In computer sciences, we usually use human as a role model to develop machine learning algorithms and concepts. Interestingly, our efforts to develop machine learning led to better self-understanding. Here, as a data scientist, I am going to use machine learning concepts to explains some of the human social behaviors.
In life, you have a limited number of observations. The observations could be scientific, social or any other observations. For the sake of this article, let’s focus on social observations. Your social observations are training data for your brain. Your future observations are your test data. What about validation data? Apparently, as a human, we don’t have access to this type of data and basically, we use test data as our validation data too.
Your experience in life, or simply your age, could be interpreted as your training epochs. Time goes, you collect data and your brain starts fitting a model on those observations.
Your initializer function is probably your family, friends, and environment. When you born, your gene was your only initializer. But, as soon as you born, your environment, family, and friends start shaping your mind and beliefs.
Your education, knowledge, and judgment is your optimizer. If you know better about social sciences, you can find better models. If you are kind and passionate about human, you try to fit a better model. If you have a bad temper, you probably try to fit a cruel model to your social observations.
Time goes and you see the society and people. You start fitting simple models to your observations. If you have limited social interactions, your models remain simple since they explain your limited observations well enough. At this stage, you form some stereotypes in your mind that might explain some behaviors well enough. If you get obsessed by your new social model and only look for more data to confirm it, you can always find those data. Simply, your brain starts ignoring observations that are not aligned with your initial social model. If you stay in a same social environment for a long time, basically, your training and test data are coming from the same dataset and your model becomes a more local model than a global model. People who travel and go outside of their origin society usually find more contradicting observations (new test data from the new dataset) and starts to develop better global models.
In another word, staying ina same social environments tend to make your model over-fitted. In the absence of different test data, your model tends to become such an over-fitted model that cannot be updated via new test data or even good optimizers.
Here, I tried to simply explain our social models using some machine learning concepts. The best way to avoid developing over-fitted social models in our minds is trying to interact with social environments outside of our comfort zone.
Source: Deep Learning on Medium