Air Force Chief Scientist confirms F-35 will include artificial intelligence -- Defense Systems

#artificialintelligence

F-35s, F-22s and other fighter jets will soon use improved artificial intelligence to control nearby drone wingmen that will be able to carry weapons, test enemy air defenses or perform intelligence, surveillance and reconnaissance missions in high risk areas, senior Air Force officials said. "This involves an attempt to have another platform fly alongside a human, perhaps serving as a weapons truck carrying a bunch of missiles," Zacharias said in an interview with Defense Systems. An F-35 computer system, Autonomic Logistics Information System, uses early applications of artificial intelligence that help computers make assessments, go through checklists, organize information and make some decisions by themselves – without needing human intervention. "We are working on making platforms more autonomous with multi-infusion systems and data from across different intel streams," Zacharias explained. ALIS serves as the information infrastructure for the F-35, transmitting aircraft health and maintenance action information to the appropriate users on a globally-distributed network to technicians worldwide, said Lockheed Martin, the contractor that built the system.


A Deep Reinforcement Learning Chatbot

arXiv.org Machine Learning

We present MILABOT: a deep reinforcement learning chatbot developed by the Montreal Institute for Learning Algorithms (MILA) for the Amazon Alexa Prize competition. MILABOT is capable of conversing with humans on popular small talk topics through both speech and text. The system consists of an ensemble of natural language generation and retrieval models, including template-based models, bag-of-words models, sequence-to-sequence neural network and latent variable neural network models. By applying reinforcement learning to crowdsourced data and real-world user interactions, the system has been trained to select an appropriate response from the models in its ensemble. The system has been evaluated through A/B testing with real-world users, where it performed significantly better than many competing systems. Due to its machine learning architecture, the system is likely to improve with additional data.


Can Behavioral Science Help in Flint?

AITopics Original Links

A week after Donald Trump's election, a thirty-year-old cognitive scientist named Maya Shankar purchased a plane ticket to Flint, Michigan. Shankar held one of the more unorthodox jobs in the Obama White House, running the Social and Behavioral Sciences Team, also known as the President's "nudge unit." When she launched the team, in early 2014, it felt, Shankar recalls, "like a startup in my parents' basement"--no budget, no mandate, no bona-fide employees. Within two years, the small group of scientists had become a staff of dozens--including an agricultural economist, an industrial psychologist, and "human-centered designers"--working with more than twenty federal agencies on seventy projects, from fixing gaps in veterans' health care to relieving student debt. Usually, the initiatives had, at their core, one question: Could the growing body of knowledge about the quirks of the human brain be used to improve public policy? For months, Shankar had been thinking about how to bring behavioral science to bear on the problems in Flint, where a crisis stemming from lead contamination of the drinking water had stretched on for almost two years. She wondered if lessons from the beleaguered city could inform the Administration's approach to the broader threat posed by lead across America--in pipes, in paint, in dust, and in soil. "Flint is not the only place poisoning kids," Shankar said. In recent years, behavioral science has become a voguish field. In 2002, the Israeli psychologist Daniel Kahneman won a Nobel Prize in Economic Sciences for his work with a colleague, Amos Tversky, exploring the peculiarities of human decision-making in the face of uncertainty. A basic premise of the discipline they'd helped to create was that people's cognition is bias-prone, and susceptible to the cognitive equivalent of optical illusions. As a result, small tweaks of presentation or circumstance could make a major difference: if a judge rendered a decision about granting parole just before a meal, the inmate's odds for a favorable outcome dipped to near zero; just after the judge ate, the chances rose to around sixty-five per cent. Grocers had learned that they could sell double the amount of soup if they placed a sign above their cans reading "limit of 12 per person." But, for all the field's potential, its advances seemed mostly to have served the private sector. A prominent exception was the "nudge," a notion advanced by the legal scholar Cass R. Sunstein, now at Harvard Law School, and the University of Chicago behavioral economist Richard Thaler, in their 2008 best-seller "Nudge: Improving Decisions About Health, Wealth, and Happiness."


Inferring Strategies for Sentence Ordering in Multidocument News Summarization

arXiv.org Artificial Intelligence

The problem of organizing information for multidocument summarization so that the generated summary is coherent has received relatively little attention. While sentence ordering for single document summarization can be determined from the ordering of sentences in the input article, this is not the case for multidocument summarization where summary sentences may be drawn from different input articles. In this paper, we propose a methodology for studying the properties of ordering information in the news genre and describe experiments done on a corpus of multiple acceptable orderings we developed for the task. Based on these experiments, we implemented a strategy for ordering information that combines constraints from chronological order of events and topical relatedness. Evaluation of our augmented algorithm shows a significant improvement of the ordering over two baseline strategies.


Inferring Strategies for Sentence Ordering in Multidocument News Summarization

Journal of Artificial Intelligence Research

The problem of organizing information for multidocument summarization so that the generated summary is coherent has received relatively little attention. While sentence ordering for single document summarization can be determined from the ordering of sentences in the input article, this is not the case for multidocument summarization where summary sentences may be drawn from different input articles. In this paper, we propose a methodology for studying the properties of ordering information in the news genre and describe experiments done on a corpus of multiple acceptable orderings we developed for the task. Based on these experiments, we implemented a strategy for ordering information that combines constraints from chronological order of events and topical relatedness. Evaluation of our augmented algorithm shows a significant improvement of the ordering over two baseline strategies.