When you accept an unknown certificate, you're communicating with whoever is on the other side regardless of their identity.
You have to settle the identity of the other side at least once when you stablish communication[1]. Actually, you're not settling the identity, just a public key with which you encrypt the communication (which in turn is settling the identity as a desirable side-effect). You still need their public key because, well, that's how encryption works: you send a message encrypted with a certain public key, which can only be decrypted with the corresponding private key.
Fortunately, web browsers alert you not to do this because it's pointless. You see both encryption and authentication as a single step because web browsers abstract the process for you. Otherwise, why am I encrypting the connection in the first place? If someone's listening and you're not checking certificates, sending plaintext or encrypted is actually the same thing and offer ZERO encryption. I'd be able to read anything you say without effort.
[1] http://en.wikipedia.org/wiki/Diffie%E2%80%93Hellman_key_exch...
The analogy for the first case is more like "The postal service guarantees that it won't be opened before whoever is reading your mail opens it", which is clearly a pretty meaningless guarantee.
Another difference: The Internet as a whole routes packets mostly concerned with reliability, not trustworthiness of the ISPs in the middle, so a trusted Postal Service doesn't work on the Internet either.
An analogy might be: 1) I seal a letter in an envelope, and the postal service guarantees that it will arrive at the address I specify, regardless of the human that opens the letter at that address; and 2) I seal a letter in an envelope, and the postal service guarantees that not only will the letter arrive at the address I specify, but that it will only be opened by the human I specify, who is identified by his postal-service-issued ID number.