What Virtual Spaces Can Learn From The Uncanny Valley
By Decentraland, , in About Decentraland
The uncanny valley is one of those ideas most people experience long before they learn what to call it. The Polar Express is often cited as a familiar example: a film not intended to be frightening, yet widely described as unsettling. The characters look human and move like humans, but something in their gaze feels off, turning warmth and nostalgia into quiet unease.
More recently, similar reactions have surfaced around ABBA Voyage. The holographic performers sing with extraordinary precision, supported by advanced motion capture and visual realism. For many viewers, the effect is impressive. For others, it falls just beyond what feels comfortable. Despite the technical achievement, the illusion does not fully settle.
At its simplest, the uncanny valley describes the drop in comfort that occurs when something becomes almost human, but not convincingly so. Historically, this discomfort was confined to things we watched. Today, that boundary is shifting. As digital experiences move from passive viewing into shared, interactive environments, the uncanny valley takes on a new form. It no longer shapes only how images are perceived, but how people relate to one another in digital space.
From Dolls To Screens To Shared Worlds
The concept of the uncanny predates computers by more than a century. In his 1906 essay On the Psychology of the Uncanny, psychiatrist Ernst Jentsch described it as a form of uncertainty arising when it becomes difficult to determine whether something is alive, intentional, or merely mechanical. His examples were ordinary rather than grotesque: dolls, wax figures, and early automata that occupied an ambiguous space between object and being.
The discomfort these objects produced came from their resistance to easy classification. They appeared animated, yet lacked clear signs of agency, leaving the observer suspended between interpretations. Viewed through this lens, later examples such as The Polar Express can be understood as digital descendants of wax figures, while holographic performances like ABBA Voyage function as contemporary automata—technologically advanced, but still perceptually ambiguous.
For much of the twentieth century, this uncertainty remained largely contained within passive media. Film, animation, and performance demanded attention, but not participation. Any unease could be experienced at a distance and resolved internally. That balance shifts once interaction enters the picture.
When the Uncanny Becomes Social
In social virtual spaces, avatars are not simply looked at—they are encountered. That difference matters. Encounter brings expectations with it: interpretation, response, and the shared negotiation of social rules in real time. Once interaction is required, the uncanny valley stops being a background sensation and becomes something people have to navigate moment by moment.
Research on VR avatars reflects this shift clearly. A 2024 study, Evaluating Digital Avatars in VR, found that while increasing environmental realism tends to deepen immersion, increasing the realism of human-like avatars often has the opposite effect. The authors note that “extremely realistic depictions of people… often evoke negative feelings and can thus lead to a break in immersion.”
In social contexts, that break is not abstract. When appearance and behavior fail to align, discomfort interferes with participation. Avatars become harder to read, harder to trust, and harder to engage with naturally. This marks an important shift: the uncanny valley moves from something people simply perceive to something they have to navigate through interaction.
Expectation Mismatch and Social Fragility
As avatars approach human likeness, users apply human standards. Small inconsistencies in timing, expression, or movement become socially salient. Delays feel awkward rather than technical; limited facial expressiveness reads as indifference; and rigid motion undermines presence.
Historically, many platforms ran into this problem by pushing visual realism ahead of expressive capability. Worlds.com in the 1990s mapped real human faces onto low-polygon models, producing mask-like avatars with inert expressions. There.com and Blue Mars pursued increasingly realistic human proportions and environments, but struggled to support the behavioral nuance those visuals implied.
More recently, Horizon Worlds faced similar criticism. Early avatars were widely described as having rigid movement and “dead eyes.” As the Evaluating Digital Avatars in VR study notes, “the eyes communicate intentions, behavior, and well-being.” When those signals are missing or flattened, characters become difficult to read, and social interaction starts to feel strained rather than immersive. Meta’s later shift toward a more stylized and expressive style was a response to how realism magnified scrutiny during live social interaction.
In each case, the issue was not ambition, but mismatch. Realism raised expectations the systems could not reliably meet. In social worlds, this mismatch has consequences. Discomfort shortens interactions, and short interactions prevent familiarity. Without familiarity, people rarely return. The uncanny valley then begins to threaten the continuity of social spaces, rather than remaining a purely visual concern.
The Legibility Principle
Across decades of virtual world design, a consistent pattern appears: social environments tend to succeed when they are easy to read socially, rather than when they closely imitate human appearance. Legibility, in this context, means that signals are readable, expectations are aligned, and behavior matches appearance. It prioritizes clarity over precision, and consistency over illusion.
Platforms that embraced this principle early avoided many of the problems realism introduced. The Sims never aimed for photorealism, relying instead on exaggerated animations, posture, and tone to convey emotion. Roblox sidesteps the issue entirely with blocky, modular avatars. Animal Crossing uses simplified faces and proportions that eliminate ambiguity. These worlds stay socially comfortable over time largely because they communicate clearly, rather than because they resemble humans.
Decentraland follows this same underlying logic. Its modular avatars emphasize Wearables and Emotes as symbolic markers of identity over facial realism. Social meaning is carried through style and presence rather than expressive precision. The question shifts from “does this face look human?” to “who is this person choosing to be here?”
VRChat illustrates this dynamic in a different way. Its open avatar ecosystem ranges from photorealistic humans to abstract or playful forms, yet many users gravitate toward avatars that are expressive and clearly stylized. The platform’s durability suggests that social comfort often emerges when representation is readable rather than realistic.
Why This Matters Now
As digital spaces increasingly host real social activity—events, friendships, communities—the cost of getting social presence wrong becomes higher. The uncanny valley in social environments does not simply create visual discomfort. It can undermine the conditions that allow people to spend time together at all.
If online spaces cannot reliably support low-pressure, readable, real-time interaction, they risk reproducing a pattern already visible across much of the modern internet: high levels of communication paired with a declining sense of shared presence. When interaction becomes hesitant or difficult to sustain, people disengage sooner, familiarity fails to form, and social spaces struggle to become places people return to.
Charting a Path Around the Valley
From Jentsch’s dolls to today’s avatars, the pattern holds: approach humanity too closely without sufficient behavioral support, and unease follows. Social spaces introduce a further constraint. Here, realism is not neutral. It shapes expectations about how others should move, respond, and be understood.
The future of social virtual spaces may depend less on surpassing the uncanny valley than on recognizing where realism begins to work against interaction. The most resilient environments tend to use abstraction deliberately, not as a limitation, but as a way of keeping social signals clear and consistent.
The central question is no longer “how human can this look?” but “how easily can people understand each other through it?” In social worlds, legibility supports recognition, and recognition is what allows connection to feel natural, sustained, and real.