This week’s list of data news highlights covers September 5, 2020 – September 11, 2020 and includes articles about using AI to improve the mental health of veterans and using drones and machine learning to prevent future wildfires.
The Africa Center for Disease Control (Africa CDC) has partnered with Novetta, an analytics company, to use machine learning to identify sources contributing to the spread of misinformation about COVID-19. The machine learning platform combines data from social media with domestic and international news media to geographically pinpoint where misinformation is coming from and to track events that may be contributing to its spread. The platform also uses natural language processing to analyze whether mistrust of local government or other countries is driving posts and tags them as either corroborating or contradicting local and global public health measures. The Africa CDC then uses the dataset of tagged posts to tailor COVID-19 guidelines to each country.
The Guardian, a UK-based newspaper, used GPT-3, a language model that uses machine learning to produce human-like text, to write an op-ed article about AI. The Guardian gave GPT-3 instructions about what to write, including how long the op-ed should be, the type of language to use, and what sentences to begin with. GPT-3 produced eight separate essays, which The Guardian edited into one final op-ed.
The U.S. Department of Veteran’s Affairs (VA) has partnered with Medallia, a customer experience management platform, to launch a program that uses AI to monitor the tone and language of veterans during crisis hotline calls. From the data collected during the call, the AI system identifies who is at-risk and routes calls to local VA offices based on priority. Since launching, the program has provided 1,400 veterans with early intervention care.
In California, fires have burned over 3 million acres and killed at least 11 people since Monday. To reduce the risk of future wildfires, gas company PG&E and other major utility companies in the state are using drones and machine learning to spot faulty equipment that could start new fires. Drone cameras take pictures of transmission towers, attached equipment, and the surrounding areas and machine learning tools analyze these images to identify any changes or abnormalities to equipment, such as corrosion or damage. This means utility companies can relieve staff of maintenance work in high-risk areas and reduce the risk that damaged gear could start fires.
The Trevor Project, an LGBTQ suicide prevention organization in the United States, is using Google’s natural language processing model ALBERT to analyze chat messages for individuals at high risk for self-harm on its crisis text-line and chat platforms. If the system flags a chat message as high-risk, the person is placed higher in the queue to speak with a human counselor. By 2024, the organization aims to serve the 1.8 million LGBTQ youths who are at risk of committing suicide each year.
Dubai’s Road and Transport Authority (RTA) is using AI and machine learning algorithms to plan bus routes that save time for riders. The RTA utilized data from the electronic ticketing cards used on public transport and analyzed which bus stops were busy all-day, experienced peak hours, or were hardly used. Based on the analysis, the RTA created express bus routes that skipped or eliminated specific stops. In a trial run, buses along 10 express routes saved 13 percent of the time that would have been wasted at unnecessary stops.
Researchers from the University of Cambridge, IBM Research, and EPFL, a research university in Switzerland, used machine learning to understand how hydrogen becomes a metal inside giant planets. The researchers combined machine learning and quantum mechanics to mimic the interactions of hydrogen atoms. Under extreme pressure conditions, the researchers discovered that hydrogen changes smoothly and gradually into a dense metallic hydrogen, which behaves like an electrical conductor.
Theoretical chemists at Linköping University in Sweden have created a new molecule that can capture and store solar energy. Using a supercomputer, researchers simulated the solar molecule undergoing chemical reactions. In one reaction, the supercomputer showed the solar molecule absorbing energy from sunlight and remaining energy-rich and stable within 200 femtoseconds, which is equivalent to a quadrillionth of a second.
Researchers from Boston University and the Massachusetts Institute of Technology have developed two AI models to reduce hospital readmissions. One model predicts the likelihood of a patient being readmitted within a 30-day period. To build this model, researchers used data from 722,101 patients and analyzed how factors such as underlying health conditions and postoperative complications impacted patient readmission. The second model offers patients’ personalized treatment recommendations based on their likelihood for readmission due to these factors, such as blood transfusions to increase a patient’s red blood cell count before surgery.
Researchers from the University of Toronto in Canada and Carnegie Mellon University are using AI to transform waste carbon into a commercially valuable product. The researchers applied a machine learning model to over 240 different materials in order to find the quickest chemical catalyst to convert carbon dioxide to ethylene, a chemical precursor for products like dish detergent. Researchers discovered the best catalyst to be a mixture of copper and aluminum.