Autoencoder - Why use autoencoder instead of PCA?

When looking for a linear and intelligible method to reduce dimensionality, principal component analysis (PCA) is a solid solution. Autoencoders are the best option when it's critical to capture complex and non-linear correlations in the data.