AI with AI

In COVID-related news, Andy and Dave discuss a commercial AI model from Biocogniv that predicts COVID-19 infection using only blood tests, with a 95% sensitivity and a 49% specificity. In a story that highlights the general challenge with algorithms, Stanford reported challenges in using a rules-based algorithm to determine priority of vaccine distribution, when it omitted front-line doctors from initial distribution. In non-COVID AI news, Vincent Boucher and Gary Marcus organize a second “AI Debate” on the topic of Moving AI Forward: An Interdisciplinary Approach, which included Daniel Kahneman, Christof Koch, Judea Pearl, Fei-Fei Li, Margaret Mitchel, and many others. Reuters reports that Google’s PR, policy, and legal teams have been editing AI research papers in order to give them a more positive tone, and to reduce discussions of the potential drawbacks of the technology. And Microsoft patents a “chat bot technology” that would seek to reincarnate deceased people. In research, Google announces MuZero, which masters chess, Go, shogi, and the Atari Learning Environment by planning with a learned model (and no information on the rules). Jeff Heaton provides the book of the week, with Applications of Deep Neural Networks. A survey paper from four universities looks at Data Security for Machine Learning. Another survey paper examines how researchers develop and use datasets for machine learning research. And the ConwayLife.com community celebrates the 50th anniversary of the Game of Life, to include an online simulator called the Exploratorium.