its a pretty interesting idea :) i dont know if you're looking for feedback or suggestions but one thing that strikes me above all is that the algorithm isnt ever "done learning", god knows its always siphoning data. Instead mb its more like if you select the cat then it sends you to a level with only cats, etc... and you could extend the reflection to filter bubbles in general.
thank you so much for your comment! I agree, you are absolutely right. I'll be honest, I struggled with variable control in Bitsy. I did originally intend to make this an infinite environment, but ended up turning it into a simpler creation (also due to time constraints). I have now included some extra dialogue to imply the algorithm's ever-changing, ever-data-consuming nature!