There are many interesting approaches to training and testing AI systems. Also interesting ways to get into trouble by accident.

Be careful what datasets you use for training! According to this article by the Guardian, much of the Scots language Wikipedia is fake.

The armies of ghost workers who help train AI algorithms.

The SUNCG dataset, built by Facebook and Princeton for computer vision research, was built using data allegedly stolen from a home design company.

AI systems imitate biases found in their training data.

Using mice to train a neural net

Carbon footprint of training a neural net model: technical paper, news article

Google vs. medical researchers about whether claims about AI performance can be verified/replicated

A project to physically model networks of neurons.