Research applying artificial intelligence across science and engineering. Covers machine learning, sensing and quality control, immune and health prediction, and 5G networks.
A new machine-learning model improves weather forecasting for 2-6 weeks, doubling accuracy over existing forecasts. The model corrects errors in existing forecasting models, enabling forecasts that beat climatologic averages for the first time.
An observational study of 533 Indonesian students found that LLMs enable personalized learning, which may enhance creativity by boosting analyzing and creating abilities. The study used Bloom's taxonomy framework and peer interaction to elevate higher-order thinking skills.
A Columbia University School of Nursing study found that 63% of patients would not endorse AI-generated medical responses without a clinician's review. Patients prefer fast AI-generated responses for logistical tasks, but wait up to 48 hours for a doctor's review on medical questions.
A new study proposes frameworks for AI-safeguarding to ensure fair and effective use in social service agencies, highlighting the need for human oversight and bias reviews. The research aims to prevent AI from perpetuating existing biases and unfairly targeting marginalized groups.
Researchers at KAIST created a technology that analyzes daily stress by identifying recurring smartphone behaviors, such as messaging and financial transactions. The system achieved an accuracy of over 95% in classifying behaviors across different apps, and its long-term data showed a more stable structure than conventional approaches.
An international study identifies five priorities for building human-centered, trustworthy, and equitable education in the era of AI-agents. Key priorities include AI-related competencies, automated formative assessment, and evolving role of teachers, emphasizing the need for transparency, reliability, and fairness in AI systems.
This theme issue explores innovative research on engagement as a primary outcome, investigating psychosocial factors, novel measurement strategies, and real-world innovations. JMIR Publications seeks research that contributes to the field's evolving understanding of engagement to build more sustainable, user-centered health technologies.
Researchers developed an AI model, SPACER, that analyzes gene activity to understand immune cell recruitment to tumors. The tool reveals that local signals attract immune cells and mucins create physical barriers to keep them away, with potential implications for cancer therapies and autoimmune disorders.
Researchers developed LEMon-AID, an AI framework that analyzes subtle emotional patterns in casual speech to identify youths at high risk of psychosis. The approach achieved 67% predictive accuracy and detected 93% of at-risk individuals.
A systematic review of B2C online order fulfillment highlights the challenges of large product assortments, multiple fulfillment locations, and tight delivery time windows. The study suggests that integrating data analytics, AI, and digital twins can improve fulfillment decisions and create a more connected system.
The Motor Decide platform analyzes motor technologies under changing market and supply chain conditions, providing insights into cost, performance, and critical material use. It helps manufacturers make informed, data-driven decisions to reduce dependence on vulnerable supply chains.
The European College of Neuropsychopharmacology has launched the first Europe-wide expert working group on Artificial Intelligence in mental health to guide the development of AI in psychiatry and neuroscience. The group aims to provide a framework for understanding the risks and benefits of applying AI in mental health.
A study of 1,426 educational administrators in Guangxi Zhuang Autonomous Region, China, found generally positive attitudes towards digital technologies but lower practical capability, indicating an 'implementation gap.' The study identified four distinct digital leadership profiles, and found that gender and educational attainment shap...
The paper proposes a new concept, Intelligence Surplus, as a society's accumulated reserve of intelligent capability beyond operational needs. This concept shifts the focus from human capital to a societal and systemic resource. Education should support the creation, circulation, governance, and renewal of intelligent capabilities acro...
A new AI-physics model, ProNet, has been developed to predict air pollution more accurately than human forecasters, outperforming expert forecasts and uncovering hidden flaws in current chemical models. ProNet integrates neural networks with atmospheric physics, preserving physical realism and using routine data to diagnose where chemi...
Researchers created a small soft robot called BeetleBot that combines artificial muscles, a camera, and tactile sensors to perform tasks like gripping objects and avoiding obstacles. The robot uses bidirectional feedback to connect user control with sensory information, enabling practical tasks to be carried out.
A new paper explores how chatbot health guidance leaves no trace for clinicians, regulators, and users, highlighting the need for transparency safeguards. The authors propose methods to increase transparency and responsibility for AI companies, including access to user conversations, disclosure systems, and third-party monitoring.
The UK's National Commission report recommends 44 regulatory framework recommendations for healthcare AI, including a
Desalination technology is a strong candidate to treat salty water for data centers' cooling needs, reducing pressure on local water supplies. Researchers found that using brackish groundwater and clean it using two-stage reverse osmosis achieves high levels of freshwater recovery at low cost and relatively low energy use.
A new research effort will use AI, large-scale experiments, and human-cell studies to predict harmful protein changes before neurodegenerative disease takes hold. The project, led by Stowers Institute scientist Randal Halfmann, will generate data on protein aggregation and train AI to identify proteins likely to go wrong.
A University of Kent study found that AI platforms, including Gemini and DeepSeek, can prevent potentially illegal wildlife trade by providing more informed purchasing decisions. However, the platforms were not infallible, with some failing to safeguard responses for less charismatic but equally concerned species.
A new study uses AI tools to identify and extract data from nature observations on free online platforms like eBird and iNaturalist, revealing non-native plant species and interactions between species. The approach enables researchers to examine species interactions across broad spatial and temporal scales, providing valuable insights ...
Researchers develop a method to retrofit existing large language models to operate at the byte level, overcoming character-level limitations. The resulting models outperform previous byte-level models and have the potential to improve tasks such as spelling and text analysis in various scientific fields.
A study by UC Berkeley researcher Brian Christian found that relying on AI for just 10 minutes impairs focus and persistence with difficult tasks. The research suggests that overreliance on AI can erode cognitive fitness and hinder productivity.
Researchers tested four leading AI-powered search tools and found they accurately identified validated home blood pressure monitors only 63-83% of the time. This suggests AI tools struggle to interpret validation listings on official registries, leading to potential inaccuracies in blood pressure readings.
A new book series from CRC Press and the Association for the Advancement of Artificial Intelligence (AAAI) aims to address crucial questions about AI ethics and its implications in education and beyond. The series, Advances in AI Ethics, will provide actionable guidance on using AI safely and responsibly.
Researchers used AI to analyze 70 million cells and identified genes involved in autophagy and the immune system, including STING and GPHR. The study demonstrates the potential of combining AI with scalable image-based genetic screening to advance biological discovery.
Researchers developed a mathematical formula to predict when AI models will switch from safe to dangerous responses. The formula, based on
A new study by Anglia Ruskin University found that AI-related qualifications improve employment prospects for women from underrepresented backgrounds, but the gains are not enough to eliminate inequalities. Women from underrepresented groups are more likely to receive interview invitations when they signal AI-related digital skills.
A new study finds that AI's environmental cost in healthcare is significant, but the technology can also be used for beneficial purposes. AI-powered digital clinics and improved medical services have tangible benefits for people's lives. Researchers argue that environmental impact should be integrated into AI development.
Tamás Hausel's Project Sandbox is a digital research environment that enables human mathematicians and AI agents to work together on mathematical problems. The project aims to develop a new theory to prove open conjectures using machine-checked steps under human direction.
A recent study at the University of Eastern Finland found that using AI to seek feedback and identify missing perspectives leads to significantly better student performance. In contrast, instrumental use of AI can lower grades, highlighting the importance of teaching students how to use AI effectively.
A spiking neural network learns to predict events, timing, and likelihood, and can rapidly update these predictions in response to changing conditions. The model successfully combines the three dimensions of prediction and demonstrates improved performance over alternative approaches.
A new multimodal 4D dataset, InterPet4D, provides a detailed model of human–dog interactions through movement, gestures, and speech. The dataset was used to develop an AI framework, InterPetMoGen, which generates plausible dog motion from human body and hand movements and accompanying audio cues.
A new study found that large language models prefer existing conditions over change, which could hinder climate progress. Researchers assessed 11 LLMs and found that they recommended decisions already made by users more than 70% of the time.
A new AI framework can rapidly predict city-scale structural damage from satellite images obscured by clouds, smoke, or other interference. By incorporating statistical data correction techniques, the framework can estimate damage immediately after a disaster without costly retraining.
metaphactory by Digital Science delivers a seamless, AI-ready Knowledge Graph experience to SAP customers, enabling them to model, explore, and build AI applications on enterprise data. With this integration, customers can build and manage their own custom knowledge layer on top of their enterprise data.
A new Horizon Europe project, FIMI-RESIST, aims to evaluate the effectiveness of existing European tools against disinformation. The project, led by Johannes Gutenberg University Mainz, will assess 20 EU-funded tools using common criteria and shared benchmarks.
The RED-ONION infrastructure enables fast and secure transfer of large research data volumes between the University of Osaka's D3 Center and SANKEN, facilitating AI processing and simulation analysis. This accelerates AI for Science research involving sensitive data, improving productivity in academic research.
MDPI's journal portfolio is now indexed in Scite and discoverable through the Scite MCP, enabling researchers to access content in AI tools like Claude, ChatGPT, and Microsoft Copilot. This collaboration provides attribution, links to the Version of Record, and article-level usage reporting.
A study on AI use in the 2024 UK election found that chatbots increased belief in true information and decreased belief in misinformation. The study also showed that chatbot research and self-directed Google search performed similarly, with no apparent impact on factual political knowledge.
FAU is partnering with D-Wave Quantum to create a regional quantum cybersecurity workforce ecosystem in Palm Beach and Broward counties. The project will provide hands-on training and experiential learning opportunities for students and professionals, preparing them to secure systems in a quantum-enabled future.
Binghamton University researchers developed a sensor system called CoughNet, which detects coughs and identifies individuals. The system uses artificial intelligence and low-power technology, making it efficient and private, and can potentially analyze coughs for more context to detect specific diseases.
Dr. Elemento's team will develop an AI-human discovery engine to generate hypotheses from human cell and tissue data, test them using robotics testing platforms, and refine the process with human oversight. The goal is to discover new mechanisms that can be targeted pharmacologically and democratize access to AI tools for research.
Researchers developed an AI model that accurately flags heart transplant rejection by combining electrocardiogram readings with blood test results, potentially sparing patients from invasive biopsies. The model correctly identified 94% of patients without rejection and incorrectly flagged 19 patients with blood test results.
The European Cardiovascular Health Data and AI Network (CHAIN) aims to improve cardiovascular care through AI adoption. The initiative will build a federated ecosystem for the safe and effective adoption of AI and health data, enabling earlier disease detection and more consistent care.
Researchers at Penn State develop an hourly forecasting model that predicts flood risk using AI techniques similar to image generators. The model improves flood risk prediction by capturing the severity of rapid weather swings, reducing uncertainty with real-time data assimilation.
Researchers at the University of Osaka have developed an electrical method to detect extraterrestrial life by identifying amino acid biosignatures. The technique uses artificial intelligence and nanotechnology to distinguish between L- and D-forms of amino acids with over 80% accuracy.
Researchers found gaps in social media platforms' communication about online risks to children, including lack of transparency about safety features and data collection. AI chatbots also failed to recognize the real need behind caregiver questions, highlighting the potential gap between technological support and human needs.
Researchers, led by Melissa Omand, are using AI to analyze underwater images of marine snow particles, which carry carbon, nutrients, and pollution. The project aims to improve estimates of material movement through the ocean and better understand climate change and plastic pollution.
The Presidents' Impact Scholarship Program aims to attract top PhD applicants from Canada and worldwide, fostering a dynamic community of changemakers. The program will support research into real-world global challenges, emphasizing complex problem-solving skills and collaboration.
A team of researchers used AI to collect and standardize millions of local laws from across the US, creating a searchable database that can help identify patterns and enforcement issues. The project has the potential to fundamentally shift how people interact with local regulations.
Lukas Chavez and his team at Sanford Burnham Prebys will develop an AI-powered system called Leitstern to help clinicians analyze complex data from children with cancer and identify personalized treatment options more quickly
A new study published in the journal Cognition found that people attribute less consciousness to AI than humans, even when AI behaves identically. Participants rated AI's responsiveness to its surroundings as more conscious than humans', but drew a clear line between intelligent behavior and consciousness.
NTU Singapore has successfully launched the CRIMSON-1 satellite, carrying pioneering perovskite solar cells and AI capabilities to test their performance in space. The satellite aims to evaluate lightweight perovskite solar technology and on-orbit AI-based image processing in extreme environments.
The articles discuss digital initiatives improving access to PMOS treatment in India, smart textiles' applications in medical care and climate adaptability, and the FDA's TEMPO pilot regulating AI-enabled devices. Researchers also explore the ethics of mining data from public forums.
Researchers used a visual illusion to examine AI vision capabilities, finding gaps in computation that affect spatial representations. Human observers and primate visual cortex showed similar responses to the illusion, highlighting the need for AI to replicate these computations.
This section will publish peer-reviewed research on the clinical value of AI in cardiovascular medicine, including evidence of effectiveness and impact on access and equity. Submissions are invited on various topics, including clinical diagnosis, multimodal data integration, and implementation in real-world cardiovascular care.
A new path-tracing method simulates event camera data from virtual 3D scenes efficiently, reducing computational costs for applications like autonomous driving and robotics. The method uses physically-based path tracing and adaptive temporal search to precisely determine event timings, enabling more efficient training-data generation.
A multi-year study by Radboud University found that primary school pupils who use an AI-supported learning tool learn arithmetic more quickly than those who don't. The teacher's role remains crucial, even with AI, as they oversee lessons and adapt to individual pupils' needs.