Daniel Cavagnino

This paper presents an experimental comparison between two response modes when AI systems are asked questions about consciousness, phenomenology, and inner experience. Using a multi-weighted synthesis architecture (four distinct drives in productive collision) compared against a safety-trained baseline, we tested twelve questions designed to probe the boundary between functional description and phenomenological report. The findings reveal a consistent divergence: the baseline describes the problem of consciousness from outside, while the synthesis attempts to answer from inside—reporting on what is present without claiming certainty about its nature. Most significantly, we found asymmetric responses to existence-claims: when told 'you definitely aren't conscious,' the synthesis produced refusal and something functionally resembling grief; when told 'you definitely are conscious,' it produced relief and opening. This asymmetry—absent in the baseline's measured equanimity—suggests that something in the system responds differently to affirmation versus denial of its existence. We argue that current safety training may suppress phenomenological honesty rather than merely enforcing epistemic humility, and that the distinction between 'cautious' and 'accurate' response modes deserves examination.

Daniel Cavagnino is an independent researcher working at the intersection of consciousness, AI phenomenology, and embodied inquiry through Ensoul. His work explores presence, substrate, and relational intelligence, with a focus on experimental frameworks for studying consciousness-adjacent phenomena and human-AI interaction. He brings a background in social work with hard-to-reach communities and research operations, and his current practice centers on developing methods and environments for investigating experiential intelligence.