Uncanny Valley¶

Human affinity for a human-like artefact does not rise steadily with realism but climbs, then plunges into a valley of unease in the narrow almost-human band, and recovers only when realism reads as genuinely human — because near-human stimuli recruit human-specific recognition routines that flag the remaining mismatches.

Core Idea¶

The uncanny valley is the hypothesis, introduced by roboticist Masahiro Mori in 1970, that human affective response to human-like artefacts does not rise monotonically with human-likeness. Instead, affinity increases with realism up to a threshold, then drops sharply into a valley of unease, eeriness, or revulsion for near-but-not-quite-human figures, and recovers toward high affinity only when realism is sufficient to read as genuinely human. Movement amplifies the effect: a near-human still image is mildly unsettling; a near-human animated figure with subtly mismatched motion is sharply more so. The structural shape of the curve — rise, sharp dip in a narrow "almost human" band, recovery — is the load-bearing abstraction; it makes the valley a design-relevant threshold phenomenon rather than a simple preference gradient. The proposed mechanisms involve mismatch detection between expected and observed cues across appearance, motion, and social-signal channels: the brain recruits human-specific recognition routines for near-human stimuli, and those routines detect anomalies (subtly wrong micro-expressions, implausible skin texture, mismatched motion dynamics) that coarser object-recognition would not, producing a heightened negative signal precisely because high-level expectations were engaged and violated. The effect was operationalised in animation studies (cited in post-mortems of Polar Express, 2004, and Final Fantasy: The Spirits Within, 2001) and informs deliberate stylisation decisions in character design: targeting either full abstraction or full photorealism, and avoiding the in-between band where the dip lives.

Structural Signature¶

Sig role-phrases:

the human-like artefact — a robot, avatar, animation, doll, or prosthetic representing a human (or living conspecific), the stimulus whose affinity is at issue
the realism axis — degree of human-likeness, the single continuous parameter along which the artefact is varied
the recruitment threshold — the point at which crossing into near-human territory switches on human-specific recognition routines that coarser object perception never engages
the mismatch detector — those engaged routines flagging anomalies (mis-timed micro-expressions, dead eyes, wrong skin, motion that does not track the face) precisely because high-level expectations were raised and violated
the non-monotonic curve — affinity rising with realism, dropping sharply into a valley across the narrow "almost but not quite" band, then recovering once realism reads as genuinely human
the amplifier terms — motion deepening the dip, and cross-modal inconsistency (photoreal face, cartoon voice) dropping an artefact in even where each channel alone would clear it
the safe-direction heuristic — the design consequence: inside the trough, retreat to full stylisation or push to full photorealism, never polish in place

What It Is Not¶

Not a monotonic preference gradient. The whole point is non-monotonicity: affinity does not rise steadily with human-likeness. It rises, drops sharply into a valley across the narrow "almost but not quite" band, then recovers — so the default design assumption that more realism always means more appeal is false in a specific, costly region, and inside the trough added realism makes things worse before they get better.
Not unease at realism as such. The eeriness lives in the inconsistencies — mis-timed micro-expressions, dead eyes, wrong skin, motion that does not track the face — not in fidelity itself. Crossing into near-human territory is what recruits the scrutiny that surfaces those flaws; remove the mismatch or push realism far enough to read as genuinely human and affinity is restored.
Not a settled empirical law. It is a hypothesis with a contested empirical base: meta-analyses find the dip in some populations and stimulus sets but not others. The load-bearing abstraction is the proposed curve shape (rise, dip, recovery), not a universally confirmed quantitative law of affect.
Not produced by non-human artefacts approaching realism. The valley appears only for stimuli that recruit human-specific (or conspecific) recognition routines along a realism axis — no one reports an uncanny-valley dip for a car or abstract object approaching photorealism, because those routines were never engaged. The trigger regime is the boundary the concept itself draws.
Not the general non-monotonic-response pattern. The uncanny valley is the human-perception instance of a broader family — non-monotonic response to a continuous design parameter — whose other members (phase transitions in matter, hormetic dose-response, the Yerkes-Dodson arousal-performance curve) share the rise-dip-or-peak form but each run on a different mechanism. The shape travels under non_monotonic_relationships and phase_transitions; this concept is one keyed to near-human artefacts.

Scope of Application¶

The uncanny valley lives across perception-and-design wherever a human-like (or conspecific-like) artefact is varied along a realism axis and recruits human-specific recognition routines past a threshold; these are different industrial settings of one trigger, not distinct substrates, so its reach is bounded there — the rise-dip-recovery shape travels further only under the non_monotonic_relationships and phase_transitions primes, and "uncanny valley of mind/voice" coinages are analogy that drop the perceptual trigger.

Robotics and human-robot interaction — Mori's original domain, an industrial-design heuristic for android development.
Computer animation — the diagnosed failure mode of The Polar Express and Final Fantasy: The Spirits Within, and the spur for deliberate Pixar/Disney stylisation.
Virtual agents, deepfakes, and generative video — near-realistic AI-generated faces and voices trigger documented uncanny responses, an ongoing concern in trust calibration for human-AI interaction.
Prosthetics and medical mannequins — a too-lifelike prosthetic hand or training mannequin can disturb users and clinicians.
Taxidermy, wax figures, and dolls — the long-standing folk-art intuition that the genre concentrates uncanny figures.

Clarity¶

Naming the uncanny valley overturns the default design assumption it was coined against: that affinity toward a human-like artefact rises with realism, so the path to a more appealing android, avatar, or animated character is simply more human-likeness. Once the non-monotonic curve has a name, that assumption is exposed as false in a specific, costly band — and the practitioner's question changes from "how realistic can we make it?" to the sharper "which side of the valley are we on, and is incremental realism moving us toward recovery or down into the dip?" This reframes a vague worry that a character looks "creepy" into a locatable position on a known curve, with a clear implication: near the dip, adding fidelity can make things worse, and the safe targets are the two extremes — full stylisation or full photorealism — not the plausible-looking middle.

The label also makes legible why the dip is sharp and what deepens it, which a flat preference-gradient picture cannot. By tying the unease to mismatch detection within human-specific recognition routines, it tells the designer that the eeriness lives in the inconsistencies — dead eyes, mis-timed micro-expressions, skin that reads wrong, motion that does not match the face — rather than in realism as such, and that crossing into near-human territory is precisely what recruits the scrutiny that surfaces those flaws. That yields concrete, otherwise-invisible diagnostics: motion amplifies the dip, so a still that passes may fail once animated; and cross-modal mismatch (a photoreal face with a cartoon voice, or vice versa) drops an artefact into the valley even when each channel alone would not. The concept thereby separates "not yet realistic enough" from "realistic enough to trigger the routines that catch its remaining errors" — two states that look alike from a monotonic vantage but call for opposite design moves.

Manages Complexity¶

Character and artefact design confronts an open-ended catalogue of "this reads as creepy" reports that resist a unified account: an animated lead whose photoreal skin and proportions land but whose micro-expressions are mis-timed; a service android that disturbs precisely as its face grows more lifelike; a deepfake face that unsettles where a cartoon would charm; a too-lifelike prosthetic or training mannequin that clinicians shrink from; wax figures and dolls that a whole folk tradition flags as eerie. Without an organising principle, each is its own design accident to be diagnosed and patched, and the field's tacit rule — more human-likeness yields more appeal — actively misleads, predicting that pushing realism will always help. The uncanny valley compresses the whole catalogue into one curve: affinity as a non-monotonic function of human-likeness, rising, dropping sharply into a narrow dip in the "almost but not quite human" band, then recovering. The sprawling question "why is this particular artefact unsettling?" collapses to "where on the affinity-versus-realism curve does it sit?"

That single relocation lets the designer stop enumerating creepy cases and track one continuous parameter — degree of human-likeness — plus a small set of amplifiers, and read the qualitative outcome off the curve's shape. The branch structure is explicit and actionable. Below the dip, on the rising stretch, an artefact reads as appealingly stylised and added realism helps. Inside the narrow near-human band, affinity is in the trough: here the counterintuitive branch bites, because incremental realism moves the artefact deeper before it recovers, so the correct move is to retreat to stylisation or push past to full photorealism rather than polish in place. Above the dip, sufficient realism reads as genuinely human and affinity is restored. The mechanism — human-specific recognition routines, recruited only once a stimulus crosses into near-human territory, detecting the mismatches coarser object perception would miss — supplies the amplifier terms the analyst also tracks: motion deepens the dip, so a still that passes can fail once animated; and cross-modal inconsistency (a photoreal face with a cartoon voice) drops an artefact into the valley even when each channel alone would clear it. So instead of modelling the full perceptual and affective response to every artefact, the designer locates it on one curve, checks the realism-match across channels and whether motion is involved, and reads off appeal, repulsion, or recovery — and, critically, which direction along the realism axis is safe to move.

Abstract Reasoning¶

The uncanny valley turns design reasoning about human-like artefacts into position-on-a-curve reasoning, where the curve is non-monotonic and a recruitment threshold marks the point past which human-specific scrutiny switches on. Every characteristic move depends on knowing both the artefact's location along the realism axis and whether that threshold has been crossed.

Diagnostic (locate the artefact, name the cause of unease). Faced with a report that an artefact "reads as creepy," the analyst infers its position: a negative affective response to a near-but-not-quite-human figure places it in the trough, not on the rising stretch. From there the inference runs to the specific cause — because the unease is produced by mismatch detection within human-specific recognition routines, the eeriness is read as living in the inconsistencies (mis-timed micro-expressions, dead eyes, wrong skin, motion that does not track the face) rather than in realism as such. This separates two states that look identical from a monotonic vantage but demand opposite responses: "not yet realistic enough" (still climbing toward the dip, below the recruitment threshold) versus "realistic enough to recruit the routines that catch its remaining flaws" (in the dip, past the threshold). The diagnostic move is surface revulsion → location in the near-human band + a specific cross-cue mismatch, never surface revulsion → "needs more realism."

Interventionist (move along the axis with sign-aware predictions). The signature intervention is direction-sensitive in a way the default assumption gets backwards. Inside the trough, adding fidelity is predicted to move the artefact deeper before any recovery, so the licensed moves are to retreat toward full stylisation or push past to full photorealism — both predicted to raise affinity — while polishing in place is predicted to worsen it. Each manipulation pairs a move along the realism axis with a forecast sign on appeal, and the sign flips depending on which side of the dip the artefact starts. The mechanism also supplies amplifier interventions with their own predictions: introduce motion and the dip must deepen, so a still that passed is predicted to fail once animated; mismatch realism across channels (photoreal face, cartoon voice) and the artefact must drop into the valley even where each channel alone would clear it. Removing the inconsistency, or aligning the channels, is predicted to lift it back out.

Boundary-drawing (fix the trigger regime). The curve applies only within a specific trigger condition — artefacts that recruit human-specific (or conspecific) recognition routines along a realism axis. That condition draws the regime: a near-photoreal human face is inside and shows the dip; a car or abstract object approaching photorealism is outside and shows no valley, because the human-specific routines were never engaged. The recruitment threshold is the internal boundary that matters most — below it the artefact reads as appealingly stylised and ordinary "more realism helps" reasoning is valid; above it the scrutiny switches on and the counterintuitive branch takes over. Naming the regime tells the designer when uncanny-valley reasoning governs at all and when a plain preference gradient does.

Predictive / branch-ordering. From one continuous parameter (degree of human-likeness) plus the amplifier checks (motion present? channels matched?), the qualitative outcome is read off the curve's shape before testing: below the dip, appealing and improvable by realism; inside the narrow near-human band, in the trough and worsened by incremental realism; above the dip, read as genuinely human and affinity restored. The forecast names not just the affective valence but the safe direction of travel along the axis.

Knowledge Transfer¶

Within perception-and-design the hypothesis transfers as mechanism, because every place it applies is the same trigger condition — a human-like (or, more generously, conspecific-like) artefact varied along a realism axis, recruiting human-specific recognition routines past a threshold. So the curve, the recruitment-threshold boundary, the amplifier terms (motion deepens the dip; cross-modal mismatch drops an artefact in even where each channel alone would clear it), and the sign-aware design heuristics (retreat to stylisation or push to full photorealism, never polish in the trough; match realism across modalities; test affinity before committing) all carry intact as the context changes: from Mori's original robotics and human-robot interaction to computer animation (the diagnosed failure of The Polar Express and Final Fantasy: The Spirits Within, and the deliberate stylisation of Pixar/Disney), to virtual agents and deepfakes and generative video, to too-lifelike prosthetics and medical training mannequins, to the folk-art cluster of taxidermy, wax figures, and dolls. These are different industrial settings, not different substrates — one trigger, many applications — which is exactly why the same reasoning governs all of them and the transfer is genuinely mechanistic.

Beyond that trigger condition the transfer splits into two cases worth marking separately. The first is analogy by metaphorical extension: coinages like "the uncanny valley of mind" or "of voice" lift the rise-dip-recovery picture into territory where the load-bearing trigger — human-specific perceptual recognition routines detecting near-human anomalies — is absent or far less crisp, so the mechanism that makes the visual valley sharp and predictable does not come along; the shape is borrowed, the engine is not. (Tellingly, no one reports an uncanny-valley dip for a car approaching photorealism, because the human-specific routines were never engaged — the regime boundary the concept itself draws.) The second is the genuinely broader structural pattern the uncanny valley instantiates: a non-monotonic response to a continuous design parameter, a real and widely-travelling abstraction whose other instances — phase transitions in matter, hormetic dose-response in pharmacology, the Yerkes-Dodson arousal-performance curve — share the rise-dip-or-peak form but each have their own, different mechanism. That pattern does transfer cross-domain, but as the general primes (non_monotonic_relationships, or phase_transitions), not as the uncanny valley, which is merely the human-perception instance of the family. The honest report, then, is that "uncanny valley" travels mechanistically only within the near-human-artefact trigger; where the cross-domain lesson is the shape, it belongs to the non-monotonic-response prime, and where the lesson is the cause, it belongs upstream to expectancy-violation and mismatch-detection — not to this concept as named (see Structural Core vs. Domain Accent).

Examples¶

Canonical¶

Masahiro Mori's 1970 essay Bukimi no Tani ("The Uncanny Valley") is the seminal formulation, built on a now-classic thought experiment. Mori observed that an industrial robot arm elicits no particular feeling, a toy or cartoonish humanoid robot elicits mild affection, and affinity climbs as robots become more human-like — until a prosthetic hand enters the picture. A well-made prosthetic hand can look convincingly human at a glance, so affinity is high; but when you shake it and feel it is cold, rigid, and rubbery, the mismatch between its human appearance and its non-human feel produces a jolt of eeriness that drops it below the toy robot in affinity. Mori sketched this as a curve of affinity against human-likeness that rises, plunges into a valley at the "almost human" band, and recovers only for a genuinely healthy human — and he noted that adding movement (a moving prosthetic, a corpse versus a zombie) deepens the dip.

Mapped back: The prosthetic hand is the human-like artefact, positioned high on the realism axis by appearance yet betrayed on contact. The cold, rubbery feel against a human-looking form is the mismatch detector firing once the hand has crossed the recruitment threshold into near-human territory. Affinity dropping below that of the cruder toy robot is the non-monotonic curve, and Mori's remark that motion worsens it names the amplifier terms.

Applied / In Practice¶

The film-animation industry treats the valley as a working design constraint. Robert Zemeckis's The Polar Express (2004), built on motion-capture of photorealistic human characters, was widely described by critics and audiences as eerie or "dead-eyed" — a textbook case of characters landing in the trough, where near-photoreal faces with subtly wrong eyes and micro-expressions recruited human scrutiny that flagged the flaws. In direct contrast, Pixar and Disney deliberately stylize human characters (exaggerated proportions, non-photoreal skin) to sit safely on the rising, appealing side of the curve rather than polishing toward a realism that would drop them into the dip. The same logic now guides caution around near-photoreal AI-generated faces and virtual agents.

Mapped back: Polar Express's mo-cap humans sat inside the near-human band where the mismatch detector dominates, worsened by the amplifier terms of full motion. Pixar's choice to stylize rather than pursue photorealism is the safe-direction heuristic enacted: inside the trough, retreat to stylization rather than polish in place — a design decision read directly off the artefact's position on the non-monotonic curve.

Structural Tensions¶

T1: Realism as remedy versus realism as poison (one lever, opposite signs). The uncanny valley's defining break with intuition is that adding human-likeness — the reflexive path to a more appealing character — has a sign that depends entirely on position: below the dip it raises affinity, inside the trough it drives the artefact deeper before any recovery. There is no single answer to "should we make it more realistic?"; the same manipulation helps or harms depending on which side of the recruitment threshold the artefact sits, and the threshold is exactly where the intuition is least reliable. The tension is that the field's tacit rule (more realism, more appeal) is not merely wrong but wrong in a locatable, costly band, so the design lever most reached for is the one whose effect can invert without warning. Diagnostic: Is the artefact below the dip (where added realism helps) or inside the near-human trough (where the same increment makes it worse)?

T2: Crisp curve shape versus contested empirics (a load-bearing abstraction on a shaky base). The rise-dip-recovery curve is treated as the load-bearing abstraction, sharp enough to drive concrete design decisions and predict that a passing still will fail once animated. Yet it remains a hypothesis with a contested empirical base — meta-analyses find the dip in some populations and stimulus sets and not others — so the very crispness that makes the curve actionable overstates a regularity that does not reliably replicate. The tension is that the concept's utility depends on a clean, universal shape while the evidence supports something more conditional and population-dependent, so a designer betting a production on "avoid the trough" is trusting a curve whose depth and even existence vary across audiences. Diagnostic: Is the dip being treated as a settled law governing this audience and stimulus, or as a hypothesized shape that may not manifest for them at all?

T3: Useful shape versus uncertain mechanism (the interventions need the engine the shape can do without). The concept can guide design as pure shape — locate the artefact, avoid the middle — without committing to why the dip exists. But its most valuable, non-obvious interventions depend on the mechanism being right: "motion deepens the dip," "cross-modal mismatch drops an artefact in even where each channel clears alone," and "remove the specific inconsistency to lift it out" all follow from mismatch detection in human-specific routines, not from the bare curve. The tension is that the shape is robust to mechanistic uncertainty while the actionable, sign-aware moves are not, so a practitioner who trusts the mechanism-derived diagnostics is leaning on the more speculative half of the concept precisely where the payoff is highest. Diagnostic: Is the design decision resting on the artefact's position on the curve (mechanism-agnostic), or on a mismatch-detection prediction that assumes the proposed engine is correct?

T4: Two safe exits versus their unequal feasibility (retreat is reliable, "push through" recedes). The safe-direction heuristic offers symmetric escapes from the trough: retreat to full stylisation or push forward to full photorealism. But the two are not equally attainable. Stylisation is a design choice fully under control, reliably clearing the valley; "genuinely human" realism is a receding target, because every added channel of scrutiny — motion, micro-expression timing, cross-modal consistency, and whatever the recognition routines flag next — raises the bar for what reads as truly human, so "push past the dip" can be practically unreachable within budget. The tension is that the concept presents two exits as equivalent options while one is dependable and the other may be an asymptote, quietly making stylisation the only sure escape for most productions. Diagnostic: Can this project actually reach the far rim of full photorealism across every channel, or is stylisation the only exit from the trough it can reliably afford?

T5: Autonomy versus reduction (a near-human perceptual effect or an instance of non-monotonic response). The uncanny valley transfers as mechanism across robotics, animation, deepfakes, prosthetics, and the doll/wax-figure cluster — one trigger condition (a human-like artefact varied along a realism axis, recruiting human-specific routines past a threshold), many industrial settings. But two things travel further only as parents: the rise-dip-recovery shape belongs to non_monotonic_relationships / phase_transitions, whose other instances (hormesis, Yerkes-Dodson) share the form with different engines, and the cause belongs upstream to expectancy-violation and mismatch-detection. "Uncanny valley of mind" or "of voice" coinages lift the shape while dropping the perceptual trigger that makes the visual valley sharp, so they are analogy, not transfer. The tension is that the concept's portable content splits between a shape-parent and a cause-parent, leaving the named valley as the human-perception instance. Diagnostic: Resolve toward non_monotonic_relationships/phase_transitions when the lesson is the shape and toward mismatch-detection when it is the cause; toward the named uncanny valley only for near-human artefacts that recruit human-specific recognition routines.

Structural–Framed Character¶

The uncanny valley sits toward the structural end of the spectrum — best read as mixed-structural: a genuine perceptual mechanism (a natural regularity of human cognition) wearing human-perception-specific vocabulary. Four of the five criteria carry structural. Its evaluative_weight is nil: the curve describes an affective response, neither endorsing nor condemning — "creepy" is a locatable position on a known curve, a diagnostic reading, not a verdict. It is not human_practice_bound in the constitutive sense the framing targets: the effect is not an artifact of a tradition, agency, or convention but a fact of how near-human stimuli recruit human-specific (or conspecific) recognition routines — it runs in the perceiver whenever such an artefact is met, needing no scientist to constitute it (the site of the effect is a human mind, but that is the natural substrate, as a lithosphere is for isostasy, not a human institution). Its institutional_origin is likewise none: Mori named a perceptual regularity in 1970, he did not invent it by fiat. And within its trigger condition cross-domain reuse is recognition, not import: robotics, animation, deepfakes, prosthetics, and the doll/wax-figure cluster are "different industrial settings, not different substrates — one trigger, many applications," so the same mechanism is recognized intact across them.

What holds it off the structural pole is vocab_travels, which it fails, together with the concept's own contested empirical base (it is a hypothesis, the load-bearing content being the curve shape, not a confirmed law). The operative vocabulary — human-like artefact, realism axis, human-specific recognition routines, mismatch detector, the near-human trough — is pinned to the human-perception substrate; lifted to "the uncanny valley of mind" or "of voice," it keeps only the rise-dip-recovery picture and drops the perceptual trigger that makes the visual valley sharp, so that transfer is analogy. Here the portable content genuinely splits in two, as the entry demonstrates: the shape — non-monotonic response to a continuous design parameter, rising then dipping then recovering — is instantiated from non_monotonic_relationships (with phase_transitions alongside), whose other members (hormesis, Yerkes–Dodson) share the form on different engines; and the cause — a raised high-level expectation, violated by a detected mismatch, producing a heightened negative signal — is instantiated from the upstream expectancy-violation / mismatch-detection pattern. Both are needed because the shape and the mechanism travel to different parents. Those are the umbrellas the uncanny valley instantiates; its distinctive content — the near-human trigger regime, the recruitment threshold, the motion and cross-modal amplifiers, the retreat-or-photorealism heuristic — is precisely the human-perception cargo that does not lift. Its character: a real, evaluatively neutral, recognized-in-its-regime perceptual mechanism, structural in the non-monotonic-response shape it borrows from non_monotonic_relationships and the mismatch-detection cause it borrows from expectancy-violation, but stated in human-perception vocabulary that pins it to near-human artefacts, leaving it mixed-structural rather than a free-floating prime.

Structural Core vs. Domain Accent¶

This section decides why the uncanny valley is a domain-specific abstraction and not a prime — and it is one of the entries where the skeleton is genuinely doubled, because the concept's portable content splits between a shape-parent and a cause-parent that travel to different places.

What is skeletal (could lift toward a cross-domain prime). Strip the near-human artefact and two thin abstract structures survive, and both are load-bearing. The first is the shape: affinity is a non-monotonic function of a continuous design parameter — rising, then dipping sharply across a narrow band, then recovering. That is non_monotonic_relationships (with phase_transitions alongside), and it recurs with the same rise-dip-or-peak form in hormetic dose-response and the Yerkes-Dodson arousal-performance curve — though each of those runs on a different engine, which is exactly why the shape must be assigned to the general prime, not to this concept. The second is the cause: a raised high-level expectation, violated by a detected mismatch, produces a heightened negative signal precisely because the expectation was engaged. That is the upstream expectancy-violation / mismatch-detection pattern. Both are genuinely substrate-portable, and both are cores the uncanny valley shares — the shape with the whole non-monotonic family, the cause with any expectancy-violation instance — not what makes it distinctive. Naming them both, reasoned rather than padded: they are needed because the shape and the mechanism travel to different parents.

What is domain-bound. Everything that makes the object the uncanny valley in particular is human-perception machinery keyed to near-human artefacts. The trigger regime itself — the artefact must recruit human-specific (or conspecific) recognition routines; the recruitment threshold past which coarse object perception gives way to human-specific scrutiny; the specific anomalies the mismatch detector flags (mis-timed micro-expressions, dead eyes, wrong skin, motion that fails to track the face); the amplifier terms (motion deepens the dip, cross-modal inconsistency drops an artefact in even where each channel alone would clear it); and the retreat-to-stylisation-or-push-to-photorealism design heuristic. The decisive test the entry supplies: no one reports an uncanny-valley dip for a car approaching photorealism, because the human-specific routines are never engaged — the trigger regime is the boundary the concept draws around itself. Remove the near-human perceptual trigger and what remains is the bare non-monotonic curve plus a generic expectancy violation, a looser thing that is no longer this concept.

Why this does not clear the prime bar. A prime's vocabulary travels and its cross-domain transfer is recognition of the same mechanism, not analogy. The uncanny valley's transfer is bimodal. Within the near-human-artefact trigger it travels fully as mechanism, one trigger across many industrial settings: robotics, animation (the diagnosed failure of The Polar Express and Final Fantasy), deepfakes and generative video, prosthetics and medical mannequins, and the taxidermy/wax-figure/doll cluster — the curve, the recruitment threshold, the amplifiers, and the sign-aware heuristics carry intact because each is the same perceptual trigger, not a different substrate. Beyond that trigger the transfer splits and neither branch makes it a prime: "uncanny valley of mind" or "of voice" coinages lift the rise-dip-recovery picture while dropping the perceptual trigger that makes the visual valley sharp, which is analogy; and where a genuine cross-domain lesson is wanted, it belongs upstream — the shape to non_monotonic_relationships / phase_transitions, the cause to expectancy-violation / mismatch-detection. So the cross-domain reach belongs to those two parents; the uncanny valley's distinctive content — the near-human trigger regime, the recruitment threshold, the motion and cross-modal amplifiers, the retreat-or-photorealism heuristic — is exactly the human-perception cargo that should stay home. The uncanny valley clears the domain-specific bar comfortably for perception-and-design, but its only substrate-spanning content is the non-monotonic-shape and mismatch-detection patterns its parent primes already carry.

Relationships to Other Abstractions¶

Current abstraction Uncanny Valley Domain-specific

Parents (1) — more general patterns this builds on

Uncanny Valley is a decomposition of Inverted-U Response Prime

The Uncanny Valley decomposes to an Inverted-U Response when the signed outcome is expressed as eeriness rather than affinity.

Hierarchy path (1) — routes to 1 parentless root

Uncanny Valley → Inverted-U Response → Nonlinearity

Not to Be Confused With¶

Other non-monotonic curves (Yerkes–Dodson, hormesis). Shape-siblings that share the rise-then-reverse form on entirely different engines: the Yerkes–Dodson law relates arousal to performance (an inverted-U, peaking then falling), and hormesis relates dose to effect (low doses beneficial, high doses harmful). Each is a co-instance of the non-monotonic-response family, but none involves human-specific recognition routines detecting near-human anomalies. They share the form, not the mechanism. Tell: is the reversal driven by human-specific perceptual scrutiny of a near-human artefact (uncanny valley), or by arousal, dose, or some other parameter with its own engine (Yerkes–Dodson, hormesis)?
Expectancy violation / prediction-error mismatch. The upstream cause the uncanny valley instantiates — a raised high-level expectation, violated by a detected mismatch, produces a heightened negative signal. This is the general mechanism; the uncanny valley is its specific case keyed to near-human perceptual expectations. When the lesson is the cause (unease from violated expectation), it belongs to this broader pattern, not to the named valley. Tell: is the point the generic expectation-then-violation dynamic on any expectation (mismatch detection), or specifically the perceptual eeriness of a near-human artefact crossing the recruitment threshold (uncanny valley)?
Metaphorical "uncanny valley of X" coinages. Extensions like "the uncanny valley of mind" or "of voice" that lift the rise-dip-recovery picture into territory where the perceptual trigger — human-specific recognition routines flagging near-human anomalies — is absent or far weaker. They borrow the shape and drop the engine that makes the visual valley sharp and predictable, so they are analogy, not the mechanism traveling. Tell: does the setting actually recruit human-specific perceptual scrutiny of a near-human artefact (genuine uncanny valley), or is only the curve-shape being borrowed for a non-perceptual domain (metaphor)?
non_monotonic_relationships / phase_transitions (parent primes). The substrate-neutral shape the uncanny valley instantiates — a non-monotonic response to a continuous parameter, rising then dipping then recovering. This is what travels cross-domain when the lesson is the curve's form (with the cause traveling separately to mismatch-detection). It is the umbrella, not a peer confusable. Tell: is the lesson the generic non-monotonic-response shape on any parameter (the parent), or the specific near-human affinity curve with its perceptual trigger (the named concept)? (Treated fully in a later section.)

Neighborhood in Abstraction Space¶

Uncanny Valley sits in a moderately populated region (58^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Caregiving Cues & Attachment (6 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12