What you can learn from machine learning about democracy, institutional education and media
This post is part of my 200 words per day challenge that I am sharing publicly on Twitter in order to improve my writing and develop a writing routine. Feel free to join and comment.
Two years ago I decided to spend a year as an Erasmus student at the department of Computer and System science of Stockholm University, located in Kista. There, I choose to focus my courses on machine learning.
I remember a professor telling us about this algorithm called Random Forest, and of the intriguing teaching it can give us about democracy and the education system.
When you want to teach something to a machine, you have to give it knowledge, in the form of data, and a brain to process it, in the form of a learning model.
There are many models. Each with pros and cons. One of such model is called a decision tree.
A decision tree uses characteristics of the data (called attributes) fed to it in order to take a learning decision. For example, such trees can learn to distinguish several objects in a picture — and thus develop some sort of computer vision. Needless to say, this model has lots of practical applications.
However, depending on the data and the attributes used in the model, decision trees can not always be right when they take decisions. Just like humans.
To decrease this problem, researchers found that using several decision trees instead of only one helped obtaining better results: each decision tree formulates its own decision, and this decision is taken into account by the end result, just like in a democratic process where each individual votes.
Obtaining better results mean getting closer to the truth, the state of reality.
Now, one important condition for this “vote” to be closer from the truth is that the trees need to make their decisions by themselves, without being influenced by the other trees. Such trees are said to be “independent”.
Also, if trees base their decisions on the same attributes, the probability of truth resulting from the vote tend to decrease, and thus we have to randomize the attributes used in the process.
So, what is the lesson here ?
For democracy to be effective, it should rely on a diverse population (the different randomized attributes) whose individual members are not educated by one standardized model but instead urged to bring out their own uniqueness through a dedicated and highly specialized education process. The same should be applied to media, as a common source of information would uniformize the thought process.
This way, the vote can reflect truth and the res publica, the common good.
It is beautiful to see how algorithms can teach you how the world works.
Liked this short post ? Follow my adventures in real time on Twitter.