On the Need to Understand Human Behavior to Do Analytics of Behavior

Artificial Intelligence and data science are rapidly gaining importance as parts of decision support systems. As these systems improve, it becomes necessary to clarify humans’ roles in the decision-making processes. Humans may not be able to improve on the choices a good algorithm makes, they may not be able to adjust the parameters of the algorithm correctly, and their role in processes that use good algorithms may be limited. However, this does not mean human involvement in data-supported decision processes is unnecessary. A closer look at the analytical process reveals that each step entails human decisions, beginning with the data preparation through the choice of algorithms, the iterative analyses, and the display and interpretation of results. These decisions may affect the following steps in the process and may alter the resulting conclusions. Furthermore, the data for the analyses often result from recordings of human actions that do not necessarily reflect the actual recorded events. Data for certain events may often not be recorded, requiring a “big-data analysis of non-existing data.” Thus, adequate use of data-based decisions requires modeling relevant human behavior to understand the decision domains and available data to prevent possible systematic biases in the resulting decisions.