Why does the specification of the batch_size parameter in a keras input object not replace the ?/None entry?

Question

I I want to compute a value (similarity/distance) between all my encoded vectors during the loss function but this is not possible because I can not access the batch dimension.