Crowdseeding Robot Design

Crowdsourcing is a well-known method in which intelligence tasks are completed by an anonymous group of human participants. These are tasks that cannot yet be adequately performed by computers. Rather than performing an intelligence task outright, one crowdsourcing strategy is to use human intelligence to complement machine intelligence. A key point in determining the potential of such a strategy is understanding the ways that human abilities most effectively complement the strengths of machine intelligence. We shed light on this relationship by 'crowdseeding' robot design: we find morphological features common to human-generated robot designs and incorporate them as an additional fitness objective in an evolutionary algorithm that searches over the same space of designs. We demonstrate that this approach outperforms the same evolutionary algorithm that is not crowdseeded in this way.