Predicting the Unprogrammed

Humans are unlike machines. We do not generate data but rather, behave. Ceteris paribus, a person may behave quite differently at two different times. So, what makes us different from machines?

Concept of time

The most important variable that differentiates us from a machine is the awareness of time. For a machine, time is just a number with constant linear increments having no context and no deeper meaning attached to it. Whereas, for a person time is not just indicative of measure of duration but it also reflects his position in his limited life span. Also, Humans perceive time as nonlinear in certain situations and stages of life. A machine functions in the same way since its birth till its death without differentiating its age. But humans are different. Similar situations in life have different meanings and different contexts at different stages of life. Identifying all the intervening variables, hence, becomes a challenge to answer the question – why someone behaves in a particular manner in a particular situation?

Risk

Risk is not an absolute measure but rather a relative one. Its perception varies from person to person.

Risk propensity of a person is the degree to which an entity is willing to take chances with respect to risk of loss. To understand or to measure an entity’s behaviour with inherent risk, we need to first understand the factors that lead to such behaviour.

There are 3 kinds of people –

1. The Irrational – No logic or reason exists for the actions. The thrill or adrenaline rush from the chance encounters are the basic motivating factors.
2. The rational – The action is calculated, reasonable, logical and depends on the expected return of an opportunity.
3. The unaware -Actions may be risky but the entity does not know it.

Even before we can delve into the deeper intricacies of each type, identifying and segregating such types is a herculean task itself.

Machines don’t lie

The corner stone for any predictive algorithm is the authenticity of data. If the data gets corrupted at any of the stages of collection, transport, or storage then no amount of polishing and/or manipulation of data would bring it back to its original, unadulterated form. If the seed itself is faulty, then you cannot expect the tree to bear flowers for you.

Case in point, the recent US presidential elections, where most of the predictions were wrong because people lied about their voting pattern to the data collectors. If you do one thing and say something else, then the prediction is not based on your action but rather on what you say.

Text vs Context

Humans have the uncanny ability to extrapolate, be intuitive and read between the lines and behave according. Human behaviour is not a result of only the facts of a given situation i.e. text but rather it is a result of assumptions and deductions, envisioning and anticipating, social norms and pressures i.e. context. Our actions knowingly or unknowingly also take into account our past experiences, stereotypes, biases and prejudices as well. Hence, measurement of human behaviour keeping in mind “Text” and “Context” would make it more useful, applicable, and reliable.

While every person is unique, there do exist some similarities in the patterns of behaviour in certain sections of the population. It is one thing to identify these patterns of behaviours and quite a different thing to predict their behaviour. Putting it differently, just because you know the “climate” of a region, it doesn’t mean that you can predict its “weather”.