Formal voting biases and weaknesses like Arrow's or Condorcet's paradoxes are known for quite a long time, but they are often considered as curiosities for mathematicians and cases that rarely occur in practice. Although, recent elections underlined the actual weaknesses of our collective decision-making processes, either concerning the resulting lack of representativity (people getting elected but representing actually a minority of the population) or more globally the efficiency of the decisions taken. A significant gap exists in between on one hand the opinions of a networked population that are structured and evolve using social media, leading to a much more visible diversity and on the other hand the ones of representatives that are still structured by political parties. Such a situation leads to questioning the processes used to choose representatives and more globally to make collective decision-making with alternatives that are proposed more and more (liquid democracy for instance (Blum and Zuber, 2016).

The article by Sueur et al. shed new light on the situation by proposing to examine further the solutions selected along time by the animal kingdom in order to make collective decision-making. Although they advocate that the decisions taken are not the same (decision for an animal group to move to another place) and do not involve the same cognitive abilities at the individual level, the existing processes could get adapted to human contexts. One of the most striking advantages of such bio-inspired approach is that animal collective decision-making processes are robust and enable to manage conflicting views, diversity of opinions and avoid forms of despotism that are not much present in animals, rather consensual decision-making being the norm. Another argument, probably less put forward by the authors is that such solutions may scale well with large networked populations as they exist for some animal species. Therefore it looks like we could have a kind of readymade library of processes for collective decision-making that are yet efficient to make timely decisions for different purposes with different population sizes and structures.

Their argument is consolidated by the possibility of using AI technologies in order to enable to support the adaptation of such solutions. Without building explicitly the link, they identify that AI could be used in order to guarantee a fair process and to scale up the proposed solutions at the level of massive populations.

This is probably the less convincing part of the proposal and it concerns the relation between human and AI. Even if the authors admit that the acceptance of AI solutions by part of the population is a key issue, as it concerns directly the legitimacy of the process and the compliance of the population with the resulting decision. They tend to minimize the gap to fill before having a fair AI with potential behavior that can be verified before using it with confidence to support a democratic process of decision-making. 

Nevertheless, the article brings forward good arguments, well formalized using relevant concepts for the use of bio-inspired solutions for collective decision-making.

References

Blum C, Zuber CI (2016) Liquid Democracy: Potentials, Problems, and Perspectives. Journal of Political Philosophy, 24, 162–182. https://doi.org/10.1111/jopp.12065

Sueur C, Bousquet C, Espinosa R, Deneubourg J-L (2021) Improving human collective decision-making through animal and artificial intelligence. hal-03299087, ver. 3 recommended and peer-reviewed by Peer Community in Network Science. https://hal.archives-ouvertes.fr/hal-03299087

In their article “Behavioural synchronization in a multilevel society of feral horses”, Maeda and colleagues (2021) use stochastic multi-agent based modeling to explore the degree to which feral horses synchronize their behavior across multiple levels of organization. The authors compare a drone-derived empirical data set on a feral population of horses with simulated data from multi-agent-based models to determine whether behavioral synchronization of resting and movement states in a multilevel society can be described by one of three models: A) independent model in which horses do not synchronize, B) anonymous model in which horses synchronize with any individual in any unit, C) unit-level social model in which horses synchronize only within units and D) herd-level social model in which horses synchronize across and within units, but internal synchronization is stronger. In a series of 100 simulations for each of seven different models, the authors conclude that evidence for the herd-level model had the strongest support in relation to the empirical data. This finding suggests that connections among individuals in such multi-level societies are rather complex in that local connections are not the only interactions driving social behavior, and specifically synchronization. This approach could be successfully applied to a number of different species that exhibit multi-level organization and possibly fission-fusion dynamics.

This study is especially innovative and interesting for three major reasons. First, the use of drone technology to successfully identify individual animals and generate social networks is highly novel and permits the study of large multi-level social groups of animals that previously have been challenging to study due to limitations in collecting data at an appropriate scale. Second, the comparison of multi-agent-based models with actual empirical data is highly applauded. Most agent-based studies design their parameters from previous empirical studies, (sometimes with questionably simple assumptions) but rarely do they actually compare model outputs to their own empirical data. This is an important next step in the burgeoning field of agent-based modeling. Finally, this study sheds light on the utility of using relatively simple mathematical models to explain highly complex behavior. It also highlights that feral horses can synchronize their behavior beyond clustered local connections which suggests that they possess the cognitive ability to track the behavior of individuals at higher social orders. As the authors state, in a multilevel society, inter-unit distance should be moderate, that is “not too close but not too far” because this strategy simultaneously avoids inter-unit competition while also providing the benefits of social buffering that comes with large group living, such as protection from bachelors or predators.

As the authors dutifully note, there were also some limitations to the study: (1) the relatively sparse empirical dataset that made it difficult to resolve the relative fitness of the two herd-level models (absolute versus proportional social models), (2) the lack of a temporal component that would provide a better understanding on how synchronization flows through the social/spatial network, and (3) the limited variation in the parameters tested which constrained identification of their true function in the model. Such limitations, however, provide fruitful avenues for further development of the model in future studies.

Overall then, this study provides new insights into the processes underlying the behavioral synchronization process and thus nicely contributes to the understanding of collective behaviors in complex animal societies as well as the evolution and functional significance of multi-level animal societies. This study is a fine addition to both the fields of agent-based modeling and the evolution of collective behavior in complex societies. I thus highly endorse its publication.

References

Maeda T, Sueur C, Hirata S, Yamamoto S (2021) Behavioural synchronization in a multilevel society of feral horses. bioRxiv, 2021.02.21.432190,  ver. 3 peer-reviewed and recommended by Peer community in Network Science. https://doi.org/10.1101/2021.02.21.432190