Neural networks for control: counter arguments