Uploaded February 2026 | Updated November 2026, 3 days ago
SORS: Collective Artificial Intelligence (CAI): From Independent Models to Autonomous Cooperative Learning Systems
Speaker:
Alfonso Amayuelas, PhD student in Computer Science in the Natural Language Processing Lab at the University of California, Santa Barbara
Abstract:
The rapid evolution of artificial intelligence has shifted focus from isolated, task-specific models to interconnected systems where multiple agents collaborate to solve complex, dynamic challenges. This presentation explores the progression from independent models to cooperative systems, focusing on LLMs as autonomous agents capable of learning not only from data but also through interaction with environments. We highlight how multi-agent debate frameworks enhance reasoning, factual accuracy, and decision-making by leveraging diverse perspectives, and how environment-grounded learning enables models to adapt through feedback and exploration. In doing so, we address critical challenges such as adversarial vulnerabilities, communication reliability, and long-term autonomy. By synthesizing insights from previous works, this talk outlines a roadmap for building trustworthy CAI systems that integrate cooperation with autonomous, environment-driven learning.
SORS: Collective Artificial Intelligence (CAI): From Independent Models to Autonomous Cooperative Learning Systems
Speaker:
Alfonso Amayuelas, PhD student in Computer Science in the Natural Language Processing Lab at the University of California, Santa Barbara
Abstract:
The rapid evolution of artificial intelligence has shifted focus from isolated, task-specific models to interconnected systems where multiple agents collaborate to solve complex, dynamic challenges. This presentation explores the progression from independent models to cooperative systems, focusing on LLMs as autonomous agents capable of learning not only from data but also through interaction with environments. We highlight how multi-agent debate frameworks enhance reasoning, factual accuracy, and decision-making by leveraging diverse perspectives, and how environment-grounded learning enables models to adapt through feedback and exploration. In doing so, we address critical challenges such as adversarial vulnerabilities, communication reliability, and long-term autonomy. By synthesizing insights from previous works, this talk outlines a roadmap for building trustworthy CAI systems that integrate cooperation with autonomous, environment-driven learning.


![SORS: ENSO-driven variability reconfigures the altitudinal frontier of dengue risk in the Andes
Speaker:
Adrià San José, Recognized Researcher at GHR, BSC
Abstract:
Despite the high burden of dengue in Latin America, high elevations have historically protected megacities such as Mexico City, Quito, and Bogotá. Climate change is expected to shift these elevational suitability frontiers, yet most long-term projections rely solely on mean-state trends and overlook the role of climate variability, potentially underestimating risk. Using spatiotemporal decomposition techniques, we identified a mechanistic cascade linking (1) the El Niño–Southern Oscillation (ENSO) to local Colombian climate, accounting for ~85% and ~40% of interannual temperature and rainfall variability, and (2) local climate to dengue dynamics, explaining [63–76]% of incidence and [42–55]% of its altitudinal range. This cascade was further validated by model selection across formulations that incorporated both local and global climate variables. Non-linear effects indicate that dengue incidence and transmission altitude expand exponentially during warm ENSO phases, producing an upward “stretch” that exposes immunologically naïve populations in highland areas previously considered unsuitable. If ENSO extremes intensify as projected, such fluctuations may trigger large outbreaks in high-elevation urban centers, fundamentally altering the global landscape of dengue. Our findings demonstrate that climate variability, acting through local temperature and rainfall, can reconfigure the geography of risk on timescales far shorter than decadal trends. This highlights critical limitations in current projections and detection-and-attribution frameworks, underscoring the need to incorporate anthropogenically modified modes of climate variability to accurately quantify the human fingerprint on global epidemic risk. SORS: ENSO-driven variability reconfigures the altitudinal frontier of dengue risk in the Andes](https://i.ytimg.com/vi/G-l95r5gecs/mqdefault.jpg)







