Summary

In this chapter, we examined many of the ethical and privacy concerns for data scientists. These range from accidentally leaking personal information about an individual to who takes the blame when a ML system leads to catastrophic consequences. We started by looking at how ML algorithms can exhibit bias, such as gender and racial bias. Some of the examples we saw were how facial recognition software used by police can cause or augment racial biases in policing, and how voice recognition software usually doesn't work as well for people who weren't in the training dataset. We learned some ways to combat this bias, including sampling techniques like SMOTE, generating synthetic data with GANs, and simply collecting more data.

Next, we saw how privacy can be breached in clever ways by combining so-called anonymized datasets with public data such as census data. We also saw that many laws have been enacted around the globe to protect privacy, and how we should...