Direct Link to File. 238 words, 1 minute read, 0 paperback pages

The main conceit of this book is that we can meaningfully talk about “understanding how humans do* classification” and how these ways may be similar to or different from the ways “[humans have programmed] machines [to] do classification.” The first phrase is problematic because not everyone may do things the same ways, and because we don’t fully understand the inner workings of cognitive processes and their neural underpinnings. The machine side of the main conceit is problematic for similar reasons: there are a host of classification algorithms available which may yield similar results but fail in different ways or offer features that others don’t, and for extremely complex (billion-parameter) machine learning systems, the issue of “interpretability” of their decisions is an ongoing area of research. Furthermore, as it is an active field, we must be careful to say that machines do or do not do certain things because someone may invent a system in a year or two that explodes such final pronouncements.

So for this book, we will have to speak in generalities, giving the reader a taste of common aspects of classification. We will not be able to hit every thinker or every algorithm, but will highlight some of the most important ones affecting the world today.

*I’m using “do” instead of “perform” because it seems more “colloquial”. Any opinion on that?

References