What two forms do unconscious biases in data typically take?

Unconscious biases in data typically manifest in two primary forms: human biases and data collection biases.

Human biases stem from the inherent prejudices or perspectives that individuals may have, which can influence how data is gathered, labeled, or interpreted. For instance, a data scientist may unintentionally prioritize certain data points over others based on their assumptions or beliefs, leading to a skewed representation of reality.

Data collection biases arise during the process of gathering data. This could include factors such as the demographic or geographic limitations of the sample population, which may not accurately reflect the broader context. For example, if data is primarily collected from a specific region or demographic, it can lead to biased conclusions that are not generalizable to the entire population.

Understanding these forms of bias is crucial for data practitioners as they can significantly impact the performance of machine learning models and their fairness. By recognizing and addressing these biases, practitioners can work towards creating more equitable and robust models.