Thousands of Google employees and contractors around the globe--many of them women--briefly walked off the job Thursday to protest Google's handling of sexual harassment claims and other workplace issues, and to demand more transparency around harassment incidents and pay levels at the company. The demonstrations took place outside about 40 Google offices, including Singapore, London, New York, San Francisco, and the company's headquarters in Mountain View, California. The protest was spurred by a recent New York Times article about Google awarding multimillion-dollar exit packages to top male executives accused of sexual misconduct, including a $90 million payment to Android founder Andy Rubin, even after Google investigators found credible a claim that Rubin coerced a female employee into performing oral sex. In San Francisco, workers carried signs saying "Not OK Google" and "Equal Pay 4 Equal Work." Organizers led the crowd in chants like "Time's up for Google," and read accounts of workplace harassment from anonymous Google employees, who did not share their names.

Personal electronic devices such as smartphones give access to a broad range of behavioral signals that can be used to learn about the characteristics and preferences of individuals. In this study we explore the connection between demographic and psychological attributes and digital records for a cohort of 7,633 people, closely representative of the US population with respect to gender, age, geographical distribution, education, and income. We collected self-reported assessments on validated psychometric questionnaires based on both the Moral Foundations and Basic Human Values theories, and combined this information with passively-collected multi-modal digital data from web browsing behavior, smartphone usage and demographic data. Then, we designed a machine learning framework to infer both the demographic and psychological attributes from the behavioral data. In a cross-validated setting, our model is found to predict demographic attributes with good accuracy (weighted AUC scores of 0.90 for gender, 0.71 for age, 0.74 for ethnicity). Our weighted AUC scores for Moral Foundation attributes (0.66) and Human Values attributes (0.60) suggest that accurate prediction of complex psychometric attributes is more challenging but feasible. This connection might prove useful for designing personalized services, communication strategies, and interventions, and can be used to sketch a portrait of people with similar worldviews.

Machine learning models trained on data from the outside world can be corrupted by data poisoning attacks that inject malicious points into the models' training sets. A common defense against these attacks is data sanitization: first filter out anomalous training points before training the model. Can data poisoning attacks break data sanitization defenses? In this paper, we develop three new attacks that can all bypass a broad range of data sanitization defenses, including commonly-used anomaly detectors based on nearest neighbors, training loss, and singular-value decomposition. For example, our attacks successfully increase the test error on the Enron spam detection dataset from 3% to 24% and on the IMDB sentiment classification dataset from 12% to 29% by adding just 3% poisoned data. In contrast, many existing attacks from the literature do not explicitly consider defenses, and we show that those attacks are ineffective in the presence of the defenses we consider. Our attacks are based on two ideas: (i) we coordinate our attacks to place poisoned points near one another, which fools some anomaly detectors, and (ii) we formulate each attack as a constrained optimization problem, with constraints designed to ensure that the poisoned points evade detection. While this optimization involves solving an expensive bilevel problem, we explore and develop three efficient approximations to this problem based on influence functions; minimax duality; and the Karush-Kuhn-Tucker (KKT) conditions. Our results underscore the urgent need to develop more sophisticated and robust defenses against data poisoning attacks.

A machine learning model may exhibit discrimination when used to make decisions involving people. One potential cause for such outcomes is that the model uses a statistical proxy for a protected demographic attribute. In this paper we formulate a definition of proxy use for the setting of linear regression and present algorithms for detecting proxies. Our definition follows recent work on proxies in classification models, and characterizes a model's constituent behavior that: 1) correlates closely with a protected random variable, and 2) is causally influential in the overall behavior of the model. We show that proxies in linear regression models can be efficiently identified by solving a second-order cone program, and further extend this result to account for situations where the use of a certain input variable is justified as a "business necessity". Finally, we present empirical results on two law enforcement datasets that exhibit varying degrees of racial disparity in prediction outcomes, demonstrating that proxies shed useful light on the causes of discriminatory behavior in models.

Sequential learning of tasks using gradient descent leads to an unremitting decline in the accuracy of tasks for which training data is no longer available, termed catastrophic forgetting. Generative models have been explored as a means to approximate the distribution of old tasks and bypass storage of real data. Here we propose a cumulative closed-loop generator and embedded classifier using an AC-GAN architecture provided with external regularization by a small buffer. We evaluate incremental learning using a notoriously hard paradigm, single headed learning, in which each task is a disjoint subset of classes in the overall dataset, and performance is evaluated on all previous classes. First, we show that the variability contained in a small percentage of a dataset (memory buffer) accounts for a significant portion of the reported accuracy, both in multi-task and continual learning settings. Second, we show that using a generator to continuously output new images while training provides an up-sampling of the buffer, which prevents catastrophic forgetting and yields superior performance when compared to a fixed buffer. We achieve an average accuracy for all classes of 92.26% in MNIST and 76.15% in FASHION-MNIST after 5 tasks using GAN sampling with a buffer of only 0.17% of the entire dataset size. We compare to a network with regularization (EWC) which shows a deteriorated average performance of 29.19% (MNIST) and 26.5% (FASHION). The baseline of no regularization (plain gradient descent) performs at 99.84% (MNIST) and 99.79% (FASHION) for the last task, but below 3% for all previous tasks. Our method has very low long-term memory cost, the buffer, as well as negligible intermediate memory storage.

A digital border guard will interrogate travellers at some European Union borders in an attempt to ramp up security at crossings. Dubbed iBorderCtrl, the agent features an AI lie detector that quizzes tourists on their trip, including the contents of their suitcase. The system is part of a six-month trial run by the Hungarian National Police at four different border crossing points in Hungary, Latvia, and Greece. If successful, the technology could be rolled out to borders across the union's member states. A digital border guard will interrogate travellers at some European Union borders in an attempt to ramp up security at crossings.

That's one of the surprising -- and unsettling -- questions Israeli historian Yuval Noah Harari asks in his much-quoted new book, 21 Lessons for the 21st Century. Whereas 20th-century technology favored democracies as they were able to distribute power to make decisions among many people and institutions, according to Harari, artificial intelligence (AI) might make centralized systems that concentrate all information and power far more efficient as machine learning works better with more information to analyze. "If you disregard all privacy concerns and concentrate all the information relating to a billion people in one database," Harari writes, "you'll wind up with much better algorithms than if you respect individual privacy and have in your database only partial information on a million people." The rise of AI swinging the pendulum from democracies toward authoritarian regimes is just one of the feared adverse impacts of technologies: Others include job displacement, concentration of power, diminishing privacy, rising income inequality and losing our "free will." Yet most people have little or no knowledge about how AI, blockchain, the Internet of Things or genetic engineering could affect their lives.

Arizona Governor Doug Ducey recently announced a new multimillion-dollar public-private research partnership in the pursuit of fully autonomous vehicles. The move is an indication that Arizona is doubling down on its status as an autonomous-vehicle testing hotbed--mere months after the first-ever human fatality involving an autonomous vehicle occurred on Arizona roads. Despite plenty of setbacks, automakers and tech companies remain committed to autonomous vehicles, with many saying that consumers will be able to hail a driverless taxi within the next couple of years. But serious cracks are beginning to emerge in this roadmap: Even younger generations are not yet onboard. Plenty of invested capital is riding on autonomous vehicles.

SAN FRANCISCO/BEIJING – DJI, the world's top seller of drones for consumers, has snagged a pair of wins in its effort to court businesses. SZ DJI Technology Co. said its latest industrial gadget -- the Mavic 2 Enterprise -- will soon survey power grids for U.S. utility Southern Co., while American Airlines Group Inc. will test the craft for plane inspections. Those are important alliances for the Chinese company, which is grappling with a U.S. government shut-out, a potentially damaging patent lawsuit and rising American tariffs. Privacy is a particularly thorny issue for DJI -- one of the few Chinese technology giants that's made major strides abroad. Escalating U.S. tensions are fueling concerns about the dominance of a Chinese company in unmanned flying craft.

Amazon knows you'll ask Alexa all sorts of questions about the midterms and politics in general, so it's been preparing the voice assistant for the event. It has teamed up with nonprofit digital encyclopedia Ballotpedia to equip Alexa with answers first-time voters will find especially helpful. You can ask the assistant when the polls will open and what's on your ballot. Alexa can even answer what it means to vote yes or no for a certain ballot measure and can list nominees running for a specific position. On Election Day itself, you can ask Alexa for a general update by asking "Alexa, what's my election update?"