Many people (including scientists involved with the field of evolution) aggressively reject the notion of devolution. They are somewhat justified. Any change in the genetic makeup of a population will occur because whatever changes which occur will be (by definition) those which increase the reproduction rate.
Consider Case 1:
A computer program is given a sequence of letters. It makes two copies of it, then it gets those copies, and makes two copies each of those, then it makes two copies each of those, and so on. Each time it makes a copy there is a small chance of an error occurring in the copy (a few of the letters might be different). Hence over time many of the strings will be different to the original.
This seems like a good model for evolution, but it is actually unrealistic. In the real world an exponential expansion stops somewhere. An infinite number of copies of something cannot exist. Something limits the number of copies (the number of atoms in the universe provides an upper limit). Populations in the real world stabilize. The species fills its niche, and the population in each generation is pretty much the same as it was in the previous one (population cycles droughts etc excepted).
Consider Case 2:
The program above starts with some number of identical sequences of letters. In every generation, half of these are chosen at random and for each of these, two copies are made. Again, there is a chance of errors being introduced at each copy.
This is more realistic, and devolution in this sense seems like the rate at which the overall population changes.
However it is still not a realistic model of evolution. In real living things, the survivors (those who replicate) are not chosen at random. They are chosen according to some criteria.
Consider Case 3:
The program above starts with some number of identical sequences of letters. In every generation, some of these are chosen according to some criteria and for each of these, some copies are made. Again, there is a chance of errors being introduced at each copy. (Assume that the number of copies made for each one is chosen so the population stays roughly constant between generations).Hence the ones which replicate are generally the ones which match the criteria more closely - the criteria is called the fitness function. Hence the fittest tend to survive, and the population evolves towards the greater fitness.
The fitness function in the Case 2 does not not have to be anything sensible. It might be the number of letter 'x's or how many instances of 'GATTACA' are in the sequence of letters.
In the real world, the DNA sequence in living things affects the characteristics of the animal, and hence defines the ability of the organism to compete and replicate.
Hence, the argument goes, any changes in the DNA makeup of the amortized population must be evolution (towards greater fitness) instead of devolution (towards less fitness).
Someone can impose a perverse criteria on the surviving population (like sterilizing all babies who are not deaf), but all they are doing is changing the fitness function - the population will evolve towards more fitness, in this case congenital deafness. It is still evolution, not devolution
The more pragmatic take the position that devolution is still a useful concept, because it is a way of measuring the error rate (mutation rate), but it is a term to be used cautiously.