Anushka Sharma

Anushka Sharma

she/her

Undergraduate AI researcher specializing in open-ended evolution and AGI safety. I design evolutionary frameworks that sustain novelty without explicit fitness functions, addressing reward hacking in AGI.

Genesis: An Empirical Platform for Studying Open-Ended Evolution Without Fitness Functions

Published in Genetic and Evolutionary Computation Conference (GECCO 2026) – Late-Breaking Abstracts, 2026

This late-breaking abstract introduces Genesis — a sham-controlled empirical platform built to study open-ended evolution in the absence of explicit fitness functions. Where most evolutionary computation research assumes a fitness landscape, Genesis is designed to rigorously test what happens when that assumption is removed.

Platform capabilities:

  • Sham-controlled experimental design to isolate the effect of constraint mechanisms from confounds
  • Reproducible multi-run statistical analysis across thousands of generations
  • Built-in support for two novel viability metrics: CARP (Cumulative Activity Rate Preservation) and PNCT (Phenotypic Novelty Continuity Threshold)
  • Modular architecture spanning V2 through V5, with each version targeting a distinct research question

Empirical results (Genesis Engine V2):

  • 58.3% sustained activity rate over 10,000 generations across 12 independent runs
  • Statistical significance: p < 0.01, Cohen’s d = 1.47 (large effect size)

Genesis Engine version history:

  • V2 — Constraint-driven sustained activity (GECCO 2026 main track submission)
  • V3 — One-million-generation sham-controlled null result study
  • V4 — Meta-evolution of physics via CPPN/NEAT
  • V5 — Structure/function decoupling as a precursor to deceptive alignment (CCS 2026)

Venue: GECCO 2026 Late-Breaking Abstracts (Poster) — accepted, to appear.

ResearchGate Preprint: View on ResearchGate

Recommended citation: Anushka Sharma. (accepted). "Genesis: An Empirical Platform for Studying Open-Ended Evolution Without Fitness Functions." In Genetic and Evolutionary Computation Conference (GECCO '26) – Late-Breaking Abstracts. To appear.
Download Paper