Explain why unsupervised learning of p(x) would not be beneficial in this scenario, and propose an alternative approach for improving pedestrian prediction accuracy.

Unsupervised learning of p(x) is not helpful because it focuses on reconstructing the input data without providing any information about the output predictions. Instead, consider using weakly-supervised techniques that leverage pre-existing labels to train a model that can accurately predict pedestrian behavior.