Can we make our robots less biased than we are?