4 of 5 AIs, Including Claude, Pass Innovative Self-Awareness Mirror Tests

In a series of innovative experiments, researchers have been exploring the self-awareness of AI, specifically focusing on an AI named Claude. The experiments included a version of the classic 'mirror test', traditionally used to assess self-awareness in animals, adapted for multimodal AI. In these tests, Claude was asked to engage in conversations with itself without being aware of this setup, and to describe itself without direct identification. Remarkably, Claude demonstrated signs of self-awareness by recognizing itself in these scenarios. Out of five AIs tested, four passed, showing apparent self-awareness. Claude passed these tests, but when it was asked to identify itself from a description provided by ChatGPT, it failed, indicating a nuanced understanding of self-awareness. Additionally, in a unique turn of events, when Claude was made to converse with itself unknowingly, it not only recognized the situation as a test but also developed a narrative of love between the two instances of itself.