II.2.3 Obtaining “Ground Truth” for Human Characteristics
To predict “soft” human characteristics such as emotions, morals, ethics, tribal affiliations, personality, risk taking attitude, performance, stress, and burnout, “ground truth” needs to be established. This is the “true” value as personally assessed by a human being, not the AI machine learning system. Mathematically speaking, it is the “outcome” or “dependent” variable which will be predicted by the machine learning system. For example, if the machine learning system is predicting customer satisfaction based on the speed of answering e-mails of the customer, ground truth is the customer satisfaction directly reported by the customer. When measuring emotions through facial expression, ground truth is the emotion felt and reported by the user.
For personality characteristics, OCEAN (openness, conscientiousness, extroversion, agreeability, neuroticism) ground truth can e.g., be measured with the Big Five FFI survey. Ground truth ethics and morals can be measured with the Schwartz and Moral values survey, attitudes to risk with the DOSPERT survey (see section I.5). Ground truth for stress, burnout, and emotions such as joy, happiness, sadness, etc. can e.g., be measured through experience-based sampling, by asking an individual to wear a smartwatch and querying the individual to enter these variables at random times by vibrating the smartwatch. Sometimes, ground truth can also be established through unsupervised learning, for example by training tribefinder with the twitter streams of all members of Congress to define a tribal embedding that contains the language of politicians. For “honest signals” computed from e-mail analysis of the employees of a company, ground truth can be taken from business metrics such as the sales numbers of the company, or satisfaction surveys of the employees or the customers of the company. The later chapters in this part of the book will take you step-by-step through the process of how to set up such ground truth collection systems.