Yup. Computational AF with intelligent object detection, recognition, and memory is the future.
Sony's lock-on AF and face registration technically works like this already. It's just that the implementation hasn't been all that great because the computational power isnt there. But but just as we've seen eye AF improve to the level it's at now, as the processing power of mirrroless cams improve, we should see even more advanced capability in terms of object recognition and tracking
It's not so much computational power but maturation of software. Object and facial recognition software has steadily been improving as "neural networks" improve. Neural networks are based on actual photo libraries that provide datasets for first analyzing and then processing algorithms in object recognition in a photo. Because of popularity, human faces have had the most software development: however, there are other open source software that's for recognition of other objects. Creating datasets of different types of animals and their types of eyes/noses/mouths needs development time.