Limitations of the Model

The preceding models are admittedly extremely simplified and very speculative. The analogs of very complex phenomena which they display capture few to none of the actual intricacies of depression. As such they are intended only as springboards for thinking about an otherwise somewhat impenetrable and tangled set of apparently disparate biases. Drew McDermott's (1981) caution that we should not let ``artificial intelligence'' approach ``natural stupidity'' by positing computers that ``think'' or actually get ``depressed'' must be interpreted in the strongest sense.

Additionally, the following concerns regarding the network's performance even in the limited domain it does try to model are notable. First, the model does not predict any processing advantages involving the semantic content of nondepressotypic negative stimuli by people. Yet, many studies show relative advantages for recall of specifically presented negative words (e.g., Bradley & Mathews, 1988, Derry & Kuiper, 1981). The current model, generalized to account for memory effects, might predict that negative stimuli not associated with the particular loss exhibited by the network should not be better recalled than other stimuli. A host of factors might explain this phenomena, such as semantic links between negative words, etc., but such explanations would require further additions to the current model. Also, the model does not yet generate multimodal response distributions detected in the human data. The model does not differentiate between short and long term memory and thus, mood state effects cannot be differentiated in the network from trait effects.