How Bidirectional Neural Networks could be used to predict future data while they need future data?!

[Amirreza Heidari]

In the following link I see that Bidirectional LSTM is used to predict the future data in time series. I know that a Bidirectional recurrent neural network use both past and future data, and therefore for predicting future data we need future data1! Can anyone explain me how they work when in a time series we are going to predict the future? 

https://machinelearningmastery.com/how-to-develop-lstm-models-for-time-series-forecasting/