Data science has been amazingly successful in exploring statistical correlations and dependences and predicting probabilities of future events. Getting causal information from data would be the next qualitative step.
This is not only because 'understanding how things work' is a central motivation of human curiosity. After all, estimating the impact of future actions before doing them - which is a causal question by definition - is required as basis for reasonable decisions. More explicitly, one wants to predict how changing a variable X would influence some target variable Y, e.g. the profit. This is challenging for two reasons:
1) The observed statistical dependences between X and Y may not be due to the influence of X on Y. Sometimes Y may also influence X, or, more often, there may be a common cause of both.
2) Often it is known that the dependences between X and Y have two reasons: X influences Y, but there are also common causes. Some of them may be known, some not. Then, computing the influence of X on Y requires the removal of the effect of the common causes with sophisticated techniques.
It has often been argued that causal conclusions can never be drawn from statistical observations alone. More recent results suggest, however, that under appropriate assumptions the joint probability distribution does contain a lot of causal information that researchers are just beginning to understand.
If you want to know more about the potential of modern causal data analysis, visit our course at the Blue Yonder Academy. This course offers an opportunity to become up to date with what is currently already possible and how to derive concrete action strategies.