AI with AI: Artificial Intelligence with Andy Ilachinski

In COVID-related AI news, MIT researchers have published a machine learning algorithm that can diagnose COVID-19 by the sounds of a person's forced cough. And the US Veterans Affairs Department rolls out a machine learning tool to predict mortality rates of COVID-19 patients. In non-COVID news, the JAIC releases the Department of Defense's AI Education Strategy, which contains a detailed description of requirements, required instruction, and competencies. DoD also releases a new electromagnetic spectrum strategy, which contains a number of machine-learning mentions. And Tesla began making available its "full self-driving beta" to a small number of "expert and careful drivers." Research from MIT CSAIL have created a machine learning system that can reportedly decipher "lost" languages; they built it on several principles from insights into historical linguistics, such as the observation that languages generally only evolve in certain predictable ways (such as sound substitutions). In other language news, Facebook makes available a machine learning model that can translate directly between 100 different languages (rather than using English as a go-between). Research from CalTech and Purdue creates a "Fourier neural operator" that can solve parametric partial differential equations, nearly 1000 times faster than traditional solvers. And research from the University of Waterloo looks at "less than one-"shot learning, attempting to allow an AI to learn with almost no data (and thus recognize more objectives than the number of examples trained on).

Click here to visit our website and explore the links mentioned in the episode.

Sam Bendett joins Andy and Dave to discuss the latest developments and happenings in Russia's research into artificial intelligence and autonomy capabilities. They discuss Russia's national strategy and the challenges that have occurred in programmatic implementation due to COVID impacts. They also discuss the status of higher education in Russia and the standing of various institutions, as well as their relationship and interaction with the global community of researchers. They cover a variety of other trends and topics, including the Army 2020 convention and some of the announcements made during that event; and they discuss CNA's Russia Program and its on-going series of newsletters dedicated to summarizing the latest in Russian advances and research in AI.

Click here to visit our website and explore the links mentioned in the episode.

In COVID-related AI news, Andy and Dave discuss the COVID-19 Grand Challenge from C3.ai. In non-COVID AI news, the Department of Defense releases its Data Strategy. The Defense Science Board publishes a report on Counter Autonomy. The National Security Commission on AI releases its 3rd Quarter interim report and recommendations. The Center for Security and Emerging Technology releases a report on Building Trust through Testing. And the US Patent and Trademark Office publishes the responses to its initial queries, in Public Views on AI and Intellectual Property Policy. Researchers from MIT and Berkeley explore the idea that children's learning has analogy to hacking, by making code better through an open-ended set of goals and activities. Nathan Benaich and Ian Hogarth release the State of AI Report 202, which examines the latest developments in AI research across a variety of areas (such as observing that only 15% of papers publish their code). And Taylor and Dorin publish Rise of the Self-Replicators: Early Visions of Machines, AI and Robots that Reproduce and Evolve.

Click here to visit our website and explore the links mentioned in the episode.

Andy and Dave have a chat with Chad Jenkins, Professor of Computer Science and Engineering at the University of Michigan, Director of the Laboratory for Perception, RObotics, and Grounded REasoning SystemS (PROGRESS), and newest member of CNA's Board of Trustees. They discuss Chad's background and his current research at Michigan, which includes interactive robot systems and human-robot interaction. And then they discuss a variety of topics ranging from movement primitives, neural networks and fat tails, the issue of reinvention, students' experiences with AI research and the role of historical research, the culture of research in AI, and much more.

Click here to visit our website and explore the links mentioned in the episode.

The bipartisan Future of Defense Task Force releases its 2020 report, which includes the suggestion of using the Manhattan Project as a model to develop AI technologies. Facebook AI releases Dynabench as a way to dynamically benchmark the performance of machine learning algorithms. Amsterdam and Helsinki launch AI registers that explain how they use algorithms, in an effort to increase transparency. In research, the Allen Institute of AI, University of Washington, and University of North Carolina publish research on X-LXMERT (learning cross-modality encoder representations from transformers), which trains GPT-3 on both text and images, to then generate images from scratch by providing descriptions (e.g., a large clock tower in the middle of a town). Researchers at Swarthmore College and Los Alamos National Labs demonstrate the challenges that neural networks of various sizes have in learning Conway's Game of Life. Maria Jeansson, Claudio Sanna, and Antoine Cully create a stunning visual infographic on the "automated futures" technologies. And the Joshua Epstein, a longtime expert in agent-based modeling, provides the European Social Stimulation Association Award Keynote speech.

Click here to visit our website and explore the links mentioned in the episode.

In COVID-related AI news, Youyang Gu provides world- and county-level COVID-19 predictions using machine learning, along with a rolling examination of accuracy. In regular AI news, a military coalition of 13 countries meets to discuss the use of and ethics of AI. Orcan Intelligence provides a deeper look into Europeans' concerns about AI technologies. Ben Lee and the Library of Congress unveil the full open version of the Newspaper Navigator, which provides access to 1.56 million photographs from newspapers. Research from Intel and Binghamton University uses the pulse of the beating heart to identify deep fake videos with a 97% accuracy. And Arthur Holland Michel publishes the Black Box, Unlocked: Predictability and Understandability in Military AI.

Click here to visit our website and explore the links mentioned in the episode.

Andy and Dave kick off Season 4.0 of AI with AI with a discussion on social media bots. CNA colleagues Meg McBride and Kasey Stricklin join to discuss the results of their recent research efforts, in which they explored the national security implications of social media bots. They describe the types of activities that social media bots engage in (distributing, amplifying, distorting, hijacking, flooding, and fracturing), how these activities might evolve in the near future, the legal frameworks (or lack thereof), and the implications for US special operations forces and the broader national security community.

Click here to visit our website and explore the links mentioned in the episode.

In COVID-related AI news, another concerning report, this time in Nature Medicine, found "serious concerns" with 20,000 studies on AI systems in clinical trials, with many reporting only the best-case scenarios; in response, an international consortium has developed CONSORT-AI, reporting guidelines for clinical trials involving AI. In Nature, an open dataset provides a collection and overview of governmental interventions in response to COVID-19. In regular AI news, the DoD wraps up its 2020 AI Symposium. And the White House nominates USMC Maj. Gen. Groen to lead the JAIC. The latest report from the NIST shows that facial recognition technology still struggles to identify people of color. Portland, Oregon passes the toughest ban on facial recognition technology in the US. And The Guardian uses GPT-3 to generate some hype. In research, OpenAI demonstrates the ability to apply transformer-based language models to the task of automated theorem proving. Research from Berkeley, Columbia, and Chicago proposes a new test to measure a text model's multitask accuracy, with 16,000 multiple choice questions across 57 task areas. A report from AI Now takes a look at regulating biometrics, which includes tech such as facial recognition. And the 37th International Conference on Machine Learning makes its proceedings available online.

Click here to visit our website and explore the links mentioned in the episode.

In COVID-related AI news, a report from Cambridge University and University of Manchester examines recent studies on using chest x-rays and CTs scans to detect and diagnose COVID, and finds that only 29 of 168 studies had reproducible results; the report further found that all of the studies had high or unclear risk of bias, such that none of the studies had value for use in clinics. CSET provides an overview of how China has used AI in its COVID-19 response. In non-COVID AI news, a GAO report finds systemic problems with facial recognition technology at U.S. airports. The University College of London provides an overview of AI's use in crime, with deepfakes ranked as the most concerning. Researchers at the University of Warwick and the Alan Turing Institute develop a machine learning algorithm to identify potential planets from astronomy data. And NASA uses an algorithm to predict more accurately when hurricanes will rapidly intensify. In research, MIT, MIT-IBM Watson AI Lab, and Columbia University present a machine learning model to abstract relations in videos about everyday actions. Researchers in the Netherlands demonstrate that (large!) adversarial patches can work for surveillance imagery of military assets on the ground. The UN Interregional Crime and Justice Research Institute releases a Special Collection on AI. Researchers in Germany and Korea provide a view of continual and open-world learning. And Georgia Tech provides the People Map as a way to discover research expertise at an institution.

Click here to visit our website and explore the links mentioned in the episode.

In COVID-related AI news, Andy and Dave discuss an effort from Google and Harvard to provide county-level forecasts on COVID-19 for hospitals and first responders. The National Library of Medicine, National Center of Biotechnology Information, and NIH provide COVID-19 literature analysis with interesting data analytic and visualization tools. In regular AI news, Elon Musk demonstrates the latest iteration of Neuralink, complete with pig implantees. The UK attempted a prediction system for Most Serious Violence, but found that it had serious flaws. Amazon awards a $500k "Alexa Prize" to Emory University students for their Emora chatbot, which scored a 3.81 average rating across categories. The Bipartisan Policy Center releases two reports on AI. And Russell Kirsch, inventor of the pixel and other groundbreaking technology, passed away on 11 August at the age of 91. In research, three papers tackle the problem of reconstructing 3D (in some cases, 4D) models of locations based on tourist photos taking from different vantage points and at different times: the NeRF (Neural Radiance Fields) model and the Plenoptic model. The Human Rights Watch releases a report summarizing Country Positions on Banning Fully Autonomous Weapons and Retaining Human Control. Springer-Verlag releases yet-another-freebie with An Introduction to Ethics and Robotics in AI. And the Conference on Computer Vision & Pattern Recognition has posted the papers and videos from its June 2020 session.