Natural systemof classification differs from artificial system in
Answers
Artificial classification systems - as they’re most commonly deployed - attempt to mimic natural classification systems. They take as input the same stimuli to which a human would be exposed, and they’re calibrated against the classes that a human would identify, given those stimuli (that is, they’re constantly tooled and retooled until they achieve the same ends as a human, with reasonable accuracy).
This is not to say that the intermediate classification steps are the same, however. Indeed, in many cases they may be quite different (or at least quite different than the ways in which we think natural classification systems work). In other words, machine learning algorithms are concerned only with mimicking human ends - not human means. For example, if a human were to look at a series of pictures of domestic animals, and were to classify them as dogs, cats, or fish, he or she might utilize characteristics like size, shape, and color (we actually don’t know that this is how people do this, but this is just a hypothetical). These might well work, but they might also be just a subset of a larger set of basic visual characteristics the distribution of which differ depending on the type of animal featured. A machine learning algorithm could achieve the same accuracy in distinguishing between categories of pet, but it might use some other, not (entirely) overlapping set of distinguishing characteristics - perhaps background objects, texture/contrast, or luminosity (to the extent that different types of animals are likely to be photographed in different types of locations, which have distinguishing features of their own).
The categories into which items are sorted, therefore, is unlikely to differ, for most ML applications into which you’d run today, and this is by design - the reason that we have developed many of these AI approaches is so that we can replace the human element of [some process], and these algorithms are therefore designed to mimic what that human would produce (such that the process is otherwise unperturbed).
It is worth noting, however, that there is nothing “natural” per se about natural classification processes, in the sense that natural tends to imply “inherent” / “ideal” / “most fundamental” / or otherwise “special.” There is nothing special about the categories that humans identify over a set of stimuli - in most cases we have identified only one of a number of different approaches to breaking up our world into smaller chunks, such that we can better understand it. In most cases, we have found one approach that appears to work well, but that does not mean it is the only or necessarily the best approach. Cross-cultural differences are a great example of this; one need only look at different languages, and the things to which they do and do not assign labels, to see this (German contains many examples with which people are famili