According to emollick, human opinions are essential for benchmarking non-verifiable AI tasks, urging adoption of qualitative research methods.