Mitchell et al. contribute their view to a long-standing controversy in psychology: Namely, can the range of human and animal behavior be sufficiently attributed to a single processing system rather than multiple competing systems? The authors propose that a single propositional learning system drives behavior. In this learning system, “controlled reasoning processes are necessary for learning to take place, and learning results in beliefs about the relationship between events” (sect. 1, para. 8). The major aspect of this position with which we disagree is: “are necessary for learning.” In this commentary, we point out several phenomena that are difficult to explain within a single propositional learning system as it is described.

Mitchell et al. contend that associative learning in the Pavlovian situation involves propositional knowledge and not the automatic formation of links between events. The authors indicate that, in their view, Aplysia do not possess propositional knowledge. This view would presumably extend to insects, such as honeybees. Rats and people show three complex conditioning effects: backward blocking, forward blocking, and relative validity. Each of these effects involve multiple conditioned stimuli (CS) occurring concurrently or in isolation along with an unconditioned stimulus (US). Mitchell et al. argue that the resulting patterns of response behaviors in these blocking designs are difficult to explain within a strictly automatic associative-link framework. Therefore, they suggest that these phenomena involve propositional knowledge in humans and possibly in rats. Because, presumably, bees do not possess propositional knowledge but display all three effects, it follows that these effects can occur in the absence of propositional knowledge (see e.g., Blaser et al. Reference Blaser, Couvillon and Bitterman2004; Guez & Miller Reference Guez and Miller2008).

So, the authors are driven to the conclusion that highly similar and complex effects in different species can occur on the basis of highly different learning mechanisms. However, the phylogenic “cut-off” point between these mechanisms is unclear. An alternative possibility is that the same learning mechanism produces the three effects in honeybees, rats, and people. According to that view, the learning mechanism must be some form of simple links rather than propositional. It is much easier to concede that humans can learn like bees than bees can learn via propositional reasoning like humans. Whether or not the same mechanism is responsible for learning in both cases, it must be assumed that links between events occur in order to accommodate the bee data.

Mitchell et al. mention that, at least in more complex animals, it is possible that their single propositional learning system may rely on the multiple connections between subcortical and cortical brain regions (sect. 5.3). As such, it is difficult to imagine an intact propositional system at work in decorticate animals in which most higher brain functions, such as those associated with reasoning skills, have been removed. However, many studies, the earliest of which include Culler and Mettler's (Reference Culler and Mettler1934) research, have found that conditioned learning still occurs following the general or localized removal of cortical regions. For example, rats with as much as 99% of their neocortex removed showed learning in a T-maze that was equivalent to that of fully intact controls (Thompson Reference Thompson1959). At the furthest extreme, the spinal cord alone is sufficient for associative learning to occur (Patterson et al. Reference Patterson, Cegavske and Thompson1973). These findings do not preclude the existence of propositional learning, but they show that it is not necessary for associative learning to take place.

Non-propositional learning also appears to occur within humans. The classic argument for learning without awareness is Claparède's (Reference Claparède1907) description of a woman suffering from Korsakoff's syndrome (Nicolas Reference Nicolas1996). Claparède pricked his patient's hand with a pin hidden in his own, and, although she did not display any declarative knowledge of the experience, she later would withdraw from Claparède when he gestured toward her with his hand. Subsequent research with anterograde amnesiacs provides further examples of similar learning in the absence of conscious, declarative knowledge. For example, Gabrieli et al. (Reference Gabrieli, Carrillo, Cermak, McGlinchey-Berroth, Gluck and Disterhoft1995) found that eyeblink conditioning remained intact in amnesiacs and suggested that declarative information was not needed for CS-US associations to occur (for more examples, see Cohen & Squire Reference Cohen and Squire1980).

It may be argued that aware, conscious reasoning does occur during the learning process in amnesiacs, and the patients are simply no longer aware of this learning experience during the expression of this information at a later time. Even if this were the case, there are two reasons that a propositional-system only account remains problematic based on evidence from amnesiacs. First, it is unclear how learned propositional information can be used later without conscious awareness that the information must be applied. As the target article authors describe their learning system, learned propositional information is expressed through beliefs about the relationships between events, and these beliefs are very unlikely to occur at all in anterograde amnesiacs. Additionally, there is some evidence that learning processes in intact individuals may occur without the need for any conscious awareness of the associations between events in the environment. In tasks such as artificial grammar learning, implicit categorization, and implicit sequence learning, the learning and expression of associations between items or events in the environment occur without the need for awareness. For example, in four-choice tasks that contain long repeating item sequences, participants show evidence of learning the sequence often without awareness or knowledge and concurrently with or in isolation from explicit learning (Song et al. Reference Song, Howard and Howard2007). This further implies that associative learning can take place without reasoning and conscious beliefs about associations.

To conclude, the authors claim that “associative learning is never automatic and always requires controlled processes” (sect. 3.1, para. 2). However, the aforementioned examples provide converging evidence that associative learning can take place without the need for reason or belief. If a single learning system is to account for all forms of learning, the system must accommodate these cases.