The use of a mouse with the replacement of one of its genes with a human version is to better mimic the human disease phenotype, or to prove something about the biology of the human phenotype with research that can't be conducted in humans.
Genes and their corresponding proteins or RNA usually differ somewhat between mouse and humans. If you wanted to study whether a certain SNP in a gene (found in humans) was the reason for part of your phenotype, you may want to use transgenic mice to test hypotheses with different exacerbations (treatments). One way to do this is to create a mouse with the human gene replacing the mouse gene, and another mouse with the human gene containing the trait-linked SNP.
If you tried to do this with the mouse gene, you couldn't be sure that the same SNP would have the same functional consequences. If you didn't remove the endogenous gene from the mouse before replacing it with the human version, you would never be sure if observed experimental phenotypes weren't confounded by both copies of the gene and their corresponding gene products being active in the same animal.
In your example, with SOD1 and ALS, creating a humanized mouse with human SOD1 allows researchers to more fully study the same protein product made in humans, while giving the mouse treatments that would be unethical in humans.
Your understanding in your last part is incorrect. Mice don't lose genes after a few generations.