The rewards-based approach in AI development, as seen in Go and Dota, has revealed its inefficiency, as it requires stumbling upon the correct booking in order to receive a reward. This approach also fails to provide a genuine understanding of the problem being solved.