Thinking about it more, a ray travels from the camera and through a pixel and out the other side until it hits an objects surface. If fixed amount of 1 is used then it returns 1 sample back to the pixel that contains information to colour that pixel. But if adaptive is used at the point it hits the surface of the object it can then decide how many samples it needs to return to the pixel in order to render it correctly. You can control the number of samples available for it to decide from through thresholds and subdivision values.