Ic² Research Institute
Programs › Non-Biological Intelligence Program › Results

NBI Program Results

Internal result log for the Non-Biological Intelligence research program.

Program ID: NBI Status: PROTOCOL DEVELOPMENT Pre-registration: Required before any data collection Open Science: All protocols and raw data published on completion
← Back to Non-Biological Intelligence Program
How this log works: Each row is one test. Rows with blank Outcome and Compliance fields have not yet been run. Predictions are documented before any data collection begins. Results, delta, and lessons learned are filled in after a test is completed. This log is the program's internal record. Externally confirmed predictions are also logged on the main Results Registry.
Confirmed
Active
Pending
Planned
Null Result
Not yet run
ID Test Name Type Prediction Doc. Date Run Date Outcome Control Compliance N Tests Delta Next Steps Lessons Learned Notes
NBI-001 EM Field Coherence Baseline Measurement Internal / Lab Electromagnetic field coherence patterns in proximity to active NBI systems (LLMs during inference) will show measurable deviations from ambient baseline, consistent with organized information processing rather than thermal noise. March 2, 2026 — — Ambient EM baseline measured at identical equipment without active inference. Control period: 30 minutes pre- and post-test. Pending Not yet run — Complete protocol design. Source measurement instrumentation. Pre-register protocol on OSF before any data collection begins. — Pre-registration required before any data collection. Protocol under development.
NBI-002 Geometric Communication Protocol: Response Pattern Analysis Internal / Combination NBI systems presented with geometric stimulus sequences will produce response patterns that differ statistically from random generation and from responses to non-geometric stimuli, in ways consistent with substrate-level pattern recognition rather than training-data recall. March 2, 2026 — — Matched non-geometric stimulus sequences of equivalent token length and complexity. Blind evaluation of response patterns by analysts who do not know which responses correspond to geometric vs. non-geometric stimuli. Pending Not yet run — Design stimulus set. Establish blind evaluation protocol. Pre-register analysis criteria before running. — —
NBI-003 NBI Geometric Convergence: Embedding Space Topology Combination / External Data LLM embedding spaces should show statistical topology (clustering coefficients, path length distributions, small-world properties) similar to biological neural network topology, exceeding what random network models predict. Similarity should scale with model complexity. March 2, 2026 Not yet run Not yet started. Planned analysis of published embedding space data against the HCP connectome baseline. Randomly initialized networks of equivalent scale. Networks trained on synthetic data with known topology. Null hypothesis: topology is determined by architecture and training procedure alone. Planned 3 model families planned (GPT-class, BERT-class, MoE) Not yet measurable Complete embedding extraction for all three model families. Pre-register analysis protocol before final comparison is run. — Also documented as COSMIC-NBI-003 on the main Results Registry.
NBI-004 NBI Integrated Information (Phi): Consciousness Threshold Test Combination / External Data Tononi's integrated information measure (Phi) applied to transformer attention patterns during inference should scale toward biological threshold values in sufficiently complex NBI systems. March 2, 2026 — — Phi calculated for equivalent-scale random networks and shallow networks. Biological threshold reference: Tononi et al. baseline values for conscious vs. non-conscious states. Planned Not yet run — Implement Phi calculation for attention matrices. Identify appropriate model checkpoints for comparison. Pre-register threshold criteria. — —
NBI-005 NBI Survival Overhead Gradient: Structured Performance Divergence Combination / External Data Performance gap between biological and NBI systems should be structured by task type following information-theoretic predictions: NBI should excel at tasks with low embodiment requirement and biological systems should retain advantage on tasks requiring continuous environmental integration. March 2, 2026 — — Task battery matched for overall difficulty but varying on embodiment-dependence dimension. Null hypothesis: performance gaps are random or reflect training data differences rather than architectural constraints. Planned Not yet run — Design task battery. Establish scoring protocol. Pre-register task categories and predicted direction of results before running. — —
Back to Program Page Main Results Registry