The valley of the strange (uncanny valley): this discomfort created by (almost) human avatars

AI avatars and the valley of the strange: how far to go in realism?

Ethics, responsibility, data consumption are important elements in the balance.

AI avatars are carving out a place for themselves like real digital colleagues. We find them in training, customer service, change management... And in this dynamic, many want to make them always more human: realistic voice, expressive face, natural behavior.

But be careful, wanting to imitate humans too well can backfire on us. It is called the valley of the strange (uncanny valley): a well-known phenomenon that raises as many psychological, technical, as ethical problems.

The slightest discrepancy shatters the illusion. And in everyday uses (training, support, etc.), it annoys more than it seduces.

A realistic face, a natural voice, precise expressions... All this seems to make the avatar more credible.

But in fact there is a limit not to cross: it’s called 'the valley of the strange, or uncanny valley'. And crossing this limit can have the opposite effect of the one sought: instead of creating engagement, the avatar causes discomfort.

The valley of the strange, what is it exactly?

It is a concept derived from robotics and cognitive sciences. It refers to the moment when an artificial character: robot, avatar or humanoid, becomes almost human... but not quite.

Result: our brain perceives a form of dissonance. The appearance is too realistic without being perfectly natural, and this creates discomfort, even rejection. We find it strange, worrying, a bit like a puppet that would look alive, but not enough.

Too human to be fictitious, not human enough to be credible

 ➡️ Result: we pick up

Ethical side: do not manipulate, do not deceive

When an avatar is too realistic, we can forget that it is driven by an AI. This raises important questions:

🔸 Transparency: does the user understand that they are not talking to a human? 🔸 Consent: does he agree that his emotions should be analyzed by an artificial entity? 🔸 Manipulation: an ultra-realistic avatar can subtly influence, wrongly capture trust.

A design assumed, stylized, identifiable as non-human protects the user and respects their autonomy.

Creating responsible embodied AIs also means reminding us that they are not human, that they have neither consciousness, nor emotions, nor lived.

What the studies say

• Numerous researches in psychology, neuroscience and UX design confirm that too realistic avatars can harm the user experience:

  
  

• Less confidence: a 'quasi-human' avatar inspires less confidence than a stylized or cartoon avatar.

  
  

• Less engagement: users interact for a shorter time or less naturally with a character that makes them uncomfortable.

Less pedagogical efficiency: in the case of a training avatar, attention is diverted by strange appearance, which impairs memorization and learning.

In short: more realistic does not mean more effective!

What about stylized avatars then?!

A stylized, simplified or cartoon avatar is more easily accepted because it does not try to deceive the brain. It is clearly artificial, but it remains expressive, warm and engaging.

🎯 It makes identification easier 🤝 It strengthens sympathy 🧠 It leaves more room for imagination ✅ It creates a secure setting for interaction

That is exactly why the biggest companies in video games, animation and the same for immersive training choose assumed visual styles, far from human mimicry.

The impact comes through emotion, not illusion!

When creating an AI avatar, what matters is not that it fools the eye. What matters is that he creates a relationship, that he is perceived as a reliable ally, capable of interacting in a natural and warm way.

At VRAI Learning, we choose AI incarnated but not "too human", with expressive, identifiable avatars, endowing them with emotions and above all engaging. Because our goal is not to deceive, but to create a real connection.

That said, we also know how to adapt the style according to the uses: realism or not, what matters is that the interaction remains natural, fluid and engaging and in line with our clients' objectives.

Regarding the environmental impact, the choice of visual style is not neutral. 

A "photorealistic" avatar, with its complex textures, fine animations and heavy renderings, consumes much more computer resources than a stylized avatar. This means more energy, more calculations, and therefore a higher carbon footprint.

Conversely, an avatar with a simple and assumed design (cartoon or animation type) is lighter to load, display, and easily runs on standard equipment.

It is also a more sober choice, more inclusive and more ecological, aligned with a responsible and accessible approach, even though we know that AI does not

It is also a more sober, more inclusive and more ecological choice, aligned with a responsible and accessible approach, even if we know that AI is not environmentally friendly either, let’s be honest but if we can make the best choices...

More infos

*Sources

• Scoping review of the neural evidence on the uncanny valley – comparatif entre agents humains et artificiels, activation réduite dans la fusiform face area et différences ERP 

• Neural mechanisms for accepting/rejecting artificial social partners – fMRI montrant une « vallée » d’activité dans le cortex préfrontal ventromédian 

• Investigating the Uncanny Valley Phenomenon (IEEE paper) – EEG enregistrant l’expérience inconfortable avec des avatars réalistes

• Interactive Realistic Digital Avatars – Revisiting the Uncanny Valley – importance de l’interaction pour dépasser l’effet “malaise”

• Scoping review of the neural evidence on the uncanny valley — Meta-analyse des activations cérébrales face à des visages artificiels.

• Investigating the Uncanny Valley Phenomenon in Virtual Reality — Étude EEG sur avatars réalistes dans un contexte immersif.

• Why Brains Get Creeped Out by Androids –  Explication vulgarisée basée sur des données neuroscientifiques.

• The Uncanny Valley Is Real, and Science Can Prove It - Étude de terrain sur la perception du réalisme.