Genesis Lets AI Evolve Without Fitness Functions – 1M Generations Tested

Biological evolution miraculously generates endless complexity without any explicit fitness function—yet every AI evolutionary algorithm typically depends on one. Genesis, an open-source platform created by researcher Anushka Sharma, attacks this paradox head-on by building a system where evolution is governed only by physical constraints, relational dominance, and adaptive regulation. No scalar fitness, no designer-specified objectives. After more than one million generations, the platform demonstrated that such constraint-driven selection can sustain genuine evolutionary activity after complete fitness removal in 7 out of 12 experimental runs (with a Wilson 95% confidence interval of [30.2%, 82.5%] and a Cohen's d of 1.47 versus baselines).

The results don't stop there. A sham-controlled negative experiment revealed that niche construction alone is insufficient to break through the complexity plateau that often stalls artificial evolution. However, when niche construction was paired with speciation protection—that is, mechanisms that allow distinct lineages to form and persist—the system initiated structural diversification that unprotected secretion could not achieve. These findings establish clear empirical boundaries for fitness-free evolution and open an entirely new research direction: the meta-evolution of physics, where the laws governing an evolutionary system can themselves be evolved. The work was accepted as a Late-Breaking Abstract at the GECCO Companion '26 conference in San Jose, Costa Rica.