Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Dealing with imperfections in validation data

Potsdam Institute for Climate Impact Research

Introduction

The standard way of assessing the accuracy of AI outputs is to compare them with “ground truth” data that is produced by humans. However, we know that humans make mistakes when they are producing this data. As well as clear mistakes, there are also areas of judgement and interpretation where disagreements are not necessarily wrong, but simply other valid and justifiable representations of the data.

If we want to evaluate well, this raises a few issues and questions.