Understanding error types through outcomes

AI teams use a framework called the confusion matrix to categorize different types of correct and incorrect predictions. This helps weigh the real-world impact of errors. Consider a hypothetical running app, RUN:

• A true positive happens when RUN suggests a trail you love and choose to run. The AI correctly predicted what you'd want.
• A true negative occurs when RUN avoids suggesting steep trails after you've indicated you dislike inclines. It correctly identified what to exclude.
• False positives frustrate users with irrelevant suggestions. RUN might recommend a mountain trail to someone who only runs on flat paths. The AI wrongly thought this would appeal to you.
• False negatives represent missed opportunities. RUN might skip suggesting a perfect waterfront trail because it misunderstood your preferences.

Not all errors carry equal weight. A false positive in RUN wastes a few seconds. A false negative in a medical AI could miss critical symptoms. A false positive in allergy detection means avoiding safe foods. A false negative could trigger dangerous reactions. Teams must consider these outcome types when optimizing their systems. The same technical error creates vastly different human consequences.