Learning and Cognition as Adaptations

The comparative analysis of learning and cognition can proceed along two lines, both consistent with the definition of evolution adopted in this book. In Chapter 1, evolution was defined as a phenomenon resulting from the interplay of evolutionary processes, such as natural selection and genetic drift, and evolutionary patterns, such as the constraints imposed by phylogenetic history. Consequently, comparative studies of learning and cognition have traditionally centered either on the assumption that learning mechanisms are the result of evolutionary processes and thus possess adaptive significance (this section), or they are the result of evolutionary patterns, thus reflecting phylogenetic history (next section).

Equipotentiality vs. CS→US Relevance

CS~^US relevance (also known as “constraints on learning,” “preparedness,” and “associative selectivity”) refers to the proposal that some combinations of signals, responses, and reinforcers are more easily associated than others. These terms reflect a departure from a principle of equipotentiality according to which the same mechanisms apply to learning involving any type of stimuli or responses (Domjan, 1997). Several findings suggest that learning mechanisms are adaptively specialized for the resolution of species-typical ecological problems. The main alternative to CS-^US relevance is that these findings are explained in terms of contextual variables (e.g., sensory-perceptual, motivational-emotional, or motor mechanisms).

Consider the following example (Garcia & Koelling, 1966). When rats were exposed to pairings between a taste or an audiovisual signal with either X-ray irradiation (which induces nausea) or electric shock (which causes peripheral pain), they seemed to learn the Taste^Nausea and Audiovisual^-^Pain associations better than the Taste^Pain and Audiovisual^-^Nausea associations. Although the original report had overlooked some important controls, subsequent experiments under controlled conditions yielded similar results. In one experiment (Miller & Domjan, 1981), rats were trained with one of two signals paired with one reinforcer, while the other reinforcer was presented by itself the following day (Table 7.5). This training procedure ensured that each rat was exposed to both reinforcers, only one of which was paired with the target signal. Licking water from a tube produced a flashing light (visual signal), whereas licking from another tube produced saccharin (taste signal). The reinforcers (Shock or LiCl) were delivered 60 s after the first lick that produced the signal (Light or Saccharin). Additional injection controls received either Light or Saccharin paired with a harmless injection of isotonic saline. To control for reinforcer exposure, rats trained with Shock were exposed to LiCl the following day, whereas rats trained with LiCl were exposed to Shock the following day. In a final preference test, rats trained with the Shock rejected the drinking tube that produced the Light more than the tube that yielded Saccharin, but rats trained with LiCl rejected the drinking tube that yielded Saccharin more than the tube that produced the Light. Thus, Lighf^Shock worked better than Lighf^LiCl, whereas Saccharin^-^LiCl worked better than Saccharin^-^Shock (Figure 7.10). Because both signals and both reinforcers were effective in one combination, failure of the other to yield evidence of conditioning cannot be attributed to the use of inadequate stimuli. These results are consistent with the notion of CS-^US relevance.

Table 7.5 A demonstration of CS^US relevance in classical conditioning

Note: Day 6 was a resting day. Light: drinking water from a spout activated a flashing light. Saccharin: drinking from a spout that yielded a 0.15% saccharin solution. LiCl: an intraperitoneal injection of lithium chloride that produces gastrointestinal sickness and nausea. Saline: an intraperitoneal injection of isotonic saline solution used as a control for the injection procedure. Notice that there was a single CS~^US pairing and that all the rats were exposed to both reinforcers (LiCl and Shock).

The ideal approach to determining that ecological pressures lead to evolutionary divergence in learning mechanisms is to compare closely related species adapted to different environments (Domjan & Galef, 1983). CS^US relevance could reflect the operation of an adaptive learning mechanism for animals with a generalized diet that are likely to sample novel food items constantly, including poisonous foods. This selective pressure could favor the evolution of mechanisms for the rapid acquisition of food aversions that would minimize the intake of poison. This hypothesis has been tested in feeding specialist vs. feeding generalist species of kangaroo rats. Generalist species consume grains, seeds, and a variety of plant materials (e.g., Dipodomys merriami), whereas specialists are almost exclusively restricted to the leaves of chenopod plants, like shadscale (e.g., D. microps). These two species differ in diet but belong to the same genus, are sympatric (they live in the Owens Valley, California), are similar in body size and shape, and can be maintained in the laboratory under similar conditions. Consistent with the notion of adaptive specializations in learning, the generalist D. marriami acquired the taste aversion faster than the specialist D. microps (Daly, Rauschen-berger, & Behrends, 1982).

Evidence consistent with the notion that CS^US relevance reflects adaptive specializations comes from a study involving artificial selection in the fruit fly (Dunlap & Stephen, 2014). Four populations of flies were exposed to training protocols that predicted different degrees of reproductive success combining colors and odors (signals) with quinine (aversive reinforcer). For one group, Odor was a reliable predictor of Quinine, but Color was not; a second group had Color reliably predicting Quinine, whereas Odor was unreliable. Finally, for two other groups, Color and Odor were either both highly reliable or both unreliable (these were controls). Once flies had been trained in these protocols, they were given a choice to lay eggs in a context marked with the Color or Odor experienced during training. The experimenter determined the probability that the eggs would be allowed to hatch in each condition, thus introducing differential reproductive success paired with the choice of context, which, in turn, depended on previous training. After 40 generations of selection, fruit flies demonstrated faster learning with Odor after Odor^-^ Quinine selection, but faster learning with Color after Coloré Quinine selection, when these signals had been reliable predictors of reproductive success. Learning about Color or Odor when signals have been unreliably paired with reproductive success in previous generations was not differential.

Figure 7.10 CS^US relevance in flavor aversion learning in rats. See Table 7.5 for a description of the design. When the signals (Light, Saccharin) were tested, their effectiveness depended on the reinforcer (LiCl, Shock) they predicted. LiCl: lithium chloride, a toxin that induces gastrointestinal discomfort.

These results suggest the possibility that some associations carry such an adaptive weight that animals would learn them faster than associations that have not influenced reproductive success in the animal’s ancestors. Whereas this is a realistic possibility, there are some problems of interpretation that highlight the complexity of these phenomena. First, could contextual factors (sensory-perceptual, motivational-emotional, and motor) explain some of these effects? For example, in the experiment with kangaroo rats, species differed not only in the strength of their food aversion learning, but also in their initial reaction to the sunflower seeds. Perhaps the generalist D. marriami showed better food aversion because of neophobia—reluctance to consume novel foods. If D. merrioni is exposed to a novel food for which it is less neophobic (wheat grain), then there is no evidence of food aversion and no difference with the specialist D. microps (Daly, Rauschenberger, & Behrends, 1982). Rapid food aversion conditioning could thus result from unconditioned neophobia, rather than from CS“^US relevance.

Second, are animals learning, but not performing? For example, Saccharin^-^ Shock pairings lead to better performance when a temporal delay is interpolated between the signal and the reinforcer (Krane & Wagner, 1975). While no aversion was detected with a 5-s delay, a substantial aversion to the Saccharin developed with delays of 30 and 210 s. Taste solutions require a considerable CS-US interval to be associated with either gastrointestinal poisoning or peripheral pain. Similarly, although the initial results suggested that audiovisual stimuli failed to become associated with nausea, subsequent studies reported positive results. In one experiment (Batson & Best, 1979), pairings of a novel context with LiCl blocked subsequent acquisition of a Saccharin^-^ LiCl association in the same context. Such blocking suggests that the context had been associated with the nausea induced by LiCl.

Third, what is the appropriate procedure for demonstrating CS^US relevance? For example, in the fruit fly study, two signals were paired with a single reinforcer. However, in the original demonstration by Garcia and Koelling (1966), two signals and two reinforcers were used, and relevance was suggested by the interaction between these conditions. This design shows that both signals and both reinforcers are viable, but only when presented in a specific arrangement. Thus, the fruit fly experiment would require the addition of a second reinforcer to demonstrate CS^US relevance (Dwyer, 2015).

An implication of the research on food aversions is that natural selection can achieve the fine tuning of a particular behavioral process by adjusting contextual factors (e.g., neophobia, stimulus trace dynamics) without modifying associative mechanisms.