Vienna, January 13, 2024 - A recent study involving Central European University (CEU) is paving way for a new field of research that merges artificial intelligence and complexity science to understand how ongoing interaction between humans and algorithms can profoundly impact social dynamics.
The study, published by an international team of AI and complexity scientists, titled "Human-AI Coevolution," explores how processes in a world where algorithms and recommendation systems increasingly guide daily decisions influence human behavior. Researchers found that these processes create a "feedback loop" in which individual choices and automated suggestions reinforce each other, and which, the researchers argue, generate complex unpredictable effects that cannot be explained by traditional models of human-machine interaction. The paper suggests a new framework to study the co-evolution of AI and society, with the aim of shaping their shared future consciously and responsibly.
"Artificial intelligence has already penetrated many areas of our lives, and the process continues unstoppably. A human-AI ecosystem is emerging, which implies mutual evolution and adaptation," said Professor Janos Kertesz from the CEU Department of Network and Data Science, an author of the study. "This process raises a number of profound questions, from ensuring social benefit, to avoiding inequalities related to accessibility or potential AI bias, and even moral-philosophical dilemmas, for example, the extent to which AI should be held responsible for decisions taken by self-driving cars in a fraction of a second. Clearly, an interdisciplinary approach is needed, involving computer scientists, network researchers and social scientists," he added.
The study's research team emphasizes the importance of a new, cross-disciplinary perspective to address the challenges of coevolution. They present concrete examples of human-AI ecosystems - not only social media and digital marketplaces, but other large online platforms, such as geographic mapping and navigation services, as well as chatbots based on generative AI. The authors also highlight the need for new regulatory and policy tools to monitor and manage the feedback loop that governs our digital interactions.
"The feedback loop between humans and AI creates unprecedented forms of interaction, and the complexity of human-AI ecosystems is in constant expansion," said Dino Pedreschi, professor of Computer Science at the University of Pisa and co-author of the study.
Complex systems consist of many interacting units and are known to produce unexpected, so-called emergent behavior, which cannot be explained from the features of the constituents. The feedback loop is based on such interactions among many humans and AI units and can therefore produce unexpected effects.
"If we want to understand the real impact of AI on our society," said Luca Pappalardo, researcher at CNR and professor at Scuola Normale Superiore in Pisa, and co-author of the study, "we need to reinterpret our understanding of complex systems in light of this continuous feedback between humans and algorithms."
According to the authors new regulatory and policy tools are, therefore, needed to mitigate inequalities and to ensure the ecosystem evolves for the public good. Emanuele Ferragina, professor of Sociology at Sciences Po in Paris and co-author of the study, underscored the urgency of addressing legal and policy barriers. "To fully understand human-AI coevolution, we need greater transparency from major online platforms," she said. "Initiatives such as the EU's Digital Services Act can make a difference, but it's also essential to ensure an equitable distribution of 'recommendation tools' in a more competitive market."
