Why biology and medicine have never been as precise I’d like: the Machine Learning and Big Data explanation
Over the past few months, I have been on a journey dedicated to understanding more about AI and machine learning. Today, as I was reading something about algorithms and various approaches to AI, I had an interesting idea/realization.
After a recent doctor’s appointment, I had been reflecting on the imprecise nature of biology as a science. A competent doctor often replies with, “I don’t know.” Most doctors are essentially trying to guess a specific outcome that correlates with the symptoms they’re seeing. The impreciseness of it all made me quite uncomfortable. Biology didn’t seem that different from the social sciences where, rather than systematic logical rules, there are just common observations accompanied by tons of exceptions/assumptions. And it seems like medicine, in particular, is filled with confounding variables.
I’ve always believed that, if you truly understand something, you can express it through a rigorous algorithm. In other words, clearly explain it through a set of rules, inputs, outcomes, and constraints. That’s probably what drew me to study physics during my undergrad. In physics, we clearly define the different rules that explain our world. Physics is beautiful in its simplicity once you understand it.
I also believed in the converse hypothesis — anything that can be truly understood can be expressed through a rigorous algorithm. What does this tell me about social sciences as areas of study? I’m not sure. Perhaps they are just inherently unexplainable. Perhaps they don’t have any universal rules, only micro-explanations. All I know is that this made me less motivated to study them.
Back to my idea, or realization, from today. It occurred to me that in physics, we are able to fit the data about the world around us into general-purpose rules, because we are able to compute the data and, more importantly, understand the computations. What if our brains are just not able to compute the data available in biology? We know that machine learning algorithms are finding new protein folding structures en-masse, at a rate that would take thousands of humans millions of hours. Can they create algorithms to explain biological phenomena, algorithms that are as rigorous as the laws of physics? Can they do the same for social sciences? Has this always just been a ‘big-data’ problem, i.e. the data is too vast for our human brains to recognize the patterns? But a deep neural network, with hundreds of layers and millions of nodes, is now able to compute it? Perhaps evolution could be considered one of the most complex ‘learning algorithms’ we know of, if not the most.
My realization centers around the ‘rigorousness’ of various fields of science. I have always loved physics as a rigorous science that explains everything we see around us. It occurred to me that many other fields may be just as rigorous, but our brains are just not capable of the level of computation required. It’s worth noting here that rigorousness or computational complexity do not imply conceptual difficulty, but rather the ability to standardize and express.
— —
My journey to understand more about AI has led to an exploration of the nature of intelligence, how ‘thinking’ works, the similarities between humans and machines in how they reason about things, what we can learn from both sides, what outcomes one is capable of that the other is not, and what this tells us about the future directions in which AI can develop. Understanding these is a tall order, and my exploration of them is just a personal hobby given my other commitments — albeit a hobby that I’ve become extremely passionate about. In the near future, I will publish some write-ups that delve into what I’ve learned and understood about how intelligence works, both in humans and in machines.