AI with AI: Artificial Intelligence with Andy Ilachinski

Andy and Dave discuss the latest in AI news, including an upgraded version of OpenAI's CoPilot, called, Codex, which can not only complete code but creates it as well (based on natural language inputs from its users). The National Science Foundation is providing $220 million in grants to 11 new National AI Research Institutes (including two fully funded by the NSF). A new DARPA program seeks to explore how AI systems can share their experiences with each other, in Shared-Experience Lifelong Learning (ShELL). The Senate Committee on Homeland Security and Governmental Affairs introduces two AI-related bills: the AI Training Act (to establish a training program to educate the federal acquisition workforce), and the Deepfake Task Force Act (to task DHS to produce a coordinated plan on how a "digital content provenance" standard might assist with decreasing the spread of deepfakes). And the Inspectors General of the NSA and DoD partner to conduct a joint evaluation of NSA's integration of AI into signals intelligence efforts. In research, DeepMind creates the Perceiver IO architecture, which works across a wide variety of input and output spaces, challenging the idea that different kinds of data need different neural network architectures. DeepMind also publishes PonderNet, which learns to adapt the amount of computation based on the complexity of the problem (rather than the size of the inputs). Research from MIT uses the corpus of US patents to predict the rate of technological improvements for all technologies. The European Parliamentary Research Service publishes a report on Innovative Technologies Shaping the 2040 Battlefield. Quanta Magazine publishes an interview with Melanie Mitchell, which includes a deeper discussion on her research in analogies. And Springer-Verlag makes available for free An Introduction to Ethics in Robotics and AI (by Christoph Bartneck, Christoph Lütge, Alan Wagner, and Sean Welsh).

Follow the link below to visit our website and explore the links mentioned in the episode.

Andy and Dave welcome the hosts of the weekly podcast AI Today, Kathleen Walch and Ronald Schmelzer. On AI Today, Kathleen and Ron discuss topics related to how AI is making impacts around the globe, with a focus on having discussions with industry and business leaders to get their thoughts and perspectives on AI technologies, applications, and implementation challenges. Ron and Kathleen also co-founded Cognilytica, an AI research, education, and advisory firm. The four podcast hosts discuss a variety of topics, including the origins of the AI Today podcast, AI trends in industry and business, AI winters, and the importance of education.

Related Links

CPMAI Methodology: https://www.cognilytica.com/cpmai/

Cognilytica website: https://www.cognilytica.com/

AI in Government community: https://www.aiingovernment.com/

Cognilytica: @Cognilytica

Kathleen Walch: @Kath0134

Ron Schmelzer: @rschmelzer

Andy and Dave discuss the latest in AI news, including a story from MIT Technology Review (which echoes observations made previously on AI with AI) that "hundreds of AI tools have been built to catch COVID. None of them helped." DeepMind has used its AlphaFold program to identify the structure for 98.5 percent of roughly 20,000 human proteins and will make the information publicly available. The Pentagon makes use of machine learning algorithms to create decision space in the latest of its Global Information Dominance Experiments. An Australian court rules that AI systems can be "inventors" under patent law (but not "owners"), and South Africa issues the world's first patent to an "AI System." The United States Special Operations Command put 300 of its personnel through a unique six-week crash course in AI, including leaders such as Google CEO Eric Schmidt and former Defense Secretary Ash Carter. And President Biden nominates Stanford professor Ramin Toloui, who has experience with AI technologies and impacts, as an Assistant Secretary of State for business. In research, DeepMind develop agents capable of "open-ended learning" in XLand, an environment with diverse tasks and challenges. A survey from the Journal of AI Research finds that AI researchers have varying amounts of trust in different organizations, companies, and governments. The Journal of Strategic Studies dedicates an issue to Emerging Technologies, with free access. Mine Cetinkaya-Rundel and Johanna Hardin make an Introduction to Modern Statistics open access with an option (or with proceeds going to OpenIntro, a US-based nonprofit). And Iyad Rahwan curates a collection of evil AI cartoons.

Follow the link below 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.

Follow the link below to visit our website and explore the links mentioned in the episode.

Andy and Dave discuss the latest in AI news and research, including the new DARPA FENCE program (Fast Event-based Neuromorphic Camera and Electronics), which seeks to create event-based cameras that only focus on pixels that have changed in a scene. NIST proposed an approach for reducing the risk of bias in AI and has invited the public to comment and help improve it. Researchers from the University of Colorado, Boulder use a machine learning model to learn physical properties in electronics building blocks (such as clumps of silicon and germanium atoms), as a way to predict how larger electronics components will work or fail. Researchers in South Korea create an artificial skin that mimics human tactile recognition, and couple it with a deep learning algorithm to classify surface structures (with an accuracy of 99.1%). A survey from IE University shows, among other things, that 75% of people surveys in China support replacing parliamentarians with AI, while in the US, 60% were opposed to it. A scientist with <> uses machine learning to learn Rembrandt's style and then recreate missing pieces of the painter's "The Night Watch." Researchers at Harvard, San Diego, Fujitsu, and MIT present methodical research on demonstrating how classification neural networks are susceptible to small 2D transformations and shifts, image crops, and changes in object colors. The GAO releases a report on Facial Recognition Technology, surveying 42 federal agencies, and finds a general lack of accountability in the use of the technology. The WHO releases a report on Ethics and Governance of AI for Health. In rebuttal to DeepMind's "Reward is enough" paper, Roitblat and Byrnes pens separate essays on why "Reward is not enough." An open-access book by Wang and Barabasi looks at the Science of Science. Julia Schneider and Lena Ziyal join forces to provide a comical essay on AI: We Need to Talk, AI. And the National Security Commission on AI holds an all-day summary on Global Emerging Technology.

Follow the link below to visit our website and explore the links mentioned in the episode.

Andy and Dave discuss the latest in AI news, including a report that the Israel Defense Forces used a swarm of small drones in mid-May in Gaza to locate, identify, and attack Hamas militants, using Thor, a 9-kilgram quadrotor drone. A paper in the Journal of American Medical Association examines an early warning system for sepsis, and finds that it misses out on most instances (67%) of cases, and frequently issued false alarms (to which the developer contests the results). A new bill, the Consumer Safety Technology Act, directs the US Consumer Product Safety Commission to run a pilot program to use AI to help in safety inspections. A survey from FICO on The State of Responsible AI (2021) shows, among other things, a disinterest in the ethical and responsible use of AI among business leaders (with 65% of companies saying that can't explain how specific AI model predictions are made, and 22% of companies have an AI ethics board to consider questions on AI ethics and fairness). In a similar vein, a survey from the Pew Research Center and Elon University's Imagining the Internet Center found that 68% of respondents (from across 602 leaders in the AI field) believe that AI ethical principles will NOT be employed by most AI systems within the next decade; the survey includes a summary of the respondents' worries and hopes, as well as some additional commentary. GitHub partners with OpenAI to launch CoPilot, a "Programming Partner" that uses contextual cues to suggest new code. Researchers from Stanford University, UC San Diego, and MIT research Physion, a visual and physical prediction benchmark to measure predictions about commonplace real world physical events (such as when objects: collide, drop, roll, domino, etc). CSET releases a report on Machine Learning and Cybersecurity: Hype and Reality, finding that it is unlikely that machine learning will fundamentally transform cyber defense. Bengio, Lecun, and Hinton join together to pen a white paper on the role of deep learning in AI, not surprisingly eschewing the need for symbolic systems. Aston Zhang and Zack C. Lipton, and Alex J Smola release the latest version of Dive into Deep Learning, now over 1000 pages, and living only as an online version.

Follow the link below to visit our website and explore the links mentioned in the episode.

Andy and Dave discuss the latest in AI news, including research from the San Diego School of Medicine, which used an AI algorithm to analyze terabytes of gene expression data in response to viral infections, identifying 20 genes that predict the severity of a patient's response (across many different viruses). Deputy Secretary of Defense Kathleen Hicks announces a new AI and Data Acceleration initiative, which includes operational data teams and flyaway technical experts. China says it has AI fighter jet pilots that can beat human pilots in simulated dogfights. A study from Stanford estimates the density of CCTV cameras in large cities around the globe (by using computer vision algorithms on street view image data). NIST held a workshop on AI Measurement and Evaluation, with an interesting 22-page read-ahead document. Appen updates its State of AI and Machine Learning report, examining various business-related views and metrics on AI, showing a general maturing of the AI market. Researchers from Tubingen and Max Planck show that the behavioral difference between human and machine vision is narrowing, but still has room for improvement (particularly with out-of-distribution data). Researchers from Stanford, University of College London, and MIT develop a counterfactual simulation model to provide quantitative predictions on how people think about causation, possibly serving as a bridge between psychology and AI. Adam Wagner uses a reinforcement learning approach to search for examples that would disprove conjectures in graph theory and finds examples that disprove five such conjectures. Justin Solomon's Numerical Algorithms provides the core methods for machine learning. And Budiansky publishes a look at the life of Kurt Gödel, in Journey to the Edge of Reason.

Follow the link below to visit our website and explore the links mentioned in the episode.

Andy and Dave discuss the latest in AI news, including the European Commission's proposal for the regulation of AI. A report in Nature Medicine examines the limitations of the evaluation process for medical devices using AI that the FDA approves. Researchers at MIT translate spider webs into sounds to explore how spiders might sense their world, and they using machine learning to classify sounds by spider activities. An NIH panel releases its preliminary ethics rules on making brain-like structures such as neural organoids and neural transplants, and finds little evidence that these structures experience humanlike consciousness or pain. And Andy and Dave spend some time with xenobioticists Sam Kriegman and Doug Blackiston, who discuss the motivations and findings behind their latest generation of xenobots, synthetic living machines that they have been designing and building in their labs.

Andy and Dave discuss the latest in AI news, including the launch of the National AI Research Resource Task Force, which will serve as a federal advisory committee and produce at least two reports to Congress (a roadmap and implementation plan) by November 2022. Google and Harvard University release a 1.4 PB reconstruction of a cubic millimeter of human brain tissue. Google reports a deep reinforcement-learning system that outperforms humans in designing floorplans for microchips, both in time and inefficiency. Researchers from the UK, Germany, and China fuse electronics to the Madagascar hissing cockroach to create an insect-computer hybrid for autonomous search and rescue. The Navy's MQ-25 tanker drone refuels a manned aircraft for the first time. Researchers use large-scale experiments and machine learning to discover a greater hierarchy of theories of human decision-making. OpenAI introduces a Process for Adapting Language Models to Society (PALMS) as a way to try to mitigate bias in transformer models such as GPT-3. A concept paper from DeepMind examines why reward systems are enough to constitute a solution to artificial general intelligence. And Richard Sutton and Andrew Barto publish the second edition of Reinforcement Learning: An Introduction.

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