Descartes conducted thought experiments that thoroughly questioned things in order to establish the scientific method. Does what I am seeing really exist? Perhaps they do not exist when I am not looking at them. Then, what is the most certain thing? He left the famous phrase, “I am because I think”(Cogito ergo sum).
In the East, Buddha realized 2,600 years ago that this world is like an illusion. Although Buddha’s awareness has nothing to do with scientific approaches, many people are now pointing out similarities with modern physics. Matter is made of atoms, and atoms are made of electrons and protons, and surprisingly, there is nothing in the space between the electrons and protons. There are topics such as the uncertainty principle, the double slit experiment, and superstring theory, but I will not go into them here. By the way, according to the Buddha, even the ego is an illusion.
At least, the world as we perceive it exists as we perceive it because we exist. If we do not perceive this world, this world does not exist as we perceive it. The colors and textures of materials that we perceive are just our perceptions, and they would not exist in that state without us. Without us, this world, this universe, would have even no “meaning”. Classification of things also belongs to our perception, and classification itself is not a real thing.
The origin of what we now call “data science” can be traced back to “pattern recognition” research in the 1960s and 1970s. At that time, automatic ZIP code readers were developed. Later, over time, it was also called real-world computing, soft computing, and data mining, leading to today’s data science. Although the different names may have slightly different purposes, they all have the same thing: to build mathematical models for classification and prediction from data.
Satoshi Watanabe is a pioneer in pattern recognition research; he published his “The Ugly Duckling Theorem” in the 1970s. See Wikipedia for more information. In a Japanese-language booklet entitled “Recognition and Patterns,” he explains that classification does not exist objectively, and that in order for classification to be valid, we must accept that some predicates (properties) are more important than others. As for cluster analysis, he explains that it should be used to discover useful new taxonomies.
When we prepare data for cluster analysis, the set of attributes (variables) selected implies the direction from which we wish to interpret the data. It is a mistake to assume that preparing as many of every available attribute as possible will yield objectively correct clustering. It only yields meaningless results that are difficult to interpret. In other words, clustering requires implicit prior knowledge.