Humans are useful for training future iterations of the closed source thinking model. Once the model is sophisticated enough to train itself humans will no longer be required.