On Our Radar

Home 5 On Our Radar

What’s on Our Radar: Top News and Must-Read Articles from UK and Beyond

Articles

“The Emergence of a Data Winter: The Growing Enclosure of Data at a Time of Rapid AI Advances” argues that society is entering a “data winter” due to increasing data hoarding and restricted access, which threatens evidence‑based policymaking and equitable AI development.
“Weight-sparse transformers have interpretable circuits” shows that training transformers with mostly zero weights produces more human-understandable internal circuits, revealing how neurons and connections correspond to natural concepts while balancing interpretability and model capability.
A Definition of AGI” — The authors propose a quantifiable framework grounded in human‑cognition theory (Cattell‑Horn‑Carroll) to define and assess Artificial General Intelligence in terms of breadth and depth of abilities.
“Why Generative Artificial Intelligence Does Not Survive First Contact with Bureaucracy” finds that GenAI outputs in public institutions are reshaped by bureaucratic and political processes, making AI insights malleable and dependent on institutional negotiation rather than purely computational logic.
“Measuring what Matters: Construct Validity in Large Language Model Benchmarks” reviews 445 LLM benchmarks and finds that many measures lack validity, offering eight recommendations to improve assessment of safety, robustness, and model capabilities.
“Beyond the Digital Divide: Unlocking Urban Economic Resilience through Integrated Digital Infrastructure and Finance” finds that coordinated development of digital infrastructure and digital finance significantly enhances urban economic resilience, particularly by supporting industry upgrading, innovation, and entrepreneurship.
“Humans and Neural Networks Show Similar Patterns of Transfer and Interference During Continual Learning” shows that patterns of learning transfer and interference in humans and neural networks align closely when exposed to sequential tasks, suggesting shared learning constraint.
“Cultural intelligence and its relationship to digital citizenship skills among university students” — The study finds that higher cultural intelligence among Saudi university students is significantly associated with stronger digital citizenship skills, highlighting the importance of education and awareness programs in developing both.
“Institutional pressure, resource orchestration, and the digital government performance: a mixed methods study in the case of China” shows how Chinese local governments orchestrate resources under multiple institutional pressures to enhance digital government performance and gain greater autonomy in shaping digital initiatives.
“Modeling public trust in AI cognitive capabilities using statistical and machine learning approaches” finds that familiarity with AI, age, and gender influence public trust in AI’s cognitive tasks, with greater trust in memory-based tasks and preference for human oversight in high-stakes decisions.
“Fair human-centric image dataset for ethical AI benchmarking” introduces FHIBE, a publicly available human image dataset designed with consent, diversity, and privacy safeguards to enable ethical evaluation and bias mitigation in computer vision AI tasks.
“Balancing innovation and integrity: AI in tax administration and taxpayer rights” examines how AI can improve tax administration efficiency while highlighting risks to transparency and fairness, proposing independent oversight to protect taxpayer rights and ensure accountability.