The Intersection of Human Life & Machines
Frankly, human lives and machines mirror each other. In my pursuit to study more about machine learning, I have seen overlaps between the works of a machine’s routine and the humdrum of one’s ritualistic movements on earth. The habit to constantly reflect on life has opened up parallels within the two experiences. As I share this observation, do pardon me for any misused terms. (still an amateur!)
Artificial vs Human Intelligence
Machine learning is a sub-category of artificial intelligence. In the development of AI, we are heavily dependent on data. Meanwhile, the development of human intelligence, we rely mostly on knowledge and experiences. Here are two concepts we often experience around us.
Iterative Process
To put things in perspective, many things that we do in life tend to be repetitive to varying degrees. How do we commit to our jobs? How do we move on from one relationship to the next? How do we manage new interactions with people we have just met?
We train machines using “training sets”, as humans practice using past year papers before major exams. It is the epitome of “practice makes perfect”, in both ways, a routine is followed to get things done. When we think of every life experience as we do with data sets, perhaps we can move on from one experience to the next more objectively. Through conscious reflections, we also learn about how we can improve in our next predictions and performance. I die a little when I hear people share about their “regrets”. Granted we don’t always get second chances, but when we do, we should see those past ourselves to become more successful at our next trial.
We have to appreciate the beauty of training sets. Nearly everything we do is an iterative process, we repeat, reflect, and seek improvements.
Overfitting & Underfitting
Overfit occurs when the machine models the training set too well and is unable to cope when new data sets are introduced.
Going back to “practice makes perfect”. Perfection is akin to an asymptote, we aim to move towards it, but it is not achievable. A model that reportedly shows 100% accuracy is certainly a red flag. Likewise, perfection is non-existent in real life.
In society, there are norms and expectations. The academia & the real world does not necessarily add up. Take a perfect straight ‘A’s student, this could be the perfect employee who is extremely intelligent or one who have simply perfected the training sets but is unable to excel when the parameters differ in the real world.
The main reason for overfitting is the inability to decipher between signal and noise. To tell the two apart is important for both machine and life. Different factors may affect our experiences and learning but one should be given more weight, while the latter should simply be ignored. Remember the times when we get too caught up in one event and obsessed too much with the details when they are not transferrable takeaways?
The quick advice? Take a step back to decipher between signal and noise before allocating too much or too little weight for each situation.
Moving forward,
Run time: as long as you need. Even the best computers take time to run the test sets; it will take even more trials for it to become intelligent.
In that light, let's agree to give ourselves time to do the necessary practices to become better and successful in our own domains.
While running tests, we can also turn to mentors and communities in order to tap on other’s wealth of knowledge and experiences.
More water in the harbour floats all boats!
