AI-generated food images cause anxiety — study

As artificial intelligence advances, it has learned to generate images that are sometimes hard to distinguish from reality. However, when it comes to food, these visual creations often evoke not appetite but unease.

They fall into what’s known as the “uncanny valley”—a realm where unnatural realism triggers discomfort. But why does this happen? Researchers suggest the reason may lie in food neophobia, an innate wariness of new or unfamiliar foods.

What the Study Revealed

According to a new study soon to be published in the journal Appetite, people perceive AI-generated food images as less appealing and more unsettling compared to real photographs or cartoon illustrations. Furthermore, attitudes toward these images follow a complex curve: initial rejection is followed by an increase in acceptance.

Scientists hypothesize that this reaction stems not from disgust, which helps avoid potentially dangerous food (like spoiled items), but from food neophobia—a natural mechanism that makes people cautious about unfamiliar products. These findings are significant for marketing, advertising, and even clinical research involving AI-generated food imagery.

Why AI Food Images Cause Discomfort

Generative models that turn text into images are widely used today across various fields, including food advertising and marketing. Yet, consumers have mixed feelings about these images. Many find them strange, unnatural, or even unpleasant, preferring traditional photography.

This effect is explained by the “uncanny valley” theory, proposed by engineer Masahiro Mori in 1970. He observed that the more realistic an object appears, the more likable it becomes—but only up to a point. When realism becomes too high yet remains imperfect, the object starts to feel eerie. This phenomenon has been observed not only with humanoid robots but also in architecture, animal depictions, and even inanimate objects.

Researchers set out to determine if this principle applies to food images. They suspected people might reject them similarly to spoiled food. However, the results showed that perception of AI-generated food is more influenced by neophobia than typical food aversion.

How the Experiment Was Conducted

The researchers aimed to answer two questions: whether AI-generated food images truly trigger an “uncanny valley” effect and which individual factors influence this perception.

They began with a pilot study involving 12 students from a German university. Participants were shown 99 AI-generated images varying in realism and rated them based on criteria such as pleasantness, eeriness, photorealism, and artistry.

This was followed by a main experiment with 95 participants, who viewed 38 images—ranging from realistic to cartoonish, alongside photos of spoiled food. They evaluated these for eeriness, warmth, and realism, and completed a test to measure their level of food neophobia.

Key Findings

The analysis revealed that people find imperfect AI-generated food images less pleasant and more anxiety-inducing. Interestingly, stylized or cartoonish images didn’t provoke this reaction—the “uncanny valley” effect was exclusive to those aiming for photorealism but featuring distortions.

The researchers also found that food neophobia plays a critical role in how these images are perceived, while standard food disgust has little impact. Notably, individuals with a higher body mass index (BMI) exhibited less anxiety when viewing AI-generated food.

What This Means for the Future

Although the study didn’t account for factors like taste preferences or exposure to AI imagery, its results confirm that imperfect yet realistic food images do trigger an “uncanny valley” effect.

These findings have implications for advertising and marketing AI-generated products. It’s particularly important to recognize that people with varying levels of food neophobia may respond differently to such images. The influence of BMI on perception is also intriguing: those with higher BMIs may be more tolerant of these visuals.

Looking ahead, scientists plan to explore how cultural differences and familiarity with AI technologies might alter this effect. As technology advances and people interact more frequently with digital food imagery, the “uncanny valley” effect may diminish, and such images could become more widely accepted.