As a researcher, one learns to value data and not go by face value. A small dataset may give great results. However, it may not be easy to reproduce those results on slightly larger dataset, say 1.5 times the size. And if the dataset is huge, maybe 100 times the original one, why, the original results would be quite pointless. How then can we generalise characteristics of people based on religion, region, caste or gender? In our limited lives, a human would only come across a limited set of people, a very small subset of the populace. And even within that, there would be contraints. In data, especially image processing and vision dataset, a solution is expected to work for specific costraints. Yet in real life, we expect an entire community to behaive similarly. We judge individuals based on our experience with other individuals of similar community be it region, religion, caste or some other social category.
In research, if a solution works for lab conditions, it may not work at a real field. If it works in a field in India, it may not work at Austria. It is not expected to. A daytime solution is not expected to work at nighttime. Yet when it comes to individuals, we generalise. We forget the contraints that people live under. What is easy for one individual may be a struggle for another. What is a negative in one may work as a positive for another living in different conditions.
The poor one shouldn't be the assumed thief. A given community may be not full of misers. A given religion may not be all about violence. Women need not be the weaker sex. The man need not always be a hero or a villain. The illiterate need not be less intelligent. The hungry beggar isn't a nuisance. The quite one need not always be proud. The jolly one can also sometimes be serious. And finally, the small children are not as unknowing as they may seem.
There is no common man. There are only special individuals, each uniquely created with a very precious life of their own. Each has an entire, little world of their own. Humans are not data points to draw patterns and make conclusions.
2 comments:
Wonderful perspective and analogy with our own dataset world :)
Thanks Piyush ☺
Post a Comment