Quantifying Human Bias and Corrections in Data Analytics- Lessons from the Recent Elections
The rage in the last few months has been around big data, small data, cognitive thinking, unsupervised learning, dark matter, artificial intelligence, % of humans being replaced by robots, etc. etc. The argument goes that the more data you have, the better off you are (sure, you cant get started without data); analysis alone (without unbiased human intelligence and input) is the silver bullet that we have all been waiting for.
Considering that the election results are very fresh, lets apply data and analysis logic and see how this could have parallels in our businesses:
1. Impact and quantification of the unknown- Everyone who has been following the election campaign for the last 18 months agrees that this was an unusual election in more ways than one. There were multiple new variables that were introduced that the analyzers (on all sides) didn’t understand, or know how to quantify and bake into their analysis. So they continued to apply the same logic that was tried and tested in 2008 and 2012. Their failure was not that of big data analysis or the prediction engine, but rather the flawed human logic that was fed into the model. The machines did what they always do, namely spew out results based on the logic.
Businesses face the same challenge. Think about a new competitor (Uber, Airbnb) who is not following a traditional path, the introduction of a new product or service that has never been tested before (iPad), entry into a developing nation with huge potential (India, Brazil), evaluating political and business risk of a nation (Russia), etc.
2. Time and value of “corrections”: Analyzers on both sides did a decent job in attaching a value or “correction (positive or negative)” every time there was a significant event, i.e. release of access Hollywood tapes or reopening the email issue. The flaw was the value associated with the event and human biases that contributed to the value. While it’s easy to comment about this without being actually involved in the process at the time of attaching a value, it goes to show the long road ahead of us for perfecting this model.
The same logic can be perfectly applied in a mergers and acquisition situation. If only I had a penny for every time a CEO announces that this is the merger of equals (1+1=11), the deal of the century, or decade if the CEO is somewhat shy!,( anyone remember AOL and Time Warner, Autonomy and HP?) then I would be very rich today. Leaders and businesses always associate a correction factor based on their inclination and bias. We are yet to reach a stage where the machines are capable of automatically attaching a value for the corrections.
3. Motivations vary and are difficult(yet) to precisely predict: Pretty much the entire media and analysts were telling us that non-college educated white voters were going to be a big support for the now President-elect, everyone was surprised to see that 28% of non-college educated white women voted for Mr. Trump, precisely the target audience that was assumed to be at risk after the surface of the Access Hollywood tapes.
The same analogy can be applied to motivation for why someone buys or does not buy your product or service. The motivations are a combinations of various factors. Some can be understood, quantified, verified and baked into your statistical modeling, while others will catch you off-guard.
Big data analytics and prediction is a work in progress. Excited as we are with the progress being made each day, we still have a long way to go to get to a stage where human bias can be precisely quantified.
PS: These opinions are not meant to be a reflection of my political leanings or commentary about the recent elections. It’s just the humble, personal curiosity of an engineer who likes data analysis and prediction. Thank you in advance for holding off all your political comments.
Ashok Muttin is the founder and CEO of SupplyCopia, a marketplace for the healthcare supply chain. He is committed to re-inventing the healthcare supply chain using big data analytics, predictive intelligence, cloud and mobile technologies delivered as a SaaS.