To encrypt a message, one uses the algorithm for doing a CBC decryption. To decrypt a message, one uses the algorithm for doing a CBC encryption. Does this work?
Reversing encryption and decryption would give you a very weak encryption scheme.
First, it would make the initialization vector effectively useless. An attacker could simply unmask the first encrypted block and subsequent blocks do not depend on the initialization vector.
Second, guessing a single block correctly (e.g. assuming it contains only space characters used for indentation) would allow an attacker to unmask the following block. If the attacker were able to do this repeatedly, she would immediately see if blocks of plaintext repeat.
Being able to see repeating patterns will enable an attacker to perform statistical analyses and can quickly lead to a complete loss of confidentiality.