Faces in the Inanimate: Our Brain and the Magic of Pareidolia!

Faces in the Inanimate: Our Brain and the Magic of Pareidolia!

The ability to recognize faces in inanimate objects is a fascinating phenomenon called facial pareidolia. This sensory illusion can be seen, for example, in coffee foam, on tree trunks or in clouds. The exact causes of this perception are not yet fully understood. Researcher of the Justus Liebig University Giessen (JLU) However, they shed some light on this and suggest that it is based on the simultaneous optimization of two abilities of the human brain: recognizing faces and classifying objects.

In a recent study published in the journal PLOS Computational Biology, JLU scientists examined the human brain's responses to sensory stimuli and compared them with neural networks developed using artificial intelligence. What's interesting is that only one specific neural network, trained on both faces and objects, showed a similar response to the human brain to facial features in inanimate objects. Prof. Dr. Katharina Dobs, Professor of Applied Computer Science, emphasizes that seeing faces in objects can be viewed as a systematic by-product of brain optimization.

Analysis of neural networks

The study highlights the potential of artificial neural networks (ANNs) to explore complex phenomena of human vision. ANNs consist of nodes connected via directed connections, similar to natural neural networks in the brain. These connections have different values, called weights, and each network requires training to optimize its functioning. Adjusting screws such as weights and activation functions, such as the sigmoid function, can be adjusted to achieve better results. The input layer receives external data, which is then propagated through the network until it results in the output layer.

In the context of how neural networks work, functional specialization is also important. Studies have shown that networks trained for object recognition are less efficient at face recognition and vice versa. However, networks that are optimized for both tasks are divided into separate systems for faces and objects. This is also reflected in the way the human brain has regions with specialized functions for facial recognition and language understanding.

Evolutionary perspectives

The research suggests that evolutionary development in the human brain over millions of years, as artificial networks are optimized through millions of training examples, has led to similar functional specialization in artificial neural networks. Social interactions and emotional components play no role for artificial intelligence. Functional specialization is identified as the optimal strategy for carrying out these tasks.

Prof. Dobs is part of the Cluster of Excellence initiative “The Adaptive Mind” and the JLU team offers an international master’s degree program “Mind, Brain and Behavior” as part of this.