This is a completely inaccurate statement. The lift produce by both wings remains unchanged regardless of the roll angle. It is defined in terms of relative airflow over the wings, being always perpendicular to the former. Lift is NOT (usefully) defined relative to terra firma, a common misconception.
So in the relative frame of the aircraft itself, the lift vector is unchanged in terms of the machine itself. Rolling the aircraft rotates the lift vector, causing a larger "sideways" component and reducing the "vertical" component. For example, at 45 degrees bank the load factor is, in fact, 1.41, meaning that the wings ought to produce 1.41 times the lift to maintain altitude.
So the real reason why a dihedral tends to "auto-correct" the bank angle is that if you rotate the airplane into a bank without adding power then you really are not going to maintain your altitude against gravitational acceleration and are going to descend through the airmass. Now, given that your airplane is now at a 45 degrees angle relative to the source of gravitational attraction, the aircraft is going to see (from it's vantage point) some"sideways" airflow over the wings. Imagine an airflow component originating from the tip of the wing, towards the fuselage.
What now happens is that the "falling" wing is going to work much more efficiently because a number of reasons, causing the aircraft to experience an aerodynamically stabilizing force.
The easiest way to see is that if you imagine a V-shape object moving sideways (let's imagine from right to left), the \ part of the V is going to deflect airflow and produce some "sideways lift" while the / part will not exhibit such a phenomena.
But in terms of the 172, it's high-wing configuration actually does not require dihedral because the actual fuselage of the aircraft "shadows" the high wing, making in naturally less efficient at producing lift, thereby again correcting the bank angle.
Airplane designers attempt to balance this out so that the final design not only is statically stable (correcting in the right direction) but also dynamically stable (the oscillations should be self-limiting).
