The Baldwin Effect and Evolution

Hillis [7, pages 204-208] points out that biologists currently believe that evolution does not proceed as a steady hill climbing process. ``Evolution moves by leaps and bounds, alternating with periods of stasis. Species remain in virtual equilibrium where fitness is suited to the environment. Then a sudden change in the environment, or an empowering mutation, causes an abrupt jump in fitness, as new and effective physical characteristics express themselves in the phenotype of the species.''

Hillis was able to produce and study this behavior in his experiment with artificial organisms. What he found was that while the population seemed to be resting as far as its phenotype were concerned, the underlying genetic makeup was actively evolving. Eventually, enough genetic change occurred to produce the phenotype change through epistasis. This latter change was the abrupt change seen in the fitness of the species.

In our studies of the Baldwin effect we have observed similar evolutionary behavior. Figure 3 shows the behavior of one learning run with p_m = 0.002.

Figure: ALI vs. generations (p_m = 0.002)

As can be seen in this figure, the improvement in the learning rate is not a smooth change. The value of ALI oscillates moving slowly downward. But suddenly (i.e., in relatively few generations) it decreases sharply. Finally, it returns to a state of relative equilibrium.