My definition of abstraction

One thing human are particularly pride of is our ability to use abstraction to solve problems. We in software development all know that “another level of abstraction can solve my problem”. But what do you call abstraction ?

Abstract is the opposite of concrete. Concrete things are simply things, the kind of things a dog can see, bite and eat: me, the bone on the floor, the bee it’s trying to catch on the fly… In my job, concrete things are the formula to compute the price of a fixed income bond, the interest rate risk in a portfolio or the report the manager wants the tool to generate.

When I first work on a thing, this thing is concrete to me. The first bond I am computing the price for is just a simple concrete bond. But the second, different bond is a revelation: while it is different to the first bond, it also has lots in common ! wow ! I can notice it, and especially I’d like to save work from this, as I am a lazy developer. So I gonna use abstraction…

Concretely, I would define abstraction as the mental action of finding the commonalities and the differences between two or more things. The common part of every thing is what I will call the abstract thing, and I will then call each thing with its differences a concrete thing.

So my (very practical) definition of abstraction has to be explicit to be of value to me, so that I can reuse the common part of things. The most obvious example of abstraction is the very notion of class against instances of a class, aka objects: the class represents the common part (the structure) of a group of objects, while each object has its own values that make it different to its mates. What is common between every bond I will be given is that they all share a group of parameters and calculation conventions, thus I will create my pricing software for this abstract bond in order to be able to price every bond by just describing its differences against the common bond.

Young children early develop their abstract skills by playing with things; They soon see the difference between one thing and two things as the notion of numbers, the difference between a red shape and a green same shape as the notion of colour.

As said before, abstraction is very much about classification: given the red square and the green square, we can see they both have a square shape, so we can classify them in the shape = square category. Given another green circle, we can also classify the green square and the green circle in the colour = green category. This way, we have identified two abstract notions, shape and colour, from very concrete things.

As I am -more and more- a lazy developer, my next question is: can my computer abstract things for me ? After all abstraction helps in handling what’s common in things in order to handle this common part only once, and I have identified the very notion of abstraction itself as an abstract notion, maybe in order to reuse it ? (by the way what I am doing at this point in this writing also follows the abstract notion of recursion, talk about abstraction, then talk about the abstraction of abstraction). Some clues: in common spreadsheets, when we enter a number sequence, say 1, 2, 3, the spreadsheet recognizes the arithmetic sequence so that it can guess its next values. I’m tempted to claim it is an abstraction ability already… Not sure in fact. Obviously, it is in AI field (artificial intelligence) that neuronal networks have natural classification abilities, so the answer is yes, software can abstract concepts.

Initially published on Jroller on Monday April 11, 2005


Software development, Domain-Driven Design, patterns and agile principles enthusiast