I believe they did change. Natural selection works, however mutations cannot ever produce "new" information, they simply remove or change the existing genetic code. More often than not, such mutations are not favorable and the majority are harmful (the protein stops working negatively affecting function) or harmless (they don't affect bodily function to the point it provides a survival advantage)
I would specifically address this point now as it has been made by several people. A gene consists of thousands of triplets that code for specific amino acids. A protein is a very very long chain of thousands and thousands of amino acids (like a short story made from words). So if a mutation changes one (or many) of these triplets, the amino acid sequence that is built the protein will be altered and hence the protein will change, altering what it does. This is how novelty is generated.
Now, you would think that if you change word and sentences in a short story blindly, the story will get incoherent. In biology talk, the protein will loose functionality.
However, in nature, this is not the case, and this is where the analogy between human codes and biological code breaks down. Human codes (programs, language etc.) are brittle, change something randomly and you will make it incoherent 100% of the time. But since nature's codes (in DNA for example) are not simply random symbols and sounds that are artificially "glued" with meaning, but are rather themselves products of regularities and laws of chemistry/physics in actual chemical networks, all nature's codes are
resilient and malleable. Almost all mutations have no impact on functionality at all, and the probability that a random mutation will cause novel or improved functionality is small but significant, and only somewhat smaller than the probability that the mutation is harmful. This is not some theoretical analysis, this is the result of exhaustive experiments where millions of DNA strands and proteins and genes were experimentally tested by altering them randomly and testing functionality. And these lab results have been complemented by even more extensive simulations of how alterations would affect gene, proteins or regulatory networks. (I will present you with sources if you want to read about it
http://www.ieu.uzh.ch/wagner/research.html).
One of the remarkable findings of the experiment is this. Take any protein from any living organism at random and any other protein from any other organism at random,
and there is over 90% probability that you will find a path of moving one step at a time from protein A to protein B by changing/adding/subtracting an amino acid while always retaining functionality. Its like having the ability of changing mahabharata into ramayana one word at a time while it being always the case that all intermediate stages make meaningful stories of their own! Such a powerful discovery is strong evidence for the inference common ancestry of at least majority of organisms and refutation of the idea that organisms have existed as independent creations from the beginning. No matter how different they look, all these organisms are different renditions of the same original song. And this is just one piece of evidence by which one can infer common ancestry through processes that conform with the pramana theory.
Isn't the diversification of One to Many one of the consistent themes of the Upanisads?
In the beginning, son, this world was simply what is existent- one only, without a second.... (Chandayoga 6.2:1)
And it thought to itself: "Let me become many, let me propagate myself."....(Chandayoga 6.2:3)
Here is also the fact that Upanisadic writers thought that the mind is material like all other,
"When one eats food it breaks down into three parts. The densest become feces, the medium becomes flesh, and the lightest becomes mind." ..(Chandayoga 6.5:1)