MSA finds common motif's (domains or homologous) sequences for a given set of sequences. Only MSA can do this kind of job.
In contrast, Pair wise alignments find motif's (domain's) among a pair of sequences !
Key words here are "set of sequences" and "a pair of sequences"
Hope this helps
There are many methods of doing multiple sequence alignments. The most popular and quickest way is to do a progressive alignment where pairwise alignments are done for all the sequences and a tree is built up as a guide for the multiple alignment. This method doesn't guarantee optimal alignment.
The slowest way that does guarantee optimal alignment is to do a traditional dynamic programming approach of a N-dimensional matrix for all your sequences. This really isn't practical as the matrix space increases exponentially for each additional sequence.
If you mean "What is the advantage of working with an MSA over 'just' a pairwise alignment", then part of the answer is that, by including 3+ sequences in the alignment (i.e. going from a pairwise to a multiple alignment), you are able to identify variation in patterns of residue "conservation" (a dangerous word to use in this context, as it comes packed with a set of assumptions about the origins of the residues aligned in the same column...) in different columns of the alignment. This can be useful, as User1029725 refers to above, when identifying "motifs" in sequences.
This is reflected, for example, in the way that a pairwise sequence similarity search method such as BLAST scores, in a given search, all alignments of A (in query sequence) against A (in target sequence), the same.
Using a search method informed by a multiple sequence alignment of the query sequence(s) (e.g. HMMER), columns in which many A's are found will generally score higher in an alignment against an A in a target sequence, than columns with few As, which provides the potential to improve the sensitivity of the search.