AI with AI
Episode 3.24: Never Give Up
In COVID-related AI topics, Andy and Dave discuss an emerging crop (no less than three!) of COVID-19 cough detectors that attempt to diagnose the presence of COVID by various voice measurements. In a similar vein, but for different purposes, the U.S. drone maker Draganfly announces it is working with the Australian Department of Defence to produce “pandemic drones,” which can detect coughing, sneezing, and respiratory rate at a difference. Folding@home has shifted its crowdsourcing computational power toward the COVID-19 problem set. In non-COVID news items, researchers at the University of California San Francisco have used deep learning algorithms to translate human brain signals for a set of 250 unique words, by recording brain signals for sentences as patients read them. In research, Uber AI and OpenAI announces their Enhanced POET (Pair Open-ended Trailblazer), which uses a procedural environment to create problems (gaps, stumps, stairs), which the agent then learns to solve, producing a diverse range of sophisticated behaviors. DeepMind reveals Agent57, the first reinforcement learning agent capable of surpassing the human benchmark for all 57 Atari games (though it still must be trained on each individually), using Never Give Up (NGU) memory to identify new environments, as well encouraging exploration and other components. The Survey of the Week takes a look at the development of deep learning for scientific discovery. A report from the BMJ suggests that studies claiming that AI outperforms doctors are “arguably exaggerated,” with a high risk of bias identified in 58 out of 81 studies. A New Conception of War, by Ian Brown, makes the Free Book of the Week, coming from the Marine Corps University Press; among many important concepts, it stresses the importance of debate and intellectual exploration among professional warfighters. Johns Hopkins APL is hosting a virtual event on Operationalizing AI in Health on 21 April. And Intelligent Heath Inspired! seeks to hold the largest summit on 25-27 May on the use of AI in medicine, with a particular focus on COVID-19.