Imagine your data points look like a U. There's no (lineral) correlation between x and y, you are equally likely to have a high value of y when x is high or low

Imagine your data points look like a U. There's no (lineral) correlation between x and y, you are equally likely to have a high value of y when x is high or low. But low values of y are associated with medium values of x, and a high value of y means x will be very high or very low.