I'll refer to what I assume you called the sokoban example, the "banana test" from part 2, which is the most relevant to our discussion. In it, a pigeon who has been trained that sometimes pushing stuff around is rewarded with food and sometimes climbing stuff is rewarded with food manages to solve a puzzle where it must push a box to below a banana and then climb the box to reach the banana.
First of all, thanks, that's a really cool video and exactly the kind of research I'm looking for.
However, I don't think it goes far enough to actually measure what I'm interested in. In the comments of this similar video [1] about a crow who has been trained to do each of A, B, C, D separately to get food doing A+B+C+D to get food, a comment captures what's wrong with it perfectly by saying:
> 2:20 The "use everything on everything" stage of playing a point-and-click adventure game.
Sure, we know birds can figure out that if step 1 is sometimes good and step 2 is sometimes good, then step 1 and then step 2 might be good. But that's not at all enough to solve, say, level 2 in [2], a video of a Sokoban game. If I trained a toddler to solve puzzles like level 1 (so it knows how to walk and it knows how to push crates), if could spend weeks moving and pushing crates around and not reach a solution. The repertoire of behaviors that allows them to play the game is just a foundation for a whole world of logic and planning that needs to be mastered to actually solve nontrivial Sokoban levels. Had I seen a crow solve level 3, I'd believe it is no less intelligent than some of my friends, and that if we put our minds to it, we could find a way to communicate, trade, and help our new corvid friends build a civilization not less sophisticated than our own. Based on that video, I see no reason to believe that about pigeons (yet).
[1] https://www.youtube.com/watch?v=cbSu2PXOTOc
[2] https://www.youtube.com/watch?v=yix2AVVUwe0