Evolution is a population process across generations. Inheritance can preserve variation, mutation can introduce new variants, natural selection can bias reproductive success, genetic drift can change frequencies by chance, migration can mix populations, and environmental change can reverse which variants are favored. None of these processes implies a goal, a ladder of worth, or an organism changing because it "needs" to.
For a diploid population with two alleles A and a, random mating and the Hardy-Weinberg assumptions give genotype frequencies p^2, 2pq and q^2 after mating. Move the allele frequency and watch the genotype proportions follow.
Set genotype fitness values and follow one allele through generations. Here, fitness means expected reproductive contribution in this model and environment. It is not a measure of worth, intelligence, complexity, or moral value.
Run repeated Wright-Fisher populations from the same starting frequency. Small populations wander more strongly, and some replicates eventually lose or fix an allele even though no allele has a selective advantage.
Use a simplified haploid two-allele model. A mutates to a at rate mu, a mutates back to A at rate nu, and selection changes reproductive contribution before mutation is applied.
Two populations experience opposite local selection but exchange migrants each generation. Increase migration and watch their allele frequencies become more tightly coupled.
Let the environment oscillate so that allele A alternates between advantage and disadvantage. The favored allele can change before the population fully catches up, creating lag, reversals and trade-offs rather than a permanent "best" type.