[1/N] Many AI people today claim that their systems are cognitively inspired (in particular inspired by the S1/S2 distinction introduced by @kahneman_daniel in his dual-process theory) just because their decision mechanisms couple fast routines and slow decision-making strategies

This is a clear example (one of the many) of the misconceptions that have been raised by the shallow ascription of labels coming from the cognitive vocabulary to the behavior of such systems. Unfortunately, it is not sufficient to just implement “fast” and “slow” mechanisms in

an artificial system to claim any kind of cognitive inspiration or of cognitive plausibility. To make one of these claims, in fact, one should build and integrate algorithms in a way that is much more constrained with respect to such a generic and shallow description of how an

intelligent system (natural or artificial) works. For example, one should consider "how” such fast or slow mechanisms are built, how they interact between them (both within the System 1/System 2 components and between them), how they evolve over time (e.g. System 2 mechanisms can

be “automatized” and become System 1 routines) etc (note: "Thinking, Fast and Slow" was written for a popular audience and therefore contains obvious oversimplifications of the dual-process theory of reasoning. Unfortunately, many people in AI have considered the book as a

scientific publication ignoring the actual scientific papers laying down the theory). In Cognitive Design for Artificial Minds ( http://amzn.com/1138207950 ) the distinction between “shallow” and “constrained” systems is made clear by introducing the “functional” and “structural"

design approaches and by exploring the different explanatory roles that such design perspectives put in place. /end