A 'Turing Test' for Self-Awareness

In this paper, I propose a test for machine self-awareness inspired by the Turing test. While my test is plain and simple, it offers a much-needed alternative to ungrounded speculation while still capturing the essence of self-awareness. I discuss how to apply the test to Large Language Models and conduct experiments on GPT-3.5 that suggest it is not self-aware. I discuss the implications of machine self-awareness and relate it to questions about meaning and understanding. Finally, I present some next steps for studying self-awareness in machines.