The Fox, the Hedgehog, and Data Science

In the well-known fable, a pack of wolves attack a fox and a hedgehog who were faring in the woods. The fox evades them by employing all kinds of maneuvers until they eventually give up, while the hedgehog just retreats inside its spiky armor protecting itself from the wolves. The morale is that there are different ways of tackling a problem, both of which are valid. Could this ancient fable be applicable in data science today?

In our case the wolves clearly symbolize the various challenges that make themselves apparent in a data science project. These can range from problems in the data (missing values, outliers, redundant features, etc.) to database issues, such as the data being too large, too varied, too unstructured, etc., to anything else related to the data science pipeline (e.g. feature engineering). The fact that there is no fool-proof way of dealing with them makes the whole situation interesting and in need of some expertise. In general, there are two strategies for dealing with these issues: the hedgehog approach and the way of the fox.

The hedgehog strategy is the most commonly used since we live in a world that values this attitude above anything else. From the factory-like development of information workers, to the factory-like pipelines in most businesses, and the not-so-imaginative approach to tackling social problems, it is no wonder that most people lean towards this approach in data science. So, if a hedgehog-like data scientist tackles an issue in his data, there are a couple of things he tries, before giving up: going back to the theory, and seeking a solution on a search engine (usually one of the mainstream ones). And although that’s perfectly fine for the majority of cases, it doesn’t really add anything to the field or to his abilities. It is the safe option, the one yielding the smallest risk. Interestingly, this same mentality is in abundance in the academic world too, which is why revolutionary technologies are hard to come by from the university labs (though there are exceptions like MIT, Georgia Tech, and other exceptional scientific institutions).

Exceptional places usually employ the fox strategy which at the very least makes the whole process intriguing and worthy of books and films (it is no coincidence that fox-like technologists, like Steve Jobs, Elon Musk, and many others, have earned the status of celebrities and left their mark in pop culture). Unlike the hedgehog-, the fox-like data scientist seeks and finds creative ways to deal with the problems he encounters. If there is no known way to deal with an issue, he often invents one, or combines existing methods in a new way, developing a new, oftentimes unique, solution. The main characteristics of this approach are out-of-the-box thinking, risk-taking, and acknowledging that there is no right answer (resulting to a focus on the essential, the function rather than the form). Unfortunately, this is an attitude that’s quite rare nowadays in the data analytics field, although it is what actually brought about data science as an independent field.

Although as the fable suggests, neither one of the two approaches is better than the other, it is clear that they are not both equally useful for a given challenge. For creativity-related work (e.g. research, development of a new product, tackling a novel challenge, etc.), a hedgehog approach would be a disaster, while the fox-like approach may not work so well for unimaginative tasks (e.g. accounting, mechanical engineering, etc.). Being a rather diverse kind of work, data science lends itself more to the fox approach, though you can easily find in it elements of both. However, the way it is taught today is quite limited, making it challenging to cultivate a holistic approach to it (though you will learn the hedgehog aspect of it quite well). For this reason, in this blog we try to encourage the development of a fox-like approach, to complement the hedgehog one, and allow you to cultivate a more holistic and perhaps enjoyable view of this fascinating field.