For some years I've been thinking about how to use modern machine learning and artificial intelligence in animal training. Most "automated" uses of animal training involve lab animals pressing some buttons, but it should be possible to teach a mouse/rat to visit a particular place in a cage, or hold a certain pose, sniff out something, or a parrot to hum a tune.
But it would also require an algorithm to score "progress" toward a multi dimensional training goal and encode that into binary reward/no reward signals.