Machine Learning and Human Capital Complementarities: Experimental Evidence on Bias Mitigation