Depends on what model you use for the AI and in what field. With ML you have a model that dictates your constraints and how it applies what it learns. If you have an AI that models let's say Hannibal and his tactics on a battlefield, well you may do well against many opponents in a battle simulation. On the other hand if you modeled "Bevis and Butthead" and had that AI run a restaurant, you may have a colorful but disastrous result. You would have to find an appropriate context for the model to be useful.

If not done right it can end up being a huge resource sink that never lives up to the promise