Tuesday, 5 July 2016

Life, Perception and Exponential Functions

Motivation to work is important. As we move forward, the sources from which we derive motivation changes. In childhood, to do well in a set of exams is all that's required and motivations were simple - the rod in early childhood, some delight in learning and competition in later stages. However, in college, there's usually no fixed goal to pursue. One tries a bunch of things, realizes the sub-field closest to one's calling and proceeds. Even so, in undergraduate college, things like maintaining a good GPA provide some form of a concrete goal.

But the cocoon is broken as one leaves for the next venture, be it a job in a company, a startup or graduate studies. One enters a stage where the goals to aspire for and the means to achieve these are no longer well-defined. Here we typically have to work with the perceived rewards in mind. And so I make the argument that our ability to perceive these rewards is detrimental to success. It is here that I believe some vestiges from evolution lie in our way.

Isn't it surprising that sounds are reported to us as Decibels instead of in a linear scale? It turns out that we, in fact, perceive sounds in a logarithmic and not linear scale. This phenomenon is not restricted to just sounds. It carries forward to all forms of sensory perception, memory, and our spatial sense. Young children, when made to place numbers from 1 to 10 on a scale, place 3 at right about the middle. Since log 3 ~ (log 10)/2, when seen from the logarithmic light, it's no longer as surprising.

In a recent work, Lav R. Varshney and John Z. Sun, using some ideas from information theory, attempt to explain this. Their work shows that logarithm as the quantization function for sensory inputs produces the least distortion if we assume sensory inputs to follow a power law distribution. In practice, many natural inputs follow power laws such as frequencies of the English language. In plain language, logarithm allows us to best store and process information. To put it less technically but at all loss of rigor, our ancestors benefitted more from knowing if they're facing one or two lions as opposed to knowing whether there are 99 or 100 lions.

Now when it comes to long-term rewards, I claim that we perceive them to be at best linear in time. A simple thought experiment proves this. Suppose we were to pose one the question, "What technological advances do you predict by the end of 2100?". The intuitive way of answering this would be to provide a figure by accounting the progress that's happened in the previous 100 years. However, a more critical analysis reveals that since innovations happen on top of existing innovations, it's more likely an exponential curve. Indeed, history suggests the same. Take a man from 1000 B.C to 500 A.D. and he might see what he expects. But on taking him from 500 A.D. to 2000 A.D., he's bound to be held speechless for a few days at the least and permanently at the most.

Also, why else do we express such surprise in rags to riches stories or glance at that super successful multi-billionaire start-up kid with smoldering jealousy? The famous 10,000 hours rule, that one can become a world-class expert in a reasonable topic of choice, by putting in 10,000 hours of effort might, in fact, have a lot of truth to it. Progress always happens in baby steps. In steps of 1,2,4,8,16,32 and before you know it, the emperor runs out of money.

In a similar sense, seemingly meaningless investments suddenly start to give results. In a venture like a PhD, where one's success depends very much on the extent of knowledge that one accumulates, this becomes crucial. Since knowledge is built upon previous knowledge, it's reasonable to assume that the utilities of studying a certain topic also grows exponentially.

After reasonable digression, getting back to the topic at hand - motivation, one strong motivation for writing this post is that it, in some ways allows me (and the readers) to realize the value in seemingly inconsequential knowledge quests. What we have is an exponential reward function f(t), which we see as a log-linear g(t). Exponentially large rewards remain within our reach if we only alter our perception!


  1. Awesome! This is a very thoughtful blog and with a deep meaning.