Base on the costs of transferring data into AWS being much cheaper, I wonder if a reverse transfer protocol could be designed with an adaptation of binary searc

Base on the costs of transferring data into AWS being much cheaper, I wonder if a reverse transfer protocol could be designed with an adaptation of binary search. You divide your data in blocks of, let's say 100 Mb, the receiver creates a block of random data and sends it to a process running on AWS. The process has to answer, higher, lower or exact (and then both parties move to the next block). Probably it will be very slow, but it will lower a lot the bandwidth usage for egression.