Towards a New Science of Collective Collaboration Between Humans and Machines
Researchers, including VUB/ULB professor Tom Lenaerts (AI Lab), are calling for closer collaboration between AI developers and scientists from the social and complexity sciences
An international group of researchers, including Professor Tom Lenaerts (VUB/ULB), argues in a new scientific opinion piece that it is time for a fundamentally new approach to collaboration between humans and intelligent machines. According to the authors, there is an urgent need for a new scientific discipline focused on collective, cooperative intelligence. In other words: how can we ensure that groups of people and intelligent machines truly collaborate to make our societies more sustainable and just?
The great challenges of our time demand cooperation on a global scale. Yet we still lack a proper understanding of how such collaboration emerges spontaneously and remains effective in complex environments. Traditional models from complexity science (Complex Systems Science, CSS) do explain group behaviour and social interaction to some extent, but they rarely account for the subtleties of individual decision-making and dynamic environments.
According to Lenaerts and his co-authors, this requires the use of so-called multi-agent reinforcement learning (MARL), a field in which multiple autonomous systems learn and make decisions together. This area of AI, the researchers argue, offers new opportunities to better understand how cooperation can arise organically between intelligent entities – both human and artificial. “By combining the computational power of MARL with insights from complexity science, we can build models that realistically capture both human behaviour and group dynamics,” says Professor Tom Lenaerts. “This is a necessary step towards making human-machine collaboration genuinely effective.”
The researchers advocate for closer collaboration between AI developers and scientists from the social and complexity sciences. In doing so, MARL can more accurately simulate individual cognitive processes, while CSS provides insights into how complex collective behaviours emerge and stabilise. “We must move beyond simplistic models if we are to make real progress in understanding and guiding collective intelligence,” Lenaerts stresses. “The future lies in systems that are not only intelligent, but capable of thinking and acting together – for the benefit of all.”
The article outlines concrete steps for further shaping this emerging field of research, including examples of projects that are already working in this direction. In this way, the authors aim to lay the scientific groundwork for systems in which cooperation, sustainability, and technology go hand in hand.
More information
Tom.lenaerts@vub.be
https://www.pnas.org/doi/full/10.1073/pnas.2319948121 .
