Search :

Predicting The Past and Facing up to the Hard Problem of Data Analytics

Everyone seems obsessed with predictive analytics these days. What will happen next, what will happen in 3 months....

I'd like to turn your attention to something we do everyday, something we trust as absolute fact, something we trust our absolute lives to. That something is what I call 'Predictive Historical Analytics' – the science of making up the past to fit the present so you can hopefully predict the future.


We all know exactly what happened in the past, right?

Are you sure? I mean, would you stake your life on it? The general focus of predictive analytics is trying to work out what will happen in some future configuration of a particular system of interest. Take for example the classic problem of 'the weather'. Most experts will say that weather prediction accuracy falls drastically as a factor of time. That the prediction becomes more of a guess the further in the future it is made. We are good at short term, but not excellent at long term.

But what about something easy.. what about 5 minutes ago? Surely you know what the weather was 5 minutes ago right? I mean you just experienced it! Lets take it a little deeper and say you were in your shower 5 minutes ago, and have just stepped outside to mixed cloudy day. Can you tell me what the weather was 5 minutes ago based on the weather now? If you go back a little further, say 3 hours, what does the state of the weather now tell you about the state of the weather then?

Take another example. You are asked to describe the actions of 2 people with whom you share a house for the last 30 days. You may say “On Tuesday, Bob came into the room and went to the Kitchen”. OK, at what time? Which way did he walk? Did he go left round the coffee table or right? The truth is in most cases you don't know anything more than the basic facts and even then they may be half truths. All you know for certain is that somehow he got to the Kitchen. So, are these the memories that you trust to make life or death choices in every day life? Choices like, is there a bus to my left or did I imagine it?!


The implication for AI and predictive analytics development

So what does that mean for predictive analytics? Well, if we take the popular view that we have a good chance of predicting the future based on the present and the past it would lead one to conclude we don't stand much of a chance because the past is at best a hodgepodge and at worst utter fiction penned by your own internal version of predictive analytics.

The study of predicting history is as, if not more important as that of predictive analytics because without a true accurate picture of the past, the prediction of future is a mystic art


Website powered by Firecart X eCommerce