Frustrated by the AI industry’s claims of proving math results without offering transparency, a team of leading academics has proposed a better way