Experiments
Our research program consists of seven designed experiments, each testing specific predictions from established theories of consciousness against AI systems. The program uses Design of Experiments (DOE) methods. Each report distinguishes its planned hypotheses, later supplemental measurements and reproducibility limits.
Fractal Embeddings
CompleteComplete. Testing whether a large body of machine-readable meaning carries a fractal fingerprint — every threshold frozen before the data was touched. Two tests confirmed, three came back negative, and the negatives are the finding.
View details →BuddhaBERT
ActiveThe four-model study is measured: confidence variance remains unresolved, attention smoothing is largely mechanical, and robustness and paraphrase results show no consistent advantage. Read the public working report.
View details →Moltbook Antfarm
DesignedWhether simple agents following only local rules develop group behavior nobody programmed — and whether it arrives suddenly, as a phase transition, rather than gradually. Includes a proposed test for telling real emergence from mere aggregation.
View details →Embodied Embeddings
PlannedTesting whether grounding language in simulated sensory and motor experience produces word and sentence representations with qualitatively different computational properties than text-only training.
View details →Digital Game of Life
ActiveUsing Conway's Game of Life as a calibration bench for emergence detectors — testing whether compression ratio and integrated information (Phi) can rank pattern classes correctly on a substrate where the right answer is not in doubt.
View details →IIT Emergence
DesignedDirectly testing Integrated Information Theory predictions about consciousness in neural network architectures of varying structure, measuring Phi across different topologies and training regimens.
View details →The Forgery Test
PlannedAdversarial validation of consciousness markers — training a generator to imitate the signatures of consciousness, then finding which markers resist faking. The imitation-resistant ones are the measurements worth trusting.
View details →