observe or imagine another individual’s pain73–75. This finding has led researchers to propose that these shared emotional representations are
involved in empathy67,68,73–75
. These regions also track reported fair-
ness of another individual in pain after a competitive game52 and empathic concern after receiving instructions to take the other’s emotional perspective76.
Socially mediated variability in affective response to another’s distress is likely to influence ensuing learning. An fMRI (functional magnetic resonance imaging) study on observational fear learning42 found activation in the ACC and anterior insula both during observation of another person receiving shocks paired with a CS and in the later test stage when the person being imaged expected to receive shocks accompanying the same stimulus, indicating that regions linked to empathy may be involved in observational fear learning. This assumption was further supported by the finding that activation in both these regions during observation predicted learning as expressed in the subsequent test stage. In addition, another region of interest, the rostral MPFC, was only activated during the observation stage. Responses in this region marginally predicted the magnitude of subsequent learning. The MPFC is implicated in thinking about one’s own and others’ mental states77–79, indicating that social cogni- tion may be involved in observational learning of fear.
In accordance with research on nonhuman animals, the findings of observational fear learning in humans demonstrate, on the one hand, an independence of conscious awareness and strategic regulation of affective responses and, on the other, a dependence on social and contextual manipulations. However, it seems likely that the amygdala, supporting automatic affective responses, interacts with the orbito- frontal cortex, the temporal lobe and the MPFC that together mediate social and contextually regulated processes, to produce an adaptive affective response (Fig. 2b). It remains to be explored what social factors cause formation of a learned fear response through social observation and what neural systems support these social influences.
Based on this research and the brain’s connectivity, it is possible that amygdala-centered observational fear learning in both rodents and primates is supported by automatically activated cortical mechanisms of shared affective representations in the anterior insula, as well as more explicit hippocampal representations about context and relevant social information about the learning model (such as social status and familiarity). Although in rodents the ventral MPFC has a role in some social behaviors59, the primate MPFC is likely to be more important in social perception and learning, as shown by deficits in social behavior after prefrontal lesions in both monkeys80 and humans81. However, the more anterior-rostral region of the MPFC is both quantitatively and qualitatively more developed in humans as compared with other primates82, implying a neural substrate for the support of more complex mental representations that might be involved in human observational learning. A meta-analysis of imaging studies reports that this region is especially sensitive to experiments involving both social and emotional tasks83. In sum, regardless of the complexity of the underlying neural representations, the research discussed above shows that a conspecific’s emotional display can serve as an US, stressing the similarity with conventional conditioning.
Instructed fear learning Humans possess the unique ability to obtain emotional information through language. Whereas fear learning through observation involves visual representation of emotional properties of a stimulus, language is arbitrarily related to, and thus detached from, its referent in the world. Language forces the receiver to rely on similar past experiences and internally generated imagery to establish an emotional memory.
NATURE NEUROSCIENCE VOLUME 10
SCR difference (CS+ vs. CS–)
0.5 0.4 0.3 0.2 0.1
Figure 4 Mean expression of learned fear as assessed with difference in skin conductance response (SCR) in three groups of subjects after conditioned, observational and instructed fear learning. CS+, CS paired with US; CS–, CS unpaired with US. Asterisks, learned fear response significantly greater than zero (*P o 0.05; **P o 0.01; NS, not significant). Dark blue bars, responses to supraliminal (perceived) CSs; light blue bars, responses to subliminal (unperceived) CSs. For all three learning groups, unmasked as compared with masked CSs elicited a stronger learning response that was equivalent across groups. In the conditioned and observational, but not in the instructed, groups, a fear response was also elicited to unseen (subliminal) CSs.
Imagery and self-projection into the future are thought to rely on neural systems similar to those involved in perception84 and episodic memory formation85. Like recollection of the past, projection into the future is impaired after hippocampal lesions86. In addition, regions of
MPFC implicated in simulation of future events87,88 involved in thinking about others’ minds.
overlap with those
Both clinical accounts that retrospectively target the etiology of phobic fears89 and experimental studies on children involving fear provoked through storytelling90 reveal that verbal instructions can be a strong stimulus for fear learning. Along the same lines, adults instructed to expect a shock paired with a specific CS and later exposed to the same CS show learned responses similar to those seen after classical fear conditioning41,91–93.
To directly compare fears acquired through conditioning, observa- tion and verbal instruction, we41 manipulated the learning procedure, keeping other factors constant. Conditioned stimuli acquired their threat value through being paired with a shock, with observed fear expression in another person or with the experimenter’s verbal instruc- tions (Fig. 1a–c). Fear responses to the CS were of comparable magnitude after the three kinds of learning (Fig. 4). In addition, replicating previous findings21, a subliminally presented (unperceived) CS triggered a response in the fear conditioning group. The observa- tional, but not the instructed, group also showed a learning response to subliminal presentations of the CS, indicating that common learning mechanisms may underlie fear learning through conditioning and observation, but a different mechanism may support learning through language. These results support the notion that there are partially dissociable systems involved in different modes of social, emotional learning. Classical conditioning and observational learning, which humans share with many other species, might be supported by an evolutionarily old system that predates the emergence of language. In contrast, learning based on language is unique to humans and is likely to be, at least initially, dependent on representa- tions in higher cortical areas that also support conscious processes. Indeed, these findings indicate that such cortically represented fear associations might depend on conscious awareness, in accordance with the observation that conscious awareness can be used to distin- guish subdivisions of conditioning (such as context versus cue or trace versus delayed).
To examine the mechanisms underlying expression of fears acquired through verbal instruction, Phelps and colleagues93 told subjects they might receive a shock when shown a square of a particular color (‘threat’ stimulus), but not another color (‘safe’ stimulus). Supporting