Andy and Dave discuss the latest in AI news and research, including:
[1:28] Researchers from several universities in biomedicine establish the AIMe registry, a community-driven reporting platform for providing information and standards of AI research in biomedicine.
[4:15] Reuters publishes a report with insight into examples at Google, Microsoft, and IBM, where ethics reviews have curbed or canceled projects.
[8:11] Researchers at the University of Tübingen create an AI method for significantly accelerating super-resolution microscopy, which makes heavy use of synthetic training data.
[13:21] The US Navy establishes Task Force 59 in the Middle East, which will focus on the incorporation of unmanned and AI systems into naval operations.
[15:44] The Department of Commerce establishes the National AI Advisory Committee, in accordance with the National AI Initiative Act of 2020.
[19:02] Jess Whittlestone and Jack Clark publish a white paper on Why and How Governments Should Monitor AI Development, with predictions into the types of problems that will occur with inaction.
[19:02] The Center for Security and Emerging Technology publishes a series of data-snapshots related to AI research, from over 105 million publications.
[23:53] In research, Google Research, Brain Team, and University of Montreal take a broad look at deep reinforcement learning research and find discrepancies between conclusions drawn from point estimates (fewer runs, due to high computational costs) versus more thorough statistical analysis, calling for a change in how to evaluate performance in deep RL.
[30:13] Quebec AI Institute publishes a survey of post-hoc interpretability on neural natural language processing.
[31:39] MIT Technology Review dedicates its Sep/Oct 2021 issues to The Mind, with articles all about the brain.
[32:05] Katy Borner publishes Atlas of Forecasts: Modeling and Mapping Desirable Futures, showing how models, maps, and forecasts inform decision-making in education, science, technology, and policy-making.
[33:16] DeepMind in collaboration with University College London offers a comprehensive introduction to modern reinforcement learning, with 13 lectures (~1.5 hours each) on the topic.
Follow the link below to visit our website and explore the links mentioned in the episode.