Computational backbone for the dissertation "Intelligence is Models for the Negative Space".
This crate reproduces all key figures and results from the 5 laws of intelligence-as-negative-space.
A reinforcement-learning agent receiving only negative feedback (no positive reward) discovers the safe subset of its environment. The agent converges to ~60% avoidance through pure negative feedback (Act0).
Key result: ≥ 55% avoidance rate from negative feedback alone.
Organisms in stochastic environments exhibit a 294:1 avoid-to-choose ratio, reflecting evolved loss-aversion where avoidance responses vastly outnumber approach responses.
Key result: Avoid:Choose ratio ≈ 294:1.
Multi-species competitive Lotka-Volterra dynamics with intra-specific competition exceeding inter-specific competition yield 100% ecological resilience — no species goes extinct.
Key result: All species persist above extinction threshold.
Population-level (group) selection outperforms individual-level selection by +0.075 fitness units, demonstrating the evolutionary advantage of collective negative-space modeling.
Key result: Population fitness advantage ≈ +0.075 over individual selection.
The avoidance ratio A = avoided / (avoided + chosen) is conserved across ecological scales, with standard deviation ≈ 0.001 across population sizes ranging from 100 to 50,000.
Key result: std(avoidance ratio) < 0.01 across all scales.
use dissertation_engine::DissertationReport;
fn main() {
let report = DissertationReport::generate(42);
println!("{}", report.render());
}cargo testThe test suite includes 25+ tests that reproduce all 5 laws and validate supporting infrastructure (RNG, statistics helpers, simulation components).
- Pure Rust, no
unsafe, no external dependencies - Deterministic via seeded PRNG (SplitMix64)
- Each law is a self-contained module with a top-level reproduction function
DissertationReportaggregates all results into a single formatted report
MIT