that's not true at all. We're trying to measure the human impact on global temperature. The natural zero is whatever our best guess for the temperature would be without human intervention.
If it was constant, I'd agree, but the natural temperature was always changing and whatever it would have been today will also be changing. So you'd be graphing a different quantity, f(t) = temperature(t) - natural_temperature(t). From such a graph, you wouldn't be able to tell if a change was a change in the actual temperature or just a change in the predicted natural temperature. A predicted natural cooling would look the same as a human-caused warming.