Machine learning methods are a general class of techniques that are quickly increasing in popularity. Unlike their standard parametric (e.g., logistic) counterparts, the validity of machine learning methods does not depend on precise knowledge of the true underlying models that generated the data under study. As such, they are often argued to be superior than routinely used approaches for prediction and causal effect estimation. In this talk, I will provide examples demonstrating that this superiority does not apply generally. Using examples from perinatal epidemiology and simulation studies, I will highlight little recognized issues that should be considered when using machine learning methods for prediction and causal effect estimation. I will show the importance of causal considerations when prediction is primarily of interest, and demonstrate how machine learning methods can be considerably biased when used to estimate causal effects. The objective of this talk will be to provide strategies to deal with potential problems that may arise when using machine learning for prediction and effect estimation.