Wow, I ADORED this idea! And it has neat execution and nice graphics too!
However, there are places where it's quite a bit unintuitive. It's completely intuitive to get him to keep going somewhere, but, for instance, if he's going somewhere and I start punishing the "go", and let's suppose he sees nothing. What DO I want him to do? Take off? Turn? And what if he takes off, and I favour that, am I saying "Yes, I did want you to take off in this istuation" or "Yes, keep doing that"?
Because when I reward the parrot for turning to the proper place, he interprets it as "being rewarded for just turning". That's actually part of the fun in this instance - just keep him spinning while I think! Bit of a "bug turned feature" situation in a game design standpoint. However, the issue is when this happens with OTHER stuff. It's specially frustrating when there are only two blocks to manuever and you have to hope that the parrot will understand that, as soon as he land on the block (because he needs to, there's no ther way), I WILL NOT want him to go forward OR turn.
You've made an idea that has HEAPS of potential and you can build on it for a long, long time. But you've also "shot yourself in the foot" of sorts: You're making an AI. You're making a game about machine learning. Even if it's a simplified version. That's both amazing for a game jam - but also terrifying to think about keep working on it. Still, kudos - it works, it's beatable, it's inventive and you've done it all in a jam!