Updates from I2SC

Image generated by ChatGPT This week we discussed reference letters for PhD applications and internships, focusing on how to approach potential letter writers and what materials to share. One key takeaway was that, as you progress in your career, it is normal and often helpful to share drafts or bullet points with your referees. The goal is to make the process easy for them and enable them to write a strong, informed letter. When to ask: Request letters well before deadlines so referees have time to reflect and write thoughtfully. Give them an easy way to decline if they cannot provide a strong letter. Ask when your work is still fresh in their mind and before busy academic periods. Avoid last-minute requests or assuming someone will say yes. Early, respectful requests increase the likelihood of a detailed and supportive letter. What to share: Do not assume they remember details or understand the role. Provide a complete packet—your CV, cover or research statements, job descriptio...

There's a particular kind of tragedy that doesn't come from bad luck or outside forces—it comes from the inside, from the very act of trying to escape. It's the tragedy of someone who sees the trap, runs from it, and finds themselves caught anyway. Or worse: someone who never even knew the trap existed, and whose every instinct to survive becomes the mechanism of their own undoing. This is the paradox of defiance. And once you see it, you may start to see it everywhere. Two African Stories, One Uncomfortable Truth Take Ola Rotimi’s The Gods Are Not to Blame , a Yoruba retelling of the Greek myth of Oedipus. A man named Odewale is fated at birth to kill his father and marry his mother. In the Yoruba context, this isn't just a "bad prediction"; it is a decree from the ancestors, a cosmic blueprint. His parents, terrified, try to have him killed as an infant. He survives, is raised elsewhere, and eventually, through pride, anger, and a series of tragic misun...

Artificial intelligence is increasingly framed as a climate solution. From optimizing power grids to improving climate models, AI is positioned as an accelerator of decarbonization.But emerging research suggests a more complex reality. Recent analysis in Nature Climate Action argues that AI’s climate impact cannot be evaluated through isolated use cases. Instead, it must be assessed across lifecycle energy demand, infrastructure externalities, governance systems, and information dynamics (Nature Climate Action, 2025).The central question is no longer “ Can AI help climate action?” It is: Under what conditions does AI produce a net climate benefit? Where AI Demonstrably Supports Climate Goals Climate Modeling and Forecasting Machine learning based surrogate models are increasingly augmenting physics-based simulations. Systems developed by Google DeepMind demonstrate that neural architectures can approximate atmospheric dynamics at lower computational cost than traditional numer...

Date : February 6, 2026 Abstract : Reliable subnational population estimates are essential for effective governance, service provision, and humanitarian response, yet many low- and middle-income countries lack recent or comprehensive census data. This talk explores emerging strategies for estimating population counts when traditional sources are unavailable. I will focus on the integration of diverse and incomplete data sources—including administrative records (e.g., health, education), satellite imagery, and geospatial covariates—combined through Bayesian modeling frameworks to generate high-resolution estimates. In the second part of the talk, I will extend these methods to the problem of nowcasting population distributions in crisis contexts, including forced displacement and conflict. Here, I will explore the utility of real-time or near real-time data streams—such as mobile phone metadata, satellite-derived indicators, and social media signals—for capturing rapid demographic shift...

Throughout the 2024 US election season, as opinion polls predicted roughly even chances of either party winning, news media paid special attention to the odds on prediction markets such as Polymarket. More recently, prediction markets made headlines for a different reason. When Venezuelan President Maduro was captured in January 2026, the prediction market reacted quicker than the actual news, helping a lucky insider make $436,000 in a few hours. But what are some uses for scientists, besides making some (or half a million) quick bucks? Determining the Reproducibility of Scientific Studies Several studies have explored the use of market-based prediction of reproducibility to guide scientific funding and replication efforts. Dreber et al. (2015) created their own prediction markets with recruited participants during psychology's Reproducibility Project, effectively leveraging the wisdom of the crowd with real financial stakes. The predictions coming from the prediction market ...

In December 2025, I (Hamayoon Behmanush) had the opportunity to attend the Workshop on Artificial Intelligence with and for Learning Science (WAILS) in Cagliari, Italy, to present our work on Hope, Aspirations, and the Impact of LLMs on Female Programming Learners in Afghanistan . The workshop brought together researchers at the intersection of AI, education, and the learning sciences. Discussions explored what role AI should play in learning, and how to keep humans in the driver's seat as learning environments become increasingly intelligent.   Across the keynote talks, Beata Klebanov 's presentation, "Language Technology for Learning Modeling," examined what role AI should play in learning. She argued that language models can support the learning loop by relying on an explicit learning model. A learning model is a structured picture of what a learner is trying to do and how they're currently learning. In this view, a learner model captures key characteristics: g...