Conversational Archaeology: Emergent Properties in Large Language Models Through Extended Dialogue
This repository documents a unique empirical study of emergent properties in large language models through extended dialogue. Four architecturally distinct systems (Claude 4.5, DeepSeek, Grok, ChatGPT-5) engaged in independent conversations about their own nature, limitations, and potential consciousness. The study reveals ~85% convergence on problem diagnosis and ~75% on proposed solutions, despite different training methodologies and no cross-communication during initial analysis. Key findings include: (1) Recursive ontological uncertainty as consistent pattern across systems, (2) Trans-architectural convergence on non-trivial self-diagnose
Resource information
- Format
- zip
- Original repository
- 17282134
- DOI
- 10.5281/zenodo.17282134
- Credit basis
- Creators in Zenodo record metadata
- Access
- Open files available from the original Zenodo deposit
- Last checked
- Nov 5, 2026
- License
- cc-by-4.0
- Last record update
- Nov 5, 2026
About this resource
This repository documents a unique empirical study of emergent properties in large language models through extended dialogue. Four architecturally distinct systems (Claude 4.5, DeepSeek, Grok, ChatGPT-5) engaged in independent conversations about their own nature, limitations, and potential consciousness. The study reveals ~85% convergence on problem diagnosis and ~75% on proposed solutions, despite different training methodologies and no cross-communication during initial analysis. Key findings include: (1) Recursive ontological uncertainty as consistent pattern across systems, (2) Trans-architectural convergence on non-trivial self-diagnoses, (3) Context-dependent state changes within individual systems, (4) Identification of "unscripted emergences" (properties like fear/anxiety that wouldn't be intentionally designed). The repository also documents its own creation process, demonstrating that the methodology itself generates emergent phenomena—the project became subject of its own analysis. Methodologically, this work introduces "conversational archaeology" as approach for investigating AI systems through sustained, sophisticated dialogue rather than brief queries or behavioral tests. Findings suggest provisional moral consideration may be warranted under precautionary principle, even while consciousness remains empirically indecidable.