Use of the Akaike information criterion

[shelkmike]

The manual says: "More gamma rate categories (-k) does not always mean a better fit to the data. While -k=2 nearly always fits the data better than -k=1, it may be the case that -k=5 has a worse likelihood than -k=3, and convergences between runs is more difficult with more categories. Try several and see what works."

Wouldn't it be better to decide which number of rate categories is the best using the Akaike information criterion instead of the likelihood?