There are enormous and ever-growing data sets about medicine, finance, engineering and many other aspects of modern life. These data sets can be mined for insights that have the potential to radically change and improve the world.
In rough terms, there are three ways in which we can extract insights from Big Data:
- FULLY AUTOMATIC COMPUTER SYSTEM
A deep learning neural network (or other learning construct) is built from Big Data, and makes decisions autonomously (without human involvement). Examples include credit card fraud detection and book recommendation.
- HUMAN ANALYST EXPLORES DATA AND PROPOSES INSIGHTS
A human manually explores the data and extracts insights from it, typically over a period of days or weeks. The human uses software tools for visualization and statistical analyses, but there is no artificial intelligence or automated learning. Examples include an expert engineer exploring engineering data in order to understand failure modes of a piece of equipment.
- HUMAN DECISION-MAKER WORKS WITH INTELLIGENT COMPUTER SYSTEM
A human makes the decision, but he/she is given recommendations, analyses and/or explanations by a software system which is built from Big Data. This scenario is especially common in real-time or near real-time decision making, when the human does not have the time to manually explore the data set. A classic example of this is medical decision support, where a learning system based on Big Data can make recommendations, but a human doctor must make (and be responsible for) the final decision.
NLG is not needed in case (1) and its utility has not been demonstrated in case (2). But in case (3), where a human and a machine have to work together, NLG can make a huge difference by allowing the machine to communicate insights, explanations, caveats, assumptions, background information and so forth using natural language: this is the most effective way to communicate this kind of information to a person.