Computational modeling of collaboration and mentalization for benevolent bots
- Acronym: COMCOMBR
- Scientific director: Jean-Claude DREHER
- Institutional coordinator: Centre national de la recherche scientifique (CNRS)
Abstract: Our social interactions are often carried out in social networks, with either humans or artificial agents (i.e. bots). At the intersection of targeted projects 1, 2, 3, and 5, we propose approaches and methods that are different from and complementary to those already existing in the PEPR eNSEMBLE project, to understand the influence of bots added or not to experimentally controlled ‘artificial’ social networks. Characterization of the computational processes involved in our social decisions which occur in social networks is necessary to enable the fluidity of our social interactions in the digital world, and to identify factors that favor collaborative modes of interactions. We consider collaboration as a form of cooperation that requires mentalization of other agents’ intentions, so as to align towards a common goal. Therefore, characterization of the computational processes engaged in inferring the intentions of others is an essential part of this project.
WP1 will develop computational models of social decision-making with other agents (human or artificial agents). Based on a taxonomy of the computational processes involved in different degrees of mentalization, WP1 will: (a) determine, for each individual, the best algorithms describing 3 types of social interaction in cognitive tasks that test these social interactions (limited depth of mentalization, hierarchical mentalization and mentalization in networked interactions); (b) analyze and compare computational algorithms to account for observed social behavior. The deliverables will be a web-based App providing individual computational assessments/profiles of how each tested individual function during network interactions.
WP2 will use an experimental and theoretical approach to understand the influence of introducing benevolent bots that promote collaboration in networked interaction systems. Based on the extension of our platform into networks), we will carry out behavioral experiments in online decision- making when many participants are simultaneously connected in networks. An essential contribution will be to test not ‘naïve’ bots, but sophisticated bots equipped with algorithms (identified in WP1) to enable mentalization of the collaborative or competitive intentions of others. Our network interaction platform will enable: (a) the causal study of interaction dynamics and network topology when humans and bots exchange in networks, and (b) the modeling of information propagation and dynamic network formation.
WP3 will test the generalizability of the cognitive computational models of WP1 to highly interactive and dynamic environments, in the sense of the HMI domain. We will study how the level of interactivity and the quantity/nature of the agents (humans and bots) involved interact to set up and stabilize collaboration. In addition, we will study the degrees of mentalization required for bots to interact with humans in this framework. We will test social interaction situations in various technological settings (screen wall, PC, virtual reality) and study how these technologies modify mentalization types (limited mentalization depth, hierarchical mentalization, network mentalization) on interactive tasks. To this end, our approach is to combine HMI models to explain and predict the behavior of a single user with the mentalization models designed in WP1 to encourage more efficient collaboration.
Taken together, our social network behavioral experiments and computational models of social interactions/inference of intentions will enable us to understand the emergence of collective intelligence and of hybrid collaboration based on human-bot interactions.
Investigating mechanisms for scaling online collectives
- Acronym: Data2Laws
- Scientific director: Cécile BOTHOREL
- Institutional coordinator: École Nationale Supérieure Mines-Télécom Bretagne Pays de la Loire
Wikipedia, which will celebrate its 25th anniversary in 2026, has been able to create encyclopedias in more than 50 languages. Free and open source software organizations, which began challenging the software industry in the 1990s, are now considered a core element of the business.
The challenge of the DATA2LAWS project is to propose a theory of the massification of such large online collectives. We will look for invariants, if any, in the constitution and consolidation of sustainable, massive projects, and identify the conditions that allow them to scale. Our aim is not to oversimplify the sociological mechanisms involved, but rather to understand the common mechanisms while highlighting the different variants.
By focusing on the notion of properties and mechanisms, this project aims to overcome the problematic alternative between analyzing a project in great detail without being able to generalize the results, on the one hand, and producing an overly abstract analysis that fails to take into account the particularities of each situation, on the other. We will use Robert Merton’s formulation of what are now commonly called “middle-range theories”. Because middle-range theories are parsimonious, they are transferable to other empirical contexts.
We will decompose a process, the massification of online projects, into different properties and mechanisms. We will first work with Wikipedian projects and in further steps, transfer our middlerange theories within other collectives, e.g. OW2, Eclipse projects, Open Street Map France, Open Food Facts, Framasoft…
The main scientific challenges are 1) to determine whether all these large collectives follow the same 3-phase evolutionary dynamics (growth, percolation, cruising), 2) whether they share common properties at the level of contributors, in their motivations and trajectories through the phases, in the way they form collective actions and according to which organizations, 3) to refine the diversity of behaviors and collective dynamics, and 4) to extract middle-range theories of scaling.
To achieve this, we will combine several complementary disciplines: data mining, signal processing, statistical physics, social simulation, sociology, and management, which is a challenge in itself. By adopting a complex network approach that brings us together, we will confront different viewpoints and methods to capture the multiple facets of the scaling phenomenon.
We expect our project to improve decision making for the management of large collectives. Our non-funded partners (Wikimedia France, OW2, The Eclipse Foundation) are interested in improving their “community building” and more generally their governance. Our results could lead to different tools adapted to the different phases, to detect when a collective doesn’t meet the conditions for scaling up, or predict if scaling up could appear with the introduction of certain organizational rules.
DATA2LAWS brings an ambitious new scientific challenge to PEPR eNSEMBLE, meeting the PC4 WP 4.2 objective of theorizing online community collaboration, while introducing two new partners (Université Littoral Côte d’Opale and the Centre de Physique Théorique) and two new disciplines (signal processing and statistical physics).
Physical deformation and collaboration
- Acronym: DECO
- Scientific coordinator: Céline COUTRIX
- Institutional coordinator: Université Grenoble Alpes
DECO focuses on studying dynamic deformation of devices for digital collaboration, with the goal of improving collaboration between humans and the system, as well as among humans through the system. DECO explores the physical deformation of system input or output, considering users and the system as potential agents. The aviation context serves as the main field of study, with particular emphasis on three collaboration scenarios: shared air/ground piloting, remote control tower and drone teleoperation. Deformation of devices provides opportunities for adaptation to user capabilities, collaboration contexts, and collaborative tasks. DECO examines how device deformation can improve operator collaboration, situational awareness and shared control. For example, interfaces can be physically deformed to ease the perception of critical information. These changes can occur in a co-located or remote environments, adapting to ongoing tasks and supporting information sharing and decision-making. DECO considers synchronous and asynchronous collaboration situations.
Additionally, DECO addresses collaboration between humans and intelligent systems, leveraging physical shape changes to provide combined visual and haptic feedback. This feedback informs users about the transformation intentions of the intelligent system, the degree of automation engaged and the possibilities for action or regaining control by the user. The goal is to reduce potential errors resulting from over-reliance on automation or under-utilization of intelligent systems.
DECO’s lines of research extend to long-term and large-scale digital collaboration, using changes in physical form for visual-haptic feedback to improve the user’s awareness of time spent. This adaptability of form can also support collaborative activities within large communities.
DECO involves collaboration between researchers in Human-Computer Interaction, Robotics, Mechatronics, Cognitive Sciences and Ergonomics, with aeronautics professionals to develop scenarios and evaluate the proposed interactions.
The project aims to produce prototypes of deformable interfaces for collaboration, conduct empirical studies and challenge existing theories on digital collaboration and reconfigurable interfaces.
DECO’s results have the potential to extend beyond aviation, influencing critical areas such as automotive or medicine but also less critical areas such as office computing. DECO’s comprehensive approach addresses scientific challenges related to the dynamic aspects of shape for digital collaboration, contributing to constructive advances in the field. DECO is intrinsically multidisciplinary and will involve new scientific communities not yet present in eNSEMBLE, such as robotics.
Prediction for shared cognition in collaboration with human or artificial agents
- Acronym: PRECOG
- Scientific coordinator: Ouriel GRINSZPAN
- Institutional coordinator: Centre national de la recherche scientifique (CNRS)
A wealth of behavioral and neuroimaging evidence highlights cognitive changes that emerge when we function in groups rather than individually. This raises several questions such as: What happens when inter-individual relations are embedded in digital spaces in either co-located or remote situations? Furthermore, can shared cognitive functioning be maintained when artificial agents are included in the group?
To successfully function with others, it is not enough for users to focus on their own goals and actions: They also need to have a representation of the other partners’ intentions to predict and adjust to their actions.
The present project seeks to investigate ways to optimize users’ abilities to anticipate others’ behaviors in hybrid digital spaces. The intelligibility of systems (i.e. the production of predictable and understandable behaviour) is a major challenge leading to the concept of mental models, i.e. the mechanisms by which human beings are able to produce explanations of how others work and predictions of their future states.
In this project, we wish to draw on the theoretical framework of joint action to better understand how to support the construction of our partner’s mental model and to improve partner predictability. Interestingly, the study of joint action is gaining momentum in the field of cognitive science where theories put forward the idea of a collective mode of brain functioning when human agents act together.
Our project aims to explore how to support the construction of the user’s mental models during both human-human interaction and human-autonomous agent interaction. The project follows an interdisciplinary approach where theory is driven by the most advanced research in cognitive science matched against the current state of the art in digital collaborative environments.
It involves three partners (UGE, LISN, ONERA) in the field of HCI with recognized expertise in cognitive psychology and one partner from the field of cognitive science (ENS).
Our project adopts the perspective of the sense of agency, which refers to the sense of being the author of one’s actions, and to the experience of controlling the effects of one’s actions on the outside world. The sense of agency results from the consistency between the prediction and the outcome of action.
Two scenarios seem possible in the context of actions in which several individuals cooperate. The first, which we favor in the present project, yields an increase in the sense of agency, where each partner in the action feels a sense of agency for the others as well as for her/himself. The second, which we believe should be avoided, is a collapse of the sense of agency, associated with the phenomenon of diffusion of responsibility within a group and attributed to difficulties in grasping the partner’s intentions.
The present project seeks to identify the factors that promote a sense of agency in collaborative situations. Our project will investigate how to convey others’ intentions in different collaborative situations, from active co-manipulation of objects to simple cohabitation in a common environment. This project will develop interoperability solutions to facilitate seamless collaboration across eXtended Reality (XR) and robotic platforms. It will test several use case scenarios such as cohabitation between humans and a swarm of autonomous agents, cooperation with robots using a remote-controlled robot, on-line share editing of 2D documents, co-manipulation of 3D items in XR.