In an era characterized by a growing convergence among evolutionary biology, neurobiology, and behavioral sciences, Mitchell et al. make a bold claim. The authors argue that in humans (and all other animals with the explicit exception of Aplysia), learning to associate stimuli requires the formation of propositions (symbolic representations with truth-value) by means of conscious and controlled reasoning. Although swimming towards the (main)stream is sometimes necessary to reach the source, there are several reasons to believe that the authors are heading in the wrong direction.
In principle, the proposal that the formation of propositions is necessary for learning to occur – a process subserved by protein synthesis – is not different from the absurd claim that successful protein syntheses in the digestive systems must also be preceded by propositional representations. In fact, given our current knowledge of biological systems, it is highly unlikely that humans, not to speak about other animals, are conscious or in control of the majority of processes underlying learning and memory formation. Nonetheless, they may be both (i.e., conscious and in control) in regards to the expression of the acquired learning.
In an attempt to avoid linking their proposal to established biological principles of learning, the authors point towards the distinction between the psychological and the neural level of explanation and argue that their thesis applies only to the former. However, this illustrates a general problem inherent in the presented approach. By limiting the phenomena under investigation by either defining out of existence critical aspects of associative learning (e.g., their biological principles) or neglecting several lines of existing research (e.g., fear conditioning and lesions studies in humans), the authors end up proving little more than their assumptions. It is also noteworthy that the notoriously problematic terms “conscious” and “awareness,” although central to the argument, are not explicated (except that the reader is reassured that “Aplysia do not have conscious beliefs”; target article, sect. 6.3, para. 3).
Here, we highlight three more specific problems with the approach presented by Mitchell et al.
1
The notion of a unitary type of associative learning resting on conscious awareness sits very uncomfortably with established ideas in evolutionary biology. Evolution is commonly conceived as a slow accumulative process, building layer upon layer of brain tissue that incorporates successful adaptations at one level into more complex functions at higher levels. As a consequence, we share many behavioral systems and their associated neural circuitry with our primitive predecessors, unlikely candidates for using awareness as their primary principle of learning. Nor does it seem a likely evolutionary feat to have reorganized the human brain for exclusive use of this principle to modify behavior. Rather, from the evolutionary perspective, many different forms of learning would be expected, as elaborated by, for example, Gregory Razran (Reference Razran1971). The MacLean (Reference MacLean, Lewis and Haviland1993) concept of a “triune brain” is one, often discussed, example of layered evolution of this kind, which directly implies that there are at least three levels of behavioral organization, each of which may incorporate associative learning: one concerned with reflexes and instincts (brain stem and striatum), a second that incorporates emotion and autonomic control (the limbic brain), and a third level concerned with instrumental behavior and cognition (thalamus and the cerebral cortex).
2
Related to the lack of compatibility with evolutionary thinking is the omission of several lines of research within the neurosciences. A contemporary version of the MacLean concept is the model of rodent fear conditioning by LeDoux (Reference LeDoux1996), Davis (Reference Davis1992), Fanselow (Reference Fanselow1994), Maren and Quirk (Reference Maren and Quirk2004), among others, which has been confirmed in human brain imaging studies (Morris et al. Reference Morris, Öhman and Dolan1998). Because this model posits that the input to and output from the central hub in the fear network do not necessarily have to go through the cortex, it strongly implies that the fear network and its modification through fear conditioning are independent of conscious awareness. Therefore, this model (and its elaboration for human fear conditioning by Öhman & Mineka Reference Öhman and Mineka2001) clearly implies two levels of learning that are partially independent but also interacting. This model provides an articulated version of dual-process theory that integrates neuroscience and behavior and is now supported by a host of both behavioral (Hamm & Vaitl Reference Hamm and Vaitl1996; Öhman & Soares Reference Öhman and Soares1998) and imaging work (Critchley et al. Reference Critchley, Mathias and Dolan2002), showing that conscious awareness of the associated stimuli or their contingency is not necessary for learning to be acquired and expressed.
Providing further support for the independence of (at least) two kinds of learning is the work on patients with lesions on the hippocampus, a structure known to be critical for the formation of declarative memories. Following fear conditioning, these patients fail to report the contingency between two associated stimuli (e.g., a neutral tone or image and an aversive shock) in a fear conditioning paradigm, but they show a normal conditioned response as measured by the skin conductance response, SCR (Bechara et al. Reference Bechara, Tranel, Damasio, Adolphs, Rockland and Damasio1995). In contrast, patients with lesions to the amygdala, a key player in the brain's fear network and known to be necessary for the implicit expression of learned fear, display the opposite response pattern with intact declarative memory, but an impaired conditioned response (spared conditioned response or SCR) (Bechara et al. Reference Bechara, Tranel, Damasio, Adolphs, Rockland and Damasio1995; Weike et al. Reference Weike, Hamm, Schupp, Runge, Schroeder and Kessler2005). These findings show a striking dissociation between explicitly (propositionally) and implicitly (SCR) expressed emotional learning. Further support along the same lines is the demonstration of fear conditioning to unseen visual stimuli in a cortically blind patient with bilateral lesions to the primary visual cortex (Hamm et al. Reference Hamm, Weike, Schupp, Treig, Dressel and Kessler2003).
Taken together, the findings listed above should make it clear that the psychological and neural levels of explanations are tightly coupled and that psychological models of learning can benefit tremendously by drawing from what is known in the neurosciences.
3
In their argumentation against the “dual-process” theory of learning, Mitchell et al. build a straw man around the claim that the SCR as a measure of learning is unaffected by conscious and controlled cognitions. This purported claim, ascribed to the dual-process camp, is then used to refute that emotional learning can occur without conscious and controlled reasoning. Indeed, some work has shown that the SCR can be used to index learned responses to both consciously and non-consciously perceived stimuli (Öhman & Mineka Reference Öhman and Mineka2001; Olsson & Phelps Reference Olsson and Phelps2004). However, the authors neglect the literature which claims that the potentiation of the startle reflex may be less affected by propositional and declarative processes. Whereas the SCR has been shown to be more sensitive to cognitive processes, such as propositional reasoning, which is likely to be cortically mediated, the potentiating of the startle reflex and eyeblink conditioning draw mainly on subcortical and cerebellar mechanisms, respectively (Clark & Squire Reference Clark and Squire1998; Davis Reference Davis2006). This is supported by the accumulating evidence of a dissociation between the startle response and SCR, in which the SCR tracks conscious awareness of stimuli contingencies, whereas the startle response tracks non-conscious learning (Hamm & Vaitl Reference Hamm and Vaitl1996; Weike et al. Reference Weike, Schupp and Hamm2007).
To sum up, the propositional account of associative learning proposed by Mitchell et al. may be parsimonious, but it critically lacks the compatibility with current evolutionary biology and neuroscience. Cultivating this form of “biological blindness” will not advance our understanding of behavioral phenomena, such as associative learning.
