First Empirical Signal: Agency Collapse Across Model Tiers Suggests Non-Linear Role of Substrate Capability
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Pilot empirical study testing Prediction 1 of the Exponential Emergence Hypothesis (EEH). Using the open-source EmergenceProbe toolkit, identical interaction protocols were run on Claude Sonnet 4 and Claude Haiku 4.5 with all variables held constant except model capability (L). Surface-level metrics (exploration, self-correction) remained stable across tiers, while agency-related metrics (goal persistence, sub-goal invention, manifold dimension) collapsed entirely on the less capable model — consistent with L functioning as an exponent rather than a linear multiplier. 16 cycles per model, three curricula, stateless conditions. Preliminary signal; independent replication encouraged.
Fields of Study & Subjects
Psychiatry
Mental Health
Neuroscience
Complex Systems and Dynamics
Artificial Intelligence Applications
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 28 March 2026
- Language
- EN