Researchers seek to injecting human uncertainty for improved AI trustworthiness