Sometimes a gene model is split into two different genes. So, there is no correct choice between the multiple new IDs if that is what happened.
Alternatively, multiple ids may have arisen because Ensembl now uses build GRCh38, and in build 38 there are alternative contigs. Hence, the same gene can be referenced by two different identifiers (one for the reference chromosome, one for the alternative contig). You can discern if this is what is going on by getting the "chromosome" names for each of the converted identifiers. Look also at the gene names.