Why Machine Learning Fascinates Me
Regardless of whether you are philosopher or a scientist or just a random person hoping to make sense of this mess called life, there will always be some questions that bug you. Unfortunately, my mind has been home to some rather controversial questions, and ideas, which if i were to openly proclaim might put me on verge of being crucified, both by liberals and conservatives, and more dangerously be declared blasphemous, and i love my dear life too much for that. But i have always believed that answer to all my questions lied in successfully debugging the machine that generates it: human brain. You might think that with all the technological advancements of the late, this has already been done probably, but no, our knowledge of working of brain, at large, is at best a theory, and at worst, pure guess work.
Despite my fascination with human brain and its marvels, i abhorred Biology, and cramming, too much to take it as a major. So, naturally, i ended up taking Math and eventually majoring in Electrical Engineering. My fascination with human brain, though not ceased, mellowed down and as i begin to care less about larger questions surrounding life, my ambition to attain answers for them became asymptotically zero.
Engineers, and computer scientists, also have a fascination with human brain, not to find answers to some weird philosophical why am i here type of questions, but just because human brain is a highly efficient machine (at least with respect to our current measures of efficiency; more on this in a later blog) and is able to perform a lot of difficult tasks (like walking, speaking multiple languages) with ridiculous ease. This fascination has resulted in the creation of the field called Artificial Intelligence whose ultimate goal is to create a (super) human level intelligent system. Even if you are not a CS or EE major, but avidly follow the news, you might have heard about this already, all those ‘concerns’ by likes of Elon Musk and Stephen Hawking regarding creation of Skynet and a Facebook bot going rogue (which was fake news by the way, welcome to the age of ‘information’). Up until 5–7 years ago, despite the catchy name, Artificial Intelligence was deceptively little intelligent when compared to humans (the gold standard of intelligence in AI; a hypothesis that i have a few problems with but lets not get into that here). Sure it was able to win a chess game around the turn of century but deep blue was not really intelligent: it was simply an elaborated search method to generate the best move. And on close inspection, Artificial Intelligence seemed little about simulating human brain and more about refined mathematical models with solid reasoning about their working.
The only AI technique that does indeed aim to simulate human brain; Artificial Neural Networks (ANN), though in existence since 1958 only really became popular post 2010. Its popularity was driven by huge increase in performance on non trivial tasks such as object detection in an image and language translation (aka google translate). A large share of the hype surrounding AI in media is in fact due to Deep Learning (DL); the name given to the field of study of ANN’s. And it was this hype that got me!
If there is one guy anyone and everyone knows in ML community, that is Andrew Ng, main instructor of Intro to ML mooc on coursera. It was this course that helped me get started in ML. My feeling during the first few days of this course was of bewilderment; here i was, a mathematical illiterate (had taken neither Linear Algebra nor Probability course at that time) programming intelligent machines (though not really intelligent, when i come to think of them now). But this bewilderment only existed for a short while and was ultimately replaced with fascination as i dwelt deeper into the theory of learning and learnt of the idea that fundamentally our brains might just be highly efficient computational machines being able to learn from data gathered by our experiences. This might not seem that deep and revolutionary, but give it a moment! If true, this means that by using laws of mathematics we can ultimately debug the human brain. And that if we are able to boost AI to level of human intelligence, we not only have a highly intelligent AI, we are also able to understand the working of our brain, and ultimately our own selves.
