Your simpler model of the mechanism would seem to suggest the very same action that the article’s more complicated model suggests, viz. find a better training method than reinforcement learning.