E-catalog        

Библиогр.: 41 назв.

There are two principal conceptions of rationality: one that bounds it with logic, and the other that associates it with efficiency. The first one has language for its model with its systematicity and regularity. Within this conception, to be rational is to follow rules of inference. The other conception is rather modeled by mathematics, as it presupposes that to be rational, in general, is to get more at a lesser cost, to which end all the data should be represented quantitatively. The history of cognitive science may well be seen as a gradual transition from logic and language as the science’s basis to parallel processing and statistic computations. The basic concepts of cognitive science in its classical era were “representation” and “computation (processing)”, i.e., mind was attributed the architecture of the von Neumann machine: processor, memory, input/output, etc. In connectionism, on the contrary, computation is focused on the most effective adaptation of the network to changing conditions. At the same time, even if individual chains of computations are slower here than in a serial architecture machine, the network as a whole benefits due to the ability to carry them out not only simultaneously and in parallel, but also interdependently – when the sequence takes into account not only the result of the previous step, but also the results of parallel processes. What the predictive processing theory has in common with cognitive symbolism is that it also relies on computations and representations, although not in all its variations. However, like connectionism and neural network vision in general, representations are considered not as symbolic, but as sub-symbolic, expressed by certain probability distributions, and, accordingly, computations are understood as probabilistic (Bayesian) inference. But where predictive processing differs from both competing computationalist paradigms is the understanding of representation as prediction: a cognitive system driven by multilevel attractors generates hypotheses about the causal structure of the environment, allowing for the prediction of incoming perceptual data. These considerations are enough for us to rationally conclude that rationality is rather capability to surf the intractable world with just an updatable engine for predictions and a feedback circuit than commitment to any imposed rules. Philosophers – and not only them – should consider the question of whether some strict and reliable logic of abduction is possible, or – the more probable option as it seems – probabilistic mathematics will remain the only assistant in explaining and creating intelligent systems.