Machine Learning, Cultural Bias and Rethinking your Own Beliefs

I cannot image a person in his/her right mind devote all of his/her life to a very pivotal idea (as one’s own belief/disbelief in God and/or religion) without getting it thoroughly validated in the first place!

Briefly, what machine learning is

Machine learning is one of the disciplines of computer science. One of its ultimate goals is to achieve human general intelligence. While this seems to be a far-fetched goal given the current state of machine learning, scientists are working on it, bit by bit.

The main pillar of machine learning is to learn from data. The simple idea is as follows: given enough amount of data, the machine (i.e. the computer) can learn the general patterns and rules governing these data. Later on, the machine can use these learned rules to do inference or prediction.

Bias problem in machine learning

Because machine learning methods depend heavily on data to learn these rules, one of the inherent – and fundamental – problems is data bias. The data the machine is trained on shape its perception about the world. In other words, the data become the prior knowledge or the prior experience, based on which, the machine infers the future.

data-bias
Figure 1: A dart with the target as the red dot in the middle and a few training examples as the blue dots. On the left, good training examples (unbiased), while on the right bad training examples (biased). Reproduced from Perceptual Enigma

Please, let me explain the bias problem in this example. Imagine we are training a robot how to dart. So, we provide it with some examples, from which, it will hopefully learn how to aim correctly to the target of the dart. As shown in figure 1, the target is the red dot in the middle, while the examples are in blue. On the left, we see good examples, as they align very well with the real target, i.e. unbiased. Unfortunately, on the right are bad examples, as they clearly deviate from the actual target, i.e. biased.

As we all expect, the robot trained on the examples on the left will be – with a high probability – a better dart player than the one trained on those on the right, right? So, we conclude that:

The data dictate the behavior of the machine it was trained on

What cultural bias is in principle

We, human beings, are social creatures. We live in societies and interact with one another. Based on this interaction, we develop a variety of habits, attitudes, laws, unwritten rules, beliefs and knowledge, all of which can be wrapped up under the title: Culture. Different societies develop their own distinctive cultures over long periods of time. This results in different interpretations of the same phenomenon.

nodding
Nodding your head right and left means no) while nodding it up and down means yes. Reproduced from StackExchange

Let’s take for example a well-known gesture: head nodding. In the majority of countries, nodding the head up and down means yes while nodding it right and left means no, right? Except this is the opposite in Bulgaria. Now, think about it for a moment and try to imagine yourself nodding left and right as a sign of acceptance! This looks counter-intuitive, right? So, we call such differences in interpreting the same behaviour or the same situation as cultural bias.

There are 2 problems with culture bias: it happens almost subconsciously and you think it is right. You don’t question whether is it easier to stand in a queue or to bypass it when getting on a bus. You don’t think about it, you just do it. To be fair, this is favourable so as to save your brain power for other things that matter more.

You cultural experience dicatates your beliefs and hence your behaviours

Data bias in analogy to cultural bias

The same idea of data bias in machine learning is analogous to the idea of cultural bias, except in the former we know the truth – the target of the dart – while in the later we don’t. This is the very pitfall of cultural bias: you always think you’re right, you never think you’re wrong, this is the norm you’ve been trained on for all of your life.

To be fair, let’s state it again: cultural bias is not necessarily wrong, it is usually helpful. It is only harmful when we inherit some bad cultures. But how to judge and correct our culture then? Let’s take the example of the machine first.

In order to judge a machine, we experiment it on never-seen-before data (called test data). We expect it to behave correctly. But when it doesn’t, we take further steps to analyse and conclude whether the error is in the test data or in the belief of the machine itself. It the case of the latter, i.e. the machine learned bad behaviours, we can fix this by training it again on more good data. This is the case for the machine and data bias, but what about us, humans and cultural bias?

In analogy to machines, we have to frequently subject ourselves to cultural tests, to have the chance of judging if our cultural bias, hence cultural beliefs, is correct or not. Travelling and interacting with foreigns from different cultures is always a good way of testing your own beliefs. If you cannot afford it, there is a magical and dirt-cheap alternative: reading.

The cultural bias in me and probably in you

I was hapless enough to live the first quarter of a century of my life in a monotonic, homogeneous, uni-directed, almost uni-ethnic and fairly ignorant society. Yet, I was lucky enough to alter this by resettling in a country with a drastically different culture. On top of it, I got the chance to travel to quite a few other countries, meet and live with people with different backgrounds, ethnicities and genders.

Having subjected to this striking difference in cultural experience, I undoubtfully began to re-think quite a big deal of my own beliefs. What if I was like a dart machine trained on a very bad training samples? What if I was subconsciously too biased towards my culture to criticize it? Doubt went like a storm in my brain uprooting some of my fundamental beliefs. Personally speaking, the culture I grew up in is deeply oriented about Islam. Consequently, I began to question Islam as an idea and as a religion. Soon after, I begin to question the very existence of ALLAH (GOD). At that time, I was completely puzzled and wasted.

Later on, my little vulnerable brain was hammered ruthlessly with lots of questions: how do I know that God exists while I’ve never read any book before about atheism nor agnosticism? How come I believe in Islam as a religion despite that I’ve never been exposed to any other religion? How, why and what? These questions were haunting my dreams.

Finding my way out of this dilemma was not easy but completely worth it. I leave the details of finding GOD to another detailed post, but the moral of this post is:

I cannot image a person in his/her right mind devote all of his/her life to a very pivotal idea (as one’s own belief/disbelief in God and/or religion) without getting it thoroughly validated in the first place!

Advertisements

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s