I've just done a test on whether the AI recognizes mount bonuses. I took the standard EDU.txt, gave Hastati the stats of Principes, but then gave Hastati a mount_effect of +50 against horses and Principes -50 against horses. I then started up a custom battle, giving myself (Julii) one Hastati and one Principes, and the AI (Carthage) one Sacred Band Cavalry and one Poeni Infantry.

On the first run, I positioned the Hastati to the left and the Principes to the right, approximately equidistant from the AI's line of approach. The AI seemed to be sending both its units toward the Hastati, with the Poeni Infantry in front and the Sacred Band Cavalry in the back, both moving at a walk, but it broke off its cavalry to charge the Principes. After a short period of battle, the Principes routed, and I disengaged my Hastati to attack the Sacred Band. The AI didn't retreat from the Hastati as it should have. (I lost anyway, because Hastati are no match for Poeni Infantry without flanking.)

On the second run, I switched the positions of Hastati and Principes. The AI performed identically, heading toward the Hastati and then breaking off its cavalry.

On the third run, I spread the Hastati in a thin line and positioned the Principes behind them. The AI approached as normal and engaged the Hastati with its infantry. It apparently couldn't figure out how to get around the Hastati, because it left its cavalry behind its infantry rather than circling around to attack the Principes. Once the Hastati routed, the AI used its cavalry to chase them (or possibly to attack the Principes—it was hard to tell, because the Hastati routed into the Principes, and the Sacred Band ended up fighting the Principes).

Conclusion: the AI does consider mount_effects to the extent of its attack selections. However, it doesn't appear to choose its fights if the player initiates combat—it fights rather than running as it should. The AI also doesn't quite seem to understand maneuvering around an unfavorable fight, at least not consistently.

-Simetrical