Just weeks after Google employees walked out of offices to protest the way the company dealt with claims of sexual misconduct, Google is bracing itself for another worldwide protest. This time, it's over Google's ominous Project Dragonfly, and human rights organization Amnesty International is throwing its whole weight behind it. Project Dragonfly has already received global backlash, with Google employees themselves calling publicly for an ethics review into the proposed censored Chinese search engine. According to leaked documents, the search app will automatically identify websites blocked by China's so-called Great Firewall. This includes information on free speech, current affairs and political opposition, plus historical references to specific events (such as the 1989 Tiananmen Square massacre) and books that negatively feature authoritarian governments.

A group of Google employees published an open letter on Tuesday calling on their employer to cancel its plans to build a censored search engine for China, the latest expression of worker unrest at a company that earlier this month saw thousands stage walkouts over its handling of sexual misconduct cases. Google's plan for returning to China, which is known as Project Dragonfly and would reportedly allow the Chinese government to blacklist certain search terms and control air quality data, has garnered significant backlash internally since it was first reported on in August. More than 1,400 Google employees signed an internal petition criticizing the lack of transparency around the project, and at least one employee resigned in protest. But Tuesday's letter, which was initially signed by nine current Google employees, is a bold step for employees of a company that prizes internal transparency but considers leaking information to be not "Googley". Organizers of the letter said they would continuously update the letter as more employees signed on; by midday there were more than 50 signers.

Mention the words artificial intelligence or machine learning and what uncontrollably springs to mind is the fictional endoskeletal, red-eyed Terminator of the same film franchise, followed closely by its fictional creator, SkyNet, the neural communication network that became self-aware. Not your typical response to new technologies, both of which are immeasurably capable and will, beyond debate, forever change the world we believe we live in. This question is centre stage in artificial intelligence research, as the guardians of our race contemplate both the risks and opportunities of creating consciousness at the high end of artificial intelligence projects that strive to create machines in our own image. Herein lies the debate about whether we are trying to create a carbon copy of ourselves - for carbon, read digital - to emulate both physical and mental attributes, or whether we are simply experimenting in all areas, to prove our intellectual prowess in addition to our physical alpha domination of the planet. But to whom are we trying to prove that?

This process, which is far more challenging than most researchers acknowledge, has driven the market capitalization of many Chinese tech companies far beyond that of the American peers they were once accused of "copying." Visionary research will always be important to A.I., which means that China still has much to learn from the United States. But as practical implementation increasingly becomes the name of the game, the United States now has much to learn from China, too. Analysts in the West often acknowledge the areas in which China has an advantage in A.I., but they tend to misunderstand the nature of China's strength in each one. With regard to China's abundant data, analysts often point to the sheer size of China's population (which owns 1.1 billion mobile internet devices) and claim that lax privacy laws allow a free-for-all with user data.

Assessing the fairness of a decision making system with respect to a protected class, such as gender or race, is challenging when class membership labels are unavailable. Probabilistic models for predicting the protected class based on observable proxies, such as surname and geolocation for race, are sometimes used to impute these missing labels for compliance assessments. Empirically, these methods are observed to exaggerate disparities, but the reason why is unknown. In this paper, we decompose the biases in estimating outcome disparity via threshold-based imputation into multiple interpretable bias sources, allowing us to explain when over- or underestimation occurs. We also propose an alternative weighted estimator that uses soft classification, and show that its bias arises simply from the conditional covariance of the outcome with the true class membership. Finally, we illustrate our results with numerical simulations and a public dataset of mortgage applications, using geolocation as a proxy for race. We confirm that the bias of threshold-based imputation is generally upward, but its magnitude varies strongly with the threshold chosen. Our new weighted estimator tends to have a negative bias that is much simpler to analyze and reason about.

The Brazilian court system is currently the most clogged up judiciary system in the world. Thousands of lawsuit cases reach the supreme court every day. These cases need to be analyzed in order to be associated to relevant tags and allocated to the right team. Most of the cases reach the court as raster scanned documents with widely variable levels of quality. One of the first steps for the analysis is to classify these documents. In this paper we present a Bidirectional Long Short-Term Memory network (Bi-LSTM) to classify these pieces of legal document.

Humans are the final decision makers in critical tasks that involve ethical and legal concerns, ranging from recidivism prediction, to medical diagnosis, to fighting against fake news. Although machine learning models can sometimes achieve impressive performance in these tasks, these tasks are not amenable to full automation. To realize the potential of machine learning for improving human decisions, it is important to understand how assistance from machine learning models affects human performance and human agency. In this paper, we use deception detection as a testbed and investigate how we can harness explanations and predictions of machine learning models to improve human performance while retaining human agency. We propose a spectrum between full human agency and full automation, and develop varying levels of machine assistance along the spectrum that gradually increase the influence of machine predictions. We find that without showing predicted labels, explanations alone do not statistically significantly improve human performance in the end task. In comparison, human performance is greatly improved by showing predicted labels (>20% relative improvement) and can be further improved by explicitly suggesting strong machine performance. Interestingly, when predicted labels are shown, explanations of machine predictions induce a similar level of accuracy as an explicit statement of strong machine performance. Our results demonstrate a tradeoff between human performance and human agency and show that explanations of machine predictions can moderate this tradeoff.

As I struggled for my first breath, Orwell was busily writing his vivid dystopian novel, 1984. That was 1948 and he switched the last two digits to get the title. I didn't read it until 1971 when it was essential reading on the youth revolution syllabus. We worried that the lust for power could create an authoritarian all-knowing state. Perhaps it is now time for the younger generation to take this more seriously before it gently creeps up and bites them.

If you've been following the latest hiring trends, you may have noticed that many recruiters are turning to artificial intelligence (AI) tools to tackle discrimination in hiring―and the expectations for success are high. However, HR technology analysts and even executives at companies offering AI solutions caution that a totally bias-free hiring process may be difficult to achieve. Amazon, the world's largest online retailer, found out the hard way. In 2015, the company discovered that a recruiting system it was building with machine-learning algorithms had begun to downgrade certain resumes that included words such as "women's club." By contrast, the system favored male candidates to whom such verbs such as "executed" and "captured" were attributed.

As artificial intelligence continues to find its way into our daily lives, its propensity to interfere with human rights only gets more severe. With this in mind, and noting that the technology is still in its infant stages, Access Now conducts this preliminary study to scope the potential range of human rights issues that may be raised today or in the near future. Many of the issues that arise in examinations of this area are not new, but they are greatly exacerbated by the scale, proliferation, and real-life impact that artificial intelligence facilitates. Because of this, the potential of artificial intelligence to both help and harm people is much greater than from technologies that came before. While we have already seen some of these consequences, the impacts will only continue to grow in severity and scope. However, by starting now to examine what safeguards and structures are necessary to address problems and abuses, the worst harms -- including those that disproportionately impact marginalized people -- may be prevented and mitigated. There are several lenses through which experts examine artificial intelligence. The use of international human rights law and its well-developed standards and institutions to examine artificial intelligence systems can contribute to the conversations already happening, and provide a universal vocabulary and forums established to address power differentials.