Decoding human emotions
- 1. Decoding human emotions By Sayanti Bhattacharya January 2016 All that is unexplained will remain unexplained. That is, until science explains it! With science slowly unraveling all mysteries and translating them into machines, somewhere deep within our darkest human sentiments, a serious question emerges. What will differentiate man from machine if all mysteries that are “human” are solved by science? Science’s journey into unexplained so far… The explanatory power of science gained pinnacle with the Theory of Evolution. But that was just the beginning of science’s journey to dig deeper into the unexplained. The reach and pace of science in this aspect has been phenomenal, validated by scientific breakthroughs in 21st century like cloning, mapping of human genome, stem cell therapy, robotic limbs, to name a few. There were 2 key factors that fueled science’s journey into the unexplained – Data and technology. With time, the avenues of generating, collecting, collating and processing data have increased manifold. Earlier the scope of data was limited to traditional sources but with the advent of social and digital media there is a huge volume of unstructured data, which when merged with traditional data, can give us deeper insights. With data becoming “Big”, it is not a surprise that relevant technology was conceived to manage this exponentially expanding data. With data and technology in place, science today has no
- 2. barriers to exploring all mysterious aspects that have fascinated us till date. This has become the universe of research for data scientists. One of the most fascinating topics that has attracted data scientists till date, is understanding what is human. This is not limited to behavior, but nudges at a deeper understanding of what makes the very core of being human – its irrationality, its emotions, intuitions, perceptions, experiences, etc. Now that is an interesting puzzle to crack! Breaking down human elements… In this blog, I would like to attempt to break down human elements into a mechanical version and see how close science can come to replicating human responses. To start, let us try and read 2 simple sentence – “I cnduo't bvleiee taht I culod aulaclty uesdtannrd waht I was rdnaieg”. “Cna Yuo Raed Thsi?” Indeed, it is gibberish, yet you could read this. How do you think you did that? Guess work? Let us remove the human element of “guessing” from the equation and see if we can still explain how our brain does this. What you will notice is that in the first sentence, the order of the letters in the word is incorrect, except for the first and last letter. Thus, as long as the first and last letters are in the right place, even if the rest are scrambled, you can still read it without a problem. In the second sentence, even the last letter in short words is incorrect, yet, you have read it correctly. This is because the human mind does not read every letter by itself, but the word as a whole. And shorter the word, the “typos” are easier to correct. How does it do that? To answer this, you would need to imagine the brain functioning like a super-computer. Under this context, all that the brain is doing is running iterations using an algorithm that systematically moves intermediate letters one at a time until it identifies what an actual, relevant and logical word could be. Thus, to read “Cna”, it needs 1 iteration (assuming first letter is correct and the current spelling is incorrect). To read “Raed”, it needs 2!-1 iteration (assuming first and last letters are correct and the current spelling is incorrect) 3!-1 iterations (assuming first letter is correct and the current spelling is incorrect) Thus, the scientific explanation for the human element of guessing could be purely a technical element of running quick algorithms and concluding on an outcome based on systematic iterations. Let us delve further with some more instances where “gut-feeling” and “intuition” come in. As a modeler, one often faces a situation when he/she retains a statistically insignificant variable in the
- 3. model with a hope that it will turn significant in subsequent iterations. And in multiple instances this actually happens. If asked, how he/she knew, the person would simply reply that it was just a “gut- feeling” that this might happen. This subjectivity, intuition, experience are what make modeling an art vs. science. But what if we remove these human elements from this scenario? Can our brain actually predict the outcome of a variable? What if the brain functioned like a repository and querying machine. Imagine the brain using its repository of modeling experience, querying previous cases and their outcomes, juxtaposing the correlation matrix and observed trends, hypotheses along with the level of statistical insignificance to create a probability of retention of the variable. Higher the probability, stronger this “gut-feeling”! Not convinced? Let us try and break down “first impressions”. You meet a stranger at a party, strike up a conversation and within seconds have formed an impression about the person. Research says it takes anywhere between 7 to 30 seconds to form a first impression. How are we able to do that, so quickly and quite accurately? How do you know he is chivalrous, party-lover, practical, friendly..? The easiest answers are “guessed that”, “felt-so”… If we take these unexplained human elements of “feelings” off the table, would we have a scientific explanation for the phenomenon of first impression? Let us think back and retrace what this stranger said and did during those crucial first 30 seconds – he perhaps shook hands and waited for you to choose your seat or be seated before he did so, and maybe spoke about how he ended up at this party with an invitation from the party he attended yesterday; he perhaps also mentioned how he took the cab to avoid parking hassles and did you notice that he waved hello to at least 3 more people while chatting with you…? What if our brain has built a dictionary that classifies and maps words, gestures, actions, behavior, etc. to certain pre-defined buckets or “characteristics”? Chivalrous = f(Shaking hands, waiting for the lady to sit..) Party-lover = f(party>1 per week) Practical = f(cab>car; context = parking) Friendly= f(acquaintances > 1; time < 30 sec) Thus, first impression could be explained as predicting personality using extraction of key words or actions and mapping of elements to pre-defined classes. In this context, the brain is quickly processing verbal and behavioral information and classifying them to create a person’s aura and defining your first impression of him. Now extend this explanation and you will see that the classes and the matches will determine if you “like”, “love” or “are attracted” to that person. Replicating this thought process across other human emotions would possibly help us break down all human emotions into technical and logical components. Does this mean that in near future, a super-
- 4. computer would be able to feel, act, react and emote like us? Or at least perhaps predict human elements and responses? Man vs. machine… For any predictive algorithm, there are 2 factors that determine its success – speed and accuracy. Let us see how these algorithmic responses fare against human responses on these 2 factors. At present, computers have definitely succeeded the human mind in calculations that involve huge amounts of data, however in cases where a logic might come to human brain at a lightning speed (read intuition), a computer still lags because it needs to run iterations until the correct logic is identified. Technology has already progressed by leaps and bounds to overcome hurdles of speed and scale to accommodate “Big” data and decisions being needed in “Real-time”. Thus, we may be able to envision tomorrow’s technology comfortably challenging the speed of a human mind in decision-taking or responding. Now comes the interesting aspect of accuracy! One characteristic of being human is its irrationality. Even after incorporating some allowable aspects of standard deviation, the accuracy element of algorithmic responses maybe up for a toss. There are likely to be many deviations in human responses and not all can be considered as outliers, as they might actually be much better outcomes. What gives this extraordinary accuracy to human responses, which lies beyond pure logic? The answer lies in the contextual elements that support the logical ones. A super-computer will be able to assimilate and assess only the information that is fed into it through a source. We, as humans, have the ability to draw context from multiple sources. The most critical sources are our very strong 5 touch-points which we use to gather information. Yes, the human sensory system – its 5 senses of sight, hearing, touch, taste and smell. Top this up with subjective elements of culture, morality, ethics and upbringing. This added context makes human emotions and responses much stronger and more accurate than the purely logic- driven algorithmic responses. Conclusion To conclude, the challenge ahead for science is to give technology a context as strong as the human sensory system and rules as strong as human ethics. The day that happens, science would have perhaps solved the puzzle called “human” and it would be another break-through day for science in its journey to explaining the unexplained!