This week’s list of data news highlights covers January 4-10, 2020, and includes articles about the White House releasing principles to regulate AI and a system that helps pilots decide if visibility is good enough for them to fly.
1. Creating Binding AI Principles for U.S. Regulators
The White House has released a draft of ten binding principles that federal regulators would need to follow when creating AI regulations. The proposal aims to remove barriers to the use of AI and includes principles that require agencies to use a risk-based approach to determine which risks are acceptable, to consider the benefits of AI applications before making regulations, and to provide ample opportunities for public comment on rule-making. The principles also require that agencies create flexible regulations that are performance-based instead of prescribing technical specifications.
2. Building a Smart City in Japan
Toyota is developing a smart city in Japan where it will test autonomous vehicles, robot-assisted living, and other smart city technology while providing residences to 2,000 individuals. The city will have autonomous delivery vehicles and will incorporate sensors and AI to perform tasks such as automatically restocking refrigerators and taking out a home’s trash. The first residents will be employees of Toyota and its project partners.
Researchers from the University of Michigan have developed an AI-enabled system that can diagnose brain tumors as accurate and faster than pathologists. The researchers trained the system using images of tissue from 415 brain surgery patients, teaching the system to identify common types of tumors by analyzing laser-generated imagery. The researchers tested the system on tissue from 278 patients and compared its performance to the performance of pathologists, finding that the system had an accuracy of 94.6 percent, compared to 93.9 percent for pathologists. The system also made its analysis in under two and a half minutes, compared to 30 minutes for the pathologists.
4. Using Data to Combat the Opioid Crisis
The U.S. Department of Health and Human Services has partnered with Millennium Health, a drug testing laboratory, to analyze near-real-time drug testing data to combat the opioid crisis. Millennium will share de-identified and aggregate data from urine drug tests. The data can help officials spot patterns and target interventions.
Police in Dayton, Ohio, have used audio sensors to track gunshots, allowing them to respond to incidents that are often not reported. The sensors use triangulation to pinpoint the location of a gunshot within 100 feet, and a machine learning system analyzes the audio to determine if a sound is actual gunfire, the type of weapon used, and if there are multiple shooters. Since its implementation in some neighborhoods in December, the system has already detected dozens of otherwise unreported gunshots, helped the police respond to a shooting victim, and led to the arrest of an armed suspect.
6. Waymo Autonomous Vehicles Have Driven 20 Million Miles On Public Roads
Autonomous vehicles from Waymo, a subsidiary of Alphabet, have driven over 20 million miles on public roads. It took more than a decade for Waymo’s fleet to drive its first 10 million miles, but it took the fleet little more than a year to drive the next 10 million. The firm’s fleet has also driven tens of billions of miles in computer simulations.
7. Using AI to Develop New Treatments for Cancer
Researchers led by an individual from Uppsala University in Sweden have developed an algorithm that can suggest new treatments to combat neuroblastoma, a form of cancer that often affects children younger than five. The algorithm analyzed genetic and pharmacological data from European and U.S. hospitals and universities, and it suggested a new treatment that involves activating a particular protein in the nervous system. The survival rate of cancer cells declined when the researchers tested the therapy using cell samples from patients and animals.
8. Measuring Vital Signs Over Video Chat
Binah.ai, a firm based in Israel, has developed AI-enabled technology that can estimate an individual’s vital signs using a video-based app. The technology analyzes an individual’s face to estimate their heart rate, stress level, oxygen saturation, and respiration. For example, the technology uses cameras to detect slight changes in facial coloring, which can indicate pulse.
9. Helping Pilots Decide If They Should Fly
A researcher from MIT has developed an algorithm that analyzes video feeds from U.S. Federal Aviation Administration cameras in Alaska to estimate an area’s visibility, which helps pilots decide if they should fly. The algorithm analyzes imagery from the previous ten days to establish how well it can detect persistent objects, such as the horizon, buildings, and mountainsides, in a “clear” image. The algorithm then compares its ability to detect objects in a clear image to the reference image to estimate visibility in miles.
10. Helping AI Analyze 3D Data
Researchers from the University of Amsterdam and Qualcomm have developed a framework for convolutional neural networks that helps the networks detect patterns in three-dimensional data, such as climate data mapped onto a sphere. Traditional convolutional neural networks can struggle to analyze 3D data because the curves can distort patterns, but the researchers’ method only allows patterns that have been converted to a common frame of reference in a consistent manner to pass through the network’s layers, thereby properly preserving patterns. The approach helped the neural network detect cyclones in climate data with nearly 98 percent accuracy, compared to 74 percent for a standard convolutional network.