We assign the value $1$ to a don't care in order to try and increase the size of prime implicants, thus reducing the complexity and cost of the circuitry. Don't cares by definition do not affect the behavior of a combinational circuit, so treating them as $1$'s and $0$'s is good. On the other hand, if you're designing a sequential circuit, problems in wiring may cause your machine to jump to an unexpected state, if don't cares are treated as such, and not assigned a value of $0$ and $1$ then the machine will be stuck in that state, and will not function properly. You can either set the don't cares in a way to always return to the initial state in the loop (costly, but a time-saver), or treat them as $1$'s and $0$'s to increase prime implicants (cheaper, but has to go through the loop from an arbitrary point before returning to start).
P.S. : I'm not sure if the question is on topic here.