When is an algorithm not an algorithm?

The only time the news media use the term “algorithm,” it seems, is for computational methods that aren’t.

Merriam-Webster defines it as “a procedure for solving a mathematical problem (as of finding the greatest common divisor) in a finite number of steps that frequently involves repetition of an operation.” Let’s forget about repetition; almost every computational procedure uses loops. The key word is “mathematical.”

An algorithm produces results that can be mathematically verified. An algorithm for calculating pi will produce the known value to the needed level of precision, or it’s wrong. A search algorithm is an algorithm when its results correspond to precise matching criteria.

Google started the misuse of the word. If its search “algorithm” is based on mathematical criteria, they’re beyond anyone’s understanding. A document can match without having all the search terms. My last name can match an entirely different last name, one with an “r” in it. The same search, run from two different computers, won’t generally produce exactly the same results. YouTube videos rank suspiciously high compared to competitors’ videos.

The notion that Google’s search is algorithmic is ludicrous.

The misuse of the term has spread. I’ve seen it often applied to machine learning methods, which by design are the opposite of algorithmic. If you have a mathematical formula for your result, you don’t need machine learning! OK, there are a few problems where scattershot guessing and comparison are a plausible way to zero in on a value, and that’s “machine learning” in a very broad sense, but it’s not what people are generally talking about.

It’s entirely OK to use non-algorithmic methods with problems that don’t have an exact mathematical description. Just stop pretending they’re algorithms.

Comments are closed.