Hopefully someone in the field will read this and be able to answer my question: I've spent a lot of time googling research about animal intelligence, and I now know a bunch of things intelligent animals can do, but I have no idea where their limits are, what they can't do. Everything published uses such... nonflexible tasks. If we're trying to measure general intelligence, why isn't anybody trying to teach corvids,…
Thinking about the project with corvids in mind, I’d suggest two nuances to your protocol.
1. Stick with plain mazes that have no crates in order to train the association with the onscreen character. Basically just spending more time with navigating mazes of increasing complexity before introducing crates at all. Then if they succeed with those, scale the maze complexity back down and add the crate mechanic.
2. Make use of audio. Birds have great eyes, but their natural interaction will be with their beaks, and I’m not sure how easy it is for them to track the onscreen motion while pecking. Using sound to signify “move to empty space” vs “tried to move into a wall” vs “pushed a crate” vs “got snack” could tighten the feedback loop for them and let you see them planning ahead: “I see I have to go ‘up, up, right‘ to get the snack, so I’ll press those buttons without checking the screen each time to confirm.” Might even be good for “move to empty space” to have a unique sound for each direction.