It's a worthy subject, that interests me from in both its computational and behavioral aspects. As is the case in most applications of computing, the crucial aspect is the "match" between the model and the domain being modeled.
The behavioral domain is particularly problematic. Humans quite naturally use abstraction as a modeling tool, and the tendency is to construct abstract categories as a basis of sorting the "raw data" of observation. The difficulty in the behavioral realm is that phenomena are indistinct and sharply bounded "boxes" don't correspond very well to the real-world behavioral complexities.
In the article the distinction between psychotic and affective disorders is pointed out as an example of the "softness" of phenomenological boundaries. If we conceptualize the distinction as an array of cardboard boxes side by side, psychosis in one box, mood disorder in another one, the compartments of the model will not fit what is seen in a clinical setting. Rather in the real-world conditions resemble sand dunes, rounded peaks flowing into surrounding soft peaks, and no particular place where it could be said one dune stops and another starts.
Furthermore the confluence of a surprisingly huge array of factors is involved in the appearance or origin of behavior we'd consider problematic (if we could even agree on that).
I'd concur with the idea that computational models need to be consistent with and informed by the reality they are intended to represent, but experience suggests that the more fundamental issue is conceptualization of behavioral reality itself which is far more complex than our usual abstraction of it lead us to consider.
The behavioral domain is particularly problematic. Humans quite naturally use abstraction as a modeling tool, and the tendency is to construct abstract categories as a basis of sorting the "raw data" of observation. The difficulty in the behavioral realm is that phenomena are indistinct and sharply bounded "boxes" don't correspond very well to the real-world behavioral complexities.
In the article the distinction between psychotic and affective disorders is pointed out as an example of the "softness" of phenomenological boundaries. If we conceptualize the distinction as an array of cardboard boxes side by side, psychosis in one box, mood disorder in another one, the compartments of the model will not fit what is seen in a clinical setting. Rather in the real-world conditions resemble sand dunes, rounded peaks flowing into surrounding soft peaks, and no particular place where it could be said one dune stops and another starts.
Furthermore the confluence of a surprisingly huge array of factors is involved in the appearance or origin of behavior we'd consider problematic (if we could even agree on that).
I'd concur with the idea that computational models need to be consistent with and informed by the reality they are intended to represent, but experience suggests that the more fundamental issue is conceptualization of behavioral reality itself which is far more complex than our usual abstraction of it lead us to consider.