Dr. T's AI brief
dtau...@gmail.com
Precise Touchscreens, Thanks to AI
ETH Zurich (Switzerland)
May 12, 2021
A new artificial intelligence (AI) technique developed by computer scientists at ETH Zurich more precisely estimates where a finger touches a mobile phone screen, to reduce typing errors. ETH's Christian Holz explained the CapContact AI “estimates the actual contact areas between fingers and touchscreens upon touch,” then “generates these contact areas at eight times the resolution of current touch sensors, enabling our touch devices to detect touch much more precisely." The researchers found their novel deep learning approach eliminates low-resolution input sensing errors, which they said are responsible for a third of errors on current touchscreen devices.
Ayanna Howard Named ACM Athena Lecturer
ACM
May 12, 2021
ACM named Ayanna Howard, dean of the Ohio State University College of Engineering, its 2021-2022 ACM Athena Lecturer for her contributions to the development of accessible human-robotic systems and artificial intelligence (AI), and for boosting participation in computing. Howard proposed some of the first concepts for simulating deformable objects via physical modeling, to enable robust robot grasping; she also introduced the modeling of environmental uncertainty through fuzzy logic, furthering the state of the art in field robotics. Howard also has spearheaded modeling trust among humans, robots, and AI systems, including conversational agents, emergency response situations, autonomous navigation, child-robot interaction, and use of lethal force. ACM president Gabriele Kotsis said, "Both as an entrepreneur and mentor, Ayanna Howard has worked to increase the participation of women and underrepresented groups in computing."
AI, Drones Will Help Pin Down Sosnovsky's Hogweed
Skolkovo Institute of Science and Technology (Russia)
May 11, 2021
To help contain the spread of Sosnovsky's hogweed, a plant hazardous to agriculture, local ecosystems, and human health, across Russia, scientists at the Skolkovo Institute of Science and Technology (Skoltech) have developed an artificial intelligence monitoring system that performs real-time image segmentation onboard drones to identify the toxic weed. The system uses drones that can capture high-resolution images even in cloudy weather via in-flight data acquisition and processing. Each drone's computer runs heavy segmentation algorithms based on Fully Convolutional Neural Networks (FCNN) that can identify an irregularly shaped object on a pixel-by-pixel basis. The researchers utilized popular architectures for the FCNN and adapted them for a single-board computer. Skoltech's Andrey Somov said, "We installed and flight-tested our monitoring system on board the drone which covered an area of up to 28 hectares in 40 minutes, flying at an altitude of 10 meters. And it did not miss a single weed!"
Patients May Not Take Advice from AI Doctors Who Know Their Names
Penn State News
Matt Swayne
May 10, 2021
Patients may be less inclined to heed artificial intelligence (AI) doctors that know their names and medical history, according to researchers at Pennsylvania State University (Penn State) and the University of California, Santa Barbara. The team designed five AI chatbots; Penn State's Jin Chen said the bots were programmed to ask questions about COVID-19 symptoms and behaviors, and to offer diagnosis and recommendations. Study participants were more likely to consider a chatbot intrusive and less likely to follow its medical advice when it used their first name and referred to their medical history, yet they expected human doctors to distinguish them from other patients, and were less likely to comply when a clinician did not recall their information. Penn State's S. Shyam Sundar said, "When an AI system recognizes a person's uniqueness, it comes across as intrusive, echoing larger concerns with AI in society."
AI Consumes a Lot of Energy. Hackers Could Make It Consume More.
MIT Technology Review
Karen Hao
May 6, 2021
Maryland Cybersecurity Center (MC2) researchers have outlined an attack that could boost the energy consumption of artificial intelligence (AI) systems by forcing a deep neural network to overuse computational resources. The team added small amounts of noise to the inputs of an input-adaptive multi-exit neural network, which were perceived as more difficult, increasing computation that required more energy to complete. In assuming the attacker had full data about the network, the researchers could max out its energy draw; in assuming the attacker had little to no data, they could still slow processing and increase energy consumption 20% to 80%. This hack remains somewhat theoretical, but MC2's Tudor Dumitras said, "What's important to me is to bring to people's attention the fact that this is a new threat model, and these kinds of attacks can be done."
Lab Launches Free Library of Virtual, AI-Calculated Organic Compounds
University of Toronto (Canada)
Dan Haves
May 5, 2021
Researchers at Canada's University of Toronto (U of T) have launched a free open access tool containing a library of 330,000 virtual machine learning-calculated organic compounds to accelerate catalysis science. The Kraken tool features organophosphorus ligands, and U of T's Théophile Gaudin said the team created a Web application "where users can search for ligands and their properties in a straightforward manner." Their hope is that the library will allow chemists to reduce the number of trials needed to realize optimal results in their work. U of T's Alan Aspuru-Guzik said, “The world has no time for science as usual; neither for science done in a silo. This is a collaborative effort to accelerate catalysis science that involves a very exciting team from academia and industry.”