If you start observing an encrypted channel in the middle, it pretty much looks like just a bunch of bytes. Symmetric encryption algorithms produce encrypted streams that are practically indistinguishable from random streams if you don't know the key. Observations may reveal the algorithm's block size, if any, by measuring the packet sizes — for a block cipher, network packets are more likely to contain whole blocks. But the block size doesn't say much about the algorithm strength; for example all AES key sizes use the same block size (16 bytes).
But observing encrypted channels from the middle only isn't very realistic: usually, an eavesdropper can start watching from the beginning. While it is possible to devise protocols that obscure the choice of algorithm to an eavesdropper, it is somewhat impractical and, in fact, counterindicated: such a protocol doesn't scale well. In practice, protocols tend to have variants, for example allowing different suites of cryptographic algorithms. Hard-coding cryptographic algorithms is bad design because algorithms can be broken or weakened — it's rare, but it happens: MD5 is thoroughly discredited for most uses, SHA-1 is considered somewhat deprecated today. More commonly than cryptographic primitives, cipher suites can have weaknesses due to design or implementation mistakes: witness attacks on TLS such as BEAST or the attacks on RC4. Supporting multiple algorithms allows reacting quickly to such attacks, without having to arrange for every participant to upgrade their software at the same time.
If the software supports multiple algorithms, then there needs to be some form of negociation between participants so that they agree on which one to use. That negociation takes place before the two parties have agreed on a way to encrypt information, so at least part of the negociation has to happen in cleartext. You need to negociate before you can encrypt, and the negociation of how to encrypt can't be encrypted.
You could do a two-phase negociation, where the parties first agree on a cipher suite to perform an encrypted negociation to agree on the cipher suite for regular communication; but that's rather convoluted for a very limited benefit.