How to make AI less racist
Source: Walter Scheirer
Affiliation: Bulletin of the Atomic Scientists
This article discusses how negative biases, for example racial biases, can be baked into artificial intelligence algorithms through the use of “big data” to train the algorithms. These “big data” datasets can have biases in the way the data is labeled or what data is included in the dataset. While some AI researchers have responded to these biases by removing datasets entirely, the author proposes to systematically remove biases from datasets as they are discovered.