From the diagram it may be concluded that the source of information processing biases in the network are twofold. First, training on negativity is reflected in a decrease in the bias term corresponding to positivity in the network, suggesting explicit inattention to positive information, regardless of a word's semantic content in depression. Next, there are larger activations to and from the negativity node in the depressed network, suggesting that when stimuli with a negative valence are processed by the network, the valence node contributes more to the semantic identification of stimuli than in other cases.
While analogs of such a simplistic system to a human brain are tenuous at best, it may be suggested that similar biases might operate in humans at a neurological level. That is, a generic predisposition towards responses to stimuli of one valence might be represented as overactivation of some neurons responsible for valence-discrimination in the limbic system , e.g., the amygdala. Similarly, depression might also be operationalized as a distribution of learned associations with negativity in connections from brain areas more traditionally associated with semantic identification, e.g., the hippocampus. Such an explanation could provide new understanding for the manner in which pharmacological interventions for depression work; they might effectively prevent the overactivation of nodes responsible for valence identification (e.g., by stopping reuptake of some neurotransmitter associated with these neurons), but might not change learned distributed biases. In this way, when drug therapies are stopped, negative information processing biases might still be present to some degree. Therapeutic interventions which target such overlearned distributed biases (e.g., cognitive therapy) might, in fact, change such diffuse representations.
