Instead of using only one notion of selection I argue for a broader typology of different types of selection. Three such types are differentiated, namely simple one-step selection, iterated one-step selection, and multi-step selection. It is argued that this more general and more inclusive typology might face more effectively the possible challenges of a general account of selection.

I argue that population-level selection does not necessarily have to be invoked to explain the polymorphism at the MHC locus. I argue that the authors' attempt to model operant conditioning in Darwinian terms faces a serious problem. Depending on how many operant responses we take to comprise a sequence, different conclusions about whether or not evolution is occurring in an operant lineage will be reached.

This commentary is an attempt to sharpen some of the issues raised in the paper and thereby increase the generality of the proposal. Some implications of an exact definition of behavior for strategies of measurement and hence behavioral variability are discussed. The role of both respondent and operant behavior in natural selection is emphasized.

Hull et al.'s construction of operant learning as an instance of selection gives rise to problems that weaken this application of selection theory beyond acceptable limits. We point out that most fundamental is a disregard for the need to include multiple concurrent replicators in any definition of selection and indicate how this problem may be solved.

Activity anorexia illustrates selection of behavior at the biological, behavioral, and neural levels. Based on evolutionary history, food depletion increases the reinforcement value of physical activity that, in turn, decreases the reinforcement effectiveness of eating – resulting in activity anorexia. Neural opiates participate in the selection of physical activity during periods of food depletion.

An analysis of biological selection aimed at deriving a mechanism-independent definition removes Hull et al.'s obligatory requirement for replication of the carriers of information, under conditions, such as those obtaining in the nervous system, where the information content of a carrier can be modified without duplication by an amount controlled by the outcome of interactions with the environment.

The interaction of the organism with the environment requires not only reactive, but also active behavior (i.e., search activity) which helps subject to meet the challenge of the uncertainty of the environment. A positive feedback between active behavior and immune system makes the selection process effective.

Campbell (1960) proposed a “blind-variation and selective retention” model of creative cognition. Subsequent researchers have developed this BVSR model into a comprehensive theory of human creativity, one that recognizes that human creativity operates by more than one cognitive process. The question is then raised of how the BVSR model can be accommodated within the Hull et al. selectionist system.

Hull et al. make a direct connection between selection and replication. My view is that selection, at its causal crux, is not inherently connected to replication. I make plain the causal crux of selection, distinguishing it from replication. I discuss implications of my results for Hull et al.'s critique of Darden and Cain (1989).

Selection as a mode of causation is central to operant psychology. Response variation and selection by consequences corresponds to phenotypic variation, to differential survivability, and to reproduction of variants. In natural selection, genes code for phylogenic history but no analogous processes exists for coding behavioral history. Neuroscience suggests potential processes but the conceptual status of these events requires clarification.

Any general account of successful selection explanations must specify how they avoid being ad hoc or vacuous, hazards that arise from their recursive form.

Hull et al. propose a set of features that can be used to identify explanatory systems as being selectionist. The commentary questions the necessity of identifying a unit of retention at the level of behavior-environment interactions. It is concluded that an answer to the question “Are units of retention necessary?” can only be given in light of a clear statement of the goals of a science of behavior.

Hull et al.'s analysis of operant behavior in terms of interaction and replication does not seem consistent with a genuine selection model. The putative replicators do not replicate, and the overall process is more reminiscent of directed mutation than of natural selection. General analogies between natural selection and operant reinforcement are too superficial to be of much scientific use.

For a long time, several natural phenomena have been considered unproblematically selection processes in the same sense of “selection.” In our target article we dealt with three of these phenomena: gene-based selection in biological evolution, the reaction of the immune system to antigens, and operant learning. We characterize selection in terms of three processes (variation, replication, and environmental interaction) resulting in the evolution of lineages via differential replication. Our commentators were largely supportive with respect to variation and environmental interaction but critical with respect to replication, in particular its appeal to information. With some reservations, our commentators think that our general analysis of selection may fit gene-based selection in biological evolution and the reaction of the immune system but not operant learning. If nothing else, this article shows that the notion of selection is not as straightforward as it may seem.