Learning from nature: uncertainty and heterogeneous groups make smart decisions possible
When groups make decisions – be it people agreeing on an idea, robots coordinating their tasks or fish determining their swimming direction - not every individual has the same influence. Some have more reliable information, others are better networked and enjoy greater social recognition.
A new study by researchers from the Cluster of Excellence ‘Science of Intelligence’ (SCIoI) shows that the combination of uncertainty and diversity is decisive for how groups reach a consensus. The work by Mohsen Raoufi, Humboldt-Universit?t zu Berlin (HU), and Vito Mengers, Technische Universit?t Berlin (TU), and other colleagues, published in the journal Scientific Reports, shows that groups make faster and more precise decisions when individuals not only take into account the opinions of their immediate counterparts, but also utilise their self-confidence and networking within the group. However, strong self-confidence alone does not always lead to better decisions. Overconfident group members can even mislead the decision-making process if they use the wrong information.
Why some groups make better decisions than others: Diversity meets uncertainty
Classic decision-making models assume that all individuals contribute equally to the group consensus. In reality, however, groups are diverse, both in terms of their knowledge and their influence. Just as some people are experts on certain topics or individual fish in a school have a more precise view of predators, some individuals have more precise and reliable information than the rest of the group. Others, on the other hand, are more networked within the group, which means that their opinions spread faster and further. Knowledge and networking, these two forms of diversity, are closely linked. Those who have more knowledge from the outset also tend to be more influential and help others to reduce uncertainty. At the same time, highly networked individuals gather more and more information through their numerous interactions and thus become more confident in their decisions over time. This dynamic process enables groups to filter out weak or distorted information and reach reliable conclusions - provided that central individuals do not overestimate their certainty.
‘Whether in animal swarms, human societies or robot collectives, groups are naturally heterogeneous because each individual brings their own perspective and experience,’ explains Mohsen Raoufi, one of the lead authors of the study. ‘The really fascinating thing is that groups can utilise this diversity - without any central control - simply by incorporating uncertainty into their decision-making process.’
How group decisions were analysed
To investigate these effects, the researchers developed a model in which individuals - whether robots, fish or humans - dynamically adapt their beliefs and the certainty about them as soon as new information is added. The results show that a mere diversity of perspectives is not enough to promote better decisions. Groups made faster and more accurate decisions when uncertainty was considered as a guiding factor. When all individuals had the same amount of certainty and were equally well connected, consensus building was slow and unreliable. In heterogeneous groups, on the other hand, taking uncertainty into account led to opinions being weighed up better, resulting in more efficient and better decisions.
One of the biggest surprises: When particularly well-connected individuals became overconfident too soon and thus acted overconfidently, they began to dominate the group - even when they were in the wrong. ‘We often assume that influential individuals should act confidently,’ says Vito Mengers, another lead author of the study. ‘However, our research shows that overconfidence quickly becomes harmful. If central actors or algorithms claim absolute certainty too early, this can lead the entire system astray - be it a group of humans or a robot network.’
Why this is important: AI, social networks and insights from nature
The results of this study offer valuable insights for the development of AI and robotic systems. Self-driving cars could not only analyse sensor data, but also take into account the reliability of information from other vehicles and thus improve road safety.
Many natural systems already utilise the principle of adapting to uncertainty. Schools of fish, flocks of birds and colonies of ants do not treat all information equally, but adapt dynamically. By studying these mechanisms, we can not only better understand nature, but also optimise human cooperation and artificial intelligence.
Further Informations
Contact
Mohsen Raoufi
Institut für Biologie der Humboldt-Universit?t zu Berlin / Exzellenzcluster Science of Intelligence (SCIoI)
Vito Mengers
Robotics and Biology Laboratory der Technischen Universit?t zu Berlin /Exzellenzcluster Science of Intelligence (SCIoI)