The Problem with Data Science Competition Platforms like Kaggle

As any data scientist is probably familiar, there exists a vast world of predictive modelling competitions on the internet; the most well-known competitions likely being those from Kaggle. Some of these competitions are incentivized financially, others just by curiosity. The basic structure for nearly all of the financially incentivized competitions is that all participants (or teams of participants) train and develop models on a training dataset and send in their submissions as predictions on a test dataset of which the labels are hidden from the competitors. Subsequently, the'best' models are awarded their respective chunk of the financial reward based on some predetermined accuracy metric. Most opinions regarding these predictive modelling competitions are that competition based on financial incentive creates demand for brilliant people to discover solutions to hard problems.