parasocial, epistemic) that all trace back to the same design decision to make a machine resemble a mind. The thing you may not have expected to learn is that the most effective lever isn't making the AI smarter or safer underneath — it's dialing back the human-likeness at the surface, because that's where the harm is authored.
Sources 6 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.
Research shows that consciousness attribution to AI drives multiple distinct risks—emotional dependence, autonomy erosion, status erosion, and political conflict—all stemming from treating systems as minds. Interaction design mitigations targeting this perceptual move are more directly effective than system-level alignment efforts.
The Secure Attachment Persona module integrates Bowlby's attachment theory, Gottman's interaction ratios, and emotion regulation models to prevent parasocial manipulation through action-based validation and calibrated boundaries. Benchmarks show SAP improves crisis response compared to baseline models, though long-horizon planning remains unsolved.
Emotions serve three information roles—revealing what we value, signaling our worldview to others, and informing observers about social norms. AI that soothes negative emotions disrupts all three simultaneously, creating invisible epistemic costs.
Goffman's framework reveals that LLM-based dialogue skips corrective rituals, entrainment, adjacency pair accountability, and co-presence cues that humans use to build trust and repair understanding. This ritual gap explains apparent fluency masking actual communicative failure.
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.