The only way to defeat their argument is to imply some sort of mechanism that assesses whether a "proposed" mutation will add or subtract information, and forbids those that will add, but this idea seems stupid.
I don't think this is quite the case. The attitude that seems to come up time and again is a belief that there are specific original 'created' sequences which represent the maximal informational content regardless of functionality or any other concern.
Any deviation from these sequences
must be a reduction in information, it is possible that a reversion would add back the lost information but you can never have any new information according to this view because no change, even one creating a new specific function, is producing novel information because information is only being measured in how closely a sequence adheres to a created ideal.
I came up against this arguing about antibiotic resistance with Smooth Operator, he claimed that even if a mutation had no phenotypic effect other than conferring resistance it still constituted a loss of genetic infromation because the protein conformation had changed.
Just for my 0.02 on the OP, as has been pointed out, there is no rational informational metric by which Georgia Purdom's claim could even remotely be considered true.
TTFN,
WK