Lessons in leadership from a school of problem-solving fishbots

In most sci-fi scenarios, the point at which robots develop traits of individual leadership alongside a sense of solidarity and willingness to work together is the beginning of the end for their human creators.

So it might be that the underground remnants of human civilisation will one day rue the work of a group of Chinese researchers who have programmed robotic fish with algorithms to give the machines “personality”, endowing them in the process with the capacity to better work together to overcome obstacles in their way.

The aim of the research, led by Wang Chen of Peking University’s State Key Laboratory of Turbulence and Complex Systems, was to use the robotic fish to investigate how personality traits evolved and effective leadership emerged in a group during increasingly difficult tasks.
 
The results, published in the journal Royal Society Open Publishing, suggest that the evolution of diverse personality types, and the different behaviours informed by personality differences, provides a collective benefit for social groups in successfully overcoming challenges.

The results of the research has implications both for developing robotic systems and for better appreciating the organic self-organisation of social animals, including humans.

For emerging applications of autonomous robots working in environments such as the deep ocean or outer space, the authors note, “robots have to learn to decide for themselves whether to co-operate with others”. That’s because engineers cannot predict and design for when and how an autonomous robot should sacrifice its own wellbeing for the good of the group in specific situations.

“In this context, a variety of robotic control strategies have been effectively developed using inspiration from social animals to accomplish team tasks, such as environmental monitoring, surveillance, exploration, pursuit and evasion, search and rescue, transportation and maintenance in harsh environments,” the authors write.

“However, the performance of such biology-inspired robotic algorithms is sometimes compromised due to those open issues for understanding co-operation mechanisms in social groups in nature, the most demanding among which is why and how group-level co-operation can be achieved through selfish individuals’ local interactions.”

In the simplified model used by the experimenters, groups of five robotic fish were programmed for the “personality” trait of either “boldness” – willingness to initiate a collective action, which could adapt with experience – or “shyness”  – willingness to follow others’ actions, which could also evolve with experience. Willingness to initiate was inversely correlated to a willingness to follow.

The fish were then set a foraging task requiring the group to overcome an obstacle to reach a “food source” (in this case a recharging station). The obstacle was a barrier of unknown and changing weight; the greater the weight, the more the fish needed to work together to move it.

Over repeat rounds of the task, each robotic fish could to choose to initiate a collective action; or, once an initiator appeared, to become a follower, joining in the effort to overcome the adversity; or be a free rider – not contributing to the effort to remove the barrier but benefiting nonetheless. A bolder fish that did not take the initial lead, being less disposed to following, was more likely to become a free rider.

What the results show, the researchers report, is that more bold fish emerged in the group when the foraging tasks were easier, while more fish participated in pushing the obstacle when the task was more difficult. That’s because, for easy tasks, an initiator was more likely to succeed without needing many co-operating fish, and the experience of success would then increase its subsequent willingness to initiate.

On the other hand, in a group with more bold fish, and thus fewer fish likely to follow, the initiator of action was likely to fail when the task was hard, so its subsequent willingness to initiate was consequently more likely to decrease.

The researchers conclude that the greater the degree of adversity the group must overcome, the greater the degree of personality trait divergence required.

So our future robot masters may not all be leaders; even among machines leaders require followers. Someone has to do the grunt work of carrying out the plans for world domination.