More than one pathway to action understanding

Abstract

Many believe that the ability to understand the actions of others is made possible by mirror neurons and a network of brain areas known as the action-observation network (AON). Despite nearly two decades of research into mirror neurons and the AON, however, there is little evidence that they enable the inference of the intention of observed actions. Instead, theories of action selection during action execution indicate that a ventral pathway, linking middle temporal gyrus with the anterior inferior frontal gyrus, might encode these abstract features during action observation. Here I propose that action understanding requires more than merely the AON, and might be achieved through interactions between a ventral pathway and the dorsal AON.

Figure 1

The human AON. (a) This schematic shows the three reciprocally connected areas of the human AON. The areas known to contain mirror neurons are the ventral IFG, shown in red, and the inferior parietal area, shown in green. These two areas are reciprocally connected [50] creating a premotor-parietal mirror system. Neurons within the STS, shown in blue, have also been shown to respond selectively to biological movements, both in monkeys [51] and in humans [52-54]. The STS is reciprocally connected to the inferior parietal area [55,56] and therefore provides visual input to the mirror system. (b) This schematic shows the predictive coding model of the AON. Predictive coding is based on minimising prediction error though recurrent or reciprocal interactions among levels of a cortical hierarchy. In the predictive coding framework, each level of a cortical hierarchy employs a generative model to predict representations in the level below. This generative model uses backward connections to convey the prediction to the lower level where it is compared to the representation in this subordinate level to produce a prediction error. This prediction error is then sent back to the higher level, via forward connections, to adjust the neuronal representation of sensory causes, which in turn changes the prediction.

Figure 2

A schematic of the two-pathway framework. In this schematic the ventral pathway of the connected areas MTG, BA47, BA45 and BA44/BA6 is shown in red and the dorsal AON pathway is shown in green.

Figure 3

An example of action understanding in the two-pathway framework. In this example the intention is to drink a cup of tea. The first step is visual processing and identification of the object as a cup. The second step is the retrieval of actions that we have learned to be associated with that object. The third step is the selection of the most probable actions given the intention. Note here that more than one action can be selected but that the likelihood of the action can be signalled through the strength of that action’s representation, indicated here by the transparency of the picture. The top action is not probable and is not selected. The fourth step is the encoding of the motor parameters to generate a prediction of the sensory consequences of the observed action. Again multiple actions can be encoded as before. The fifth step is the prediction of the sensory consequences of the most probable action. Here only the most probable action is encoded. In this schematic, steps 2–4 would be encoded in the ventral pathway of the connected areas MTG, BA47, BA45 and BA44/BA6 with the representation of the action changing from the abstract to the concrete through these steps. Steps 4–5 would represent the generation of the predicted sensory consequences of the action encoded in the dorsal AON pathway.