A New Perspective on the “Uncanny Valley Phenomenon” at the Intersection of AI Technology and Human Art
You may have felt uneasy when looking at an android or avatar that resembles humans but is strangely different in video games, animated movies, or robotics. This is known as the uncanny valley phenomenon. As the accessibility of generative AI increases, interesting directions for research on this phenomenon are being proposed. Until now, most scientists have focused on studying the uncanny valley phenomenon itself that occurs in video games, animated movies, and robotics. However, recently, there has been a focus on this phenomenon at two interesting intersections with generative AI, which can bring about new twists.
The first intersection is that the human figures drawn by AI also trigger the aforementioned anxiety. Most AI tools find patterns in existing photos posted online, so they generate human faces very well. Since there are countless human facial data available online, they can reproduce human faces more intricately than in reality. However, this ability causes the uncanny valley phenomenon.
“For demonstration purposes, we used the keywords ‘person eating spaghetti’ and ‘studio-style photo’ to generate four images using DALL-E 2. At first glance, they look like convincing human photos, but upon closer inspection, they still trigger a sense of unease. In the first photo, the man’s empty and melancholic gaze, the rounded shape of the ears in the second photo, and the partially generated teeth in the fourth photo all hint that these may not be real humans. They are almost human, but they violate our expectations, triggering the uncanny valley phenomenon.
The second characteristic and more innovative intersection of generative AI and the uncanny valley phenomenon is the fact that the generated images, although appearing to be created by humans, are not actually so. Artworks produced by generative AI such as Jason Allen’s Théâtre D’opéra Spatial (2021) and Boris Eldagsen’s The Electrician (2022) have recently won first place in art competitions due to their compelling nature. This indicates the need for a more nuanced understanding of our concept of the uncanny valley phenomenon.
The previous theories of the uncanny valley have discussed the strange feeling caused by creating “non-human” things as “human.” However, generative AI presents a completely different type of uncanny valley. GAI attempts to depict humans by finding patterns in existing data created by humans, which means dealing with “non-human things” that are “human-like expressions” based on human input.
Understanding how the uncanny valley phenomenon is experienced in relation to GAI has significant implications as a new form of communication. If AI tools continue to advance in the future, the aforementioned strange discomfort may disappear. However, this highlights the urgency of researching the uncanny valley phenomenon. The previous uncanny valley in video games or animations was simple, but increasingly convincing GAI technology is opening the door to content that can shake the foundation of democratic societies, such as deepfakes or bot spam. Therefore, understanding how we perceive generative AI content must be done carefully and with a nuanced approach, and understanding the uncanny valley phenomenon will be an important puzzle piece.
Developing this understanding is not the sole responsibility of cognitive scientists. Empirical understanding of the cultural and societal importance and impact of generative AI must be studied by artists, who are both consumers and leading researchers in digital culture. With advancements in data science and cognition, and with increased public awareness, we can fully utilize the benefits of generative AI and avoid the threats of a destructive future.
