When assessing the risk of an AI tool for use in the courts, what critical factor must be considered in relation to its intended function?