it’s too difficult to fix these biases, arguing that modern-day AI models are so complex that even their creators don’t understand why they make certain decisions. Deep-learning models, like neural networks, are made up of millions or billions of parameters, also known as “weights,” that act as adjusters in complex mathematical functions between connected layers. Think of the layers of a neural network as being a bit like a factory with an assembly line, where each person on the line has a certain job like painting a toy car or adding the wheels. By the end of the line, you have a toy car.
...more
