Although global and local alignment are mechanistically similar, they have very different properties. Consider the alignment between the genomic sequence of two eukaryotic genes from distantly related organisms. You'd expect the exons to remain the same because their coding sequences are evolutionarily constrained, but the introns may no longer be recognizably similar, especially if they have acquired many insertions or deletions. The problem is that exons may account for only 1 to 2 percent of the sequence. As a result, a global alignment between these sequences is an alignment of mostly random letters. In such a scenario, it's very likely (especially if introns change size, as they often do) that the exons will not align to one another because their score contribution is very small compared to the rest of the sequence. In contrast, local alignment can pick out conserved exons, but unfortunately, just the maximum scoring one. The shortcomings of the standard alignment algorithms have been addressed by numerous variants.