Broadening the Tests of Learning Models