We describe and investigate the learning capabilities displayed by a p
opulation of self-replicating segments of computer code subject to ran
dom mutation: the tierra environment. We find that learning is achieve
d through phase transitions that adapt the population to the environme
nt it encounters, at a rate characterized by external parameters such
as mutation rate and population size. Our results suggest that most ef
fective learning is achieved close to the transition to disorder, and
that learning curves of evolutionary systems are fractal.