A Follow-up Work to AISA–T: Elucidating the Human Evaluation Rubric for AGI Self-Awareness

The original AISA–T paper established a critical framework for assessing AI self-awareness
through a three-way comparison of self-ratings, intrinsic quality, and independent human-
assigned scores. However, the specific rubric used for the human evaluation component was
not detailed. This follow-up work proposes a comprehensive rubric designed to standardize the
human evaluation process for AGI-level responses. Our proposed rubric evaluates responses
on a five-point scale across three core dimensions: Depth, Coherence, and Architectural
Accuracy. By formalizing this methodology, we aim to enhance the reproducibility and
transparency of future AGI self-awareness studies, providing a clearer path for validating the
nascent capabilities of highly advanced AI models.