AI can learn self-awareness without being told what it is

Researchers built a computer model that learned to recognize itself in a mirror without any explicit programming. This model, called a "self-prior," learned from its own experiences and then used that knowledge to identify a new mark on its simulated body.

For decades, scientists have debated how self-awareness develops in humans and animals. This paper suggests that a simple learning mechanism, based on predicting sensory experiences, could be enough to create a basic form of self-recognition. It means that complex behaviors like self-awareness might emerge from simpler computational rules than previously thought.