This week’s list of data news highlights covers January 25-31, 2020, and includes articles about treating OCD and an AI system that predicted the outbreak of the coronavirus.
Covariant, a startup based in California, has developed an AI platform that allows robotic arms to pick up and move 95 percent of products with perfect accuracy. Covariant trained the platform’s algorithms both by manually guiding the robot to pick up objects and by conducting millions of trials, helping the robot learn techniques that are successful and fast. The platform is already in use in multiple warehouses.
2. Creating a Drug to Treat OCD
Exscientia, a biotechnology startup based in the UK, and Japanese pharmaceutical firm Sumitomo Dainippon Pharma have used AI to create a drug to treat OCD. The firms developed the drug, which will soon enter human clinical trials, by having an AI platform generate millions of potential molecules and decide which ones to synthesize and test. The AI system’s ability to learn which molecules could be successful allowed the firms to test a fifth of the normal number of compounds, helping the drug reach the point of clinical trials in one year instead of the typical four and a half.
3. Predicting the Development of the Coronavirus
BlueDot, a Canadian firm that monitors public health, has developed a system that used AI to predict the outbreak of the coronavirus a week before the U.S. Centers for Disease Control and Prevention issued its alert. The system uses natural language processing and other machine learning techniques to analyze news reports, airline data, and reports of animal disease outbreaks. The system also correctly predicted the disease would spread to Bangkok, Seoul, Taipei, and Tokyo after its initial appearance in the Chinese city of Wuhan.
Researchers from the University of Southern California and the University of California, Los Angeles, have shown that AI can accurately detect changes in patients’ mental health by analyzing audio of patients speaking. The researchers used a machine learning algorithm to analyze hundreds of patient voicemails, and the algorithm providing ratings of the patients’ mental health that matched clinicians’ ratings. The algorithm can help track how a patient’s condition changes over time.
Researchers led by an individual from the French National Center for Scientific Research attached sensors to albatrosses to discover that more than a quarter of vessels in the Indian Ocean could be illegally fishing. Boats use an identification system to avoid collisions with other vessels and to allow authorities to track their movements, but vessels that engage in illegal fishing frequently turn the system off. The sensors on the albatrosses identified boats within 30 kilometers of the birds, and the researchers compared the boats’ locations to data from the identification system, finding that 25.8 percent had shut the system off.
6. Predicting and Treating Sepsis
The Cleveland Clinic is using a machine learning system to predict which patients will develop sepsis, a condition that can lead to multiple organ failure and death. The system analyzes data from electronic health records to identify patterns that indicate patients have a high-risk profile for sepsis, such as the use of medications that raise blood pressure. The system also makes recommendations to clinicians, including blood monitoring and administering medications.
7. Predicting the Number of Icebergs
Researchers from the University of Sheffield in the UK have developed a forecasting model that uses AI to predict the number of icebergs that will enter shipping lanes in a year. The model analyzes data about ocean surface temperatures, atmospheric pressure, and the Greenland ice sheet’s accumulation or loss of mass on its surface. The researchers tested the model on data from 1997 and 2016, finding that it was 80 percent accurate.
8. Helping Blind Individuals Find Familiar Faces
Researchers from Microsoft have developed a headband that uses cameras, a depth sensor, speakers, and AI to help visually impaired individuals better understand their surroundings. The headband uses facial recognition to identify people the user knows and click sounds to alert the user to a person’s location; for example, if an individual is approaching the user from the left, the click will sound as if it’s coming from the left. The tool can help users detect the position of people in their environment, which can help them maintain social interactions.
Researchers from Google have created a chatbot named Meena that outperforms existing state-of-the-art chatbots at engaging in conversations. The researchers trained Meena on 341 gigabytes of text from public domain social media conversations. They also tested multiple chatbots and humans on their ability to provide responses that made sense and were specific to a conversation topic, finding that only humans (86 percent) scored higher than Meena (79 percent).
Harrison.ai, an Australian firm that creates AI-enabled tools for healthcare, has developed a system that uses AI to select embryos with the highest chance of resulting in a successful pregnancy during in vitro fertilization. The system analyzes time-lapse videos of fertilized eggs during incubation to assess their growth and make selections. The system has helped choose embryos for several thousand women in Australia.
Image: Will Kennard