I would argue that in the driving example, there is in fact correlation, but the data set (a single drive) is insufficient to detect it. If you analyzed the data from a thousand different drives over the same road, the correlation between speed and gas pedal pressure would indeed become apparent. Perhaps the book's authors might better have said that if causation exists, there must exist a set of measurements that can reveal the corresponding correlation.
Or are there cases where causation exists, but no amount of data will reveal correlation?
Or are there cases where causation exists, but no amount of data will reveal correlation?