We clarify what fairness guarantees we can and cannot expect to follow from unconstrained machine learning. Specifically, we characterize when unconstrained learning on its own implies group calibration, that is, the outcome variable is conditionally independent of group membership given the score. We show that under reasonable conditions, the deviation from satisfying group calibration is upper bounded by the excess risk of the learned score relative to the Bayes optimal score function. A lower bound confirms the optimality of our upper bound. Moreover, we prove that as the excess risk of the learned score decreases, it strongly violates separation and independence, two other standard fairness criteria. Our results show that group calibration is the fairness criterion that unconstrained learning implicitly favors. On the one hand, this means that calibration is often satisfied on its own without the need for active intervention, albeit at the cost of violating other criteria that are at odds with calibration. On the other hand, it suggests that we should be satisfied with calibration as a fairness criterion only if we are at ease with the use of unconstrained machine learning in a given application.

A new book by Conflict Analytics project member Jacob Turner is delving into some of the key issues surrounding AI. Published by Palgrave, Robot Rules argues that AI's ability to make independent decisions makes it unique – and unpredictable. "In Robot Rules: Regulating Artificial Intelligence, I explore what makes AI unique, what legal and ethical problems this will cause, and how we can solve them," Turner says. Turner lays out three issues: the responsibility for harm caused by AI, the rights surrounding'legal personality' for AI, and the ethics behind an AI decision-making process. Robot Rules suggests that in order to address these questions we need to develop new institutions and regulations on a cross-industry and international level.

As the human population has grown, our demands on the ocean have increased rapidly. These demands have similarly increased the pressure we place on these systems, and we now cause considerable damage globally. If we want to maintain healthy ocean ecosystems into the future, we must learn to use ocean resources in a sustainable way and facilitate recovery in regions that have declined. Determining how to make these goals a reality, however, is no small challenge. Ingeman et al. review the challenge presented by attempting both to recover and to use ecosystems simultaneously and discuss several approaches for facilitating this essential dual goal. Ocean defaunation and loss of marine ecosystem services present an urgent need to recover degraded ocean ecosystems. Growing scientific awareness, strong regulations, and effective management have begun to fulfill the promise of recovery. Unfortunately, many efforts remain unsuccessful, in part because marine ecosystems and human societies are changing. Rapid shifts in environmental conditions are undermining previously effective recovery strategies. Moreover, divergent perceptions of recovery exist. Efforts toward reversing marine degradation must address the dynamic social-ecological landscape in which recoveries occur, or forever chase a moving target. Recovery efforts of tomorrow will require institutional and tactical flexibility to keep pace with a changing ocean, and an inclusive concept of recovery. Further, vital population-level efforts will be most successful when complemented by a broader ecosystem and social-ecological perspective. In this Review, we provide a synthesis of ocean-recovery goals as moving targets and highlight promising steps forward. While acknowledging the priority of basic conservation imperatives, successful recoveries can encompass a range of outcomes in the space between minimum ecological viability and maximum carrying capacity. Ongoing advances are improving our ability to predict the effects of environmental change on ocean productivity and to calibrate recovery targets to changing conditions. As a complement to predict-and-prescribe methods, research can also point the way toward robust approaches in the face of irreducible uncertainty.

Artificial intelligence (AI) presents limitless opportunity, but not without potential pitfalls and risks. This paradox has become increasingly evident for government leaders. They want to give domestic companies an edge over the competition, but are also expected to protect their citizens and use AI for social good. They want to support innovation, while still maintaining some level of control over how new technologies impact society at large. With a huge payoff on the line -- by our own estimates, AI has the potential to increase worldwide GDP by 14 percent by 2030, an infusion of US$15.7 trillion into the global economy -- it should come as no surprise that governments are eager to claim their share.

Monitoring land use changes are important to identify new buildings coming up, crop lands converting to plantations, water bodies getting dumped with land fills etc. Manually monitoring these changes is cumbersome and costly. We are trying to use machine learning and satellite data and check whether these changes can be identified. This blog is about our effort and some preliminary results. NASA and ESA have launched satellites which help us to try out our algorithms. For works like this we use ESA's Sentinel 2 satellite which is the best high resolution satellite data available in open domain.

The July 2, 1978 issue of the New York Times was the final one the paper sent to print under the linotype process. After decades of relying on Gutenburg printing press-style technology, the newspaper invested in a computerized method that would eliminate the need to physically cast each letter of every page into lead plates for the presses. The automation and digitization of the "hot type" process did not leave linotype operators jobless, however. Those same employees who had run the hot metal typesetting machines were sitting in front of computers the next day, typing stories into a digital format rather than hammering them into place. Asked what the technological upgrade would mean for him personally, one employee responded, "it means I'll have to learn a new process."1

As companies from IBM to Samsung Electronics Co. to Halliburton Co. scramble to find the next great invention using artificial intelligence, they may hit a roadblock when trying to patent their ideas. The U.S. Patent and Trademark Office is making it increasingly difficult to obtain legal protections for inventions related to AI, a field that encompasses autonomous cars, virtual assistants and financial analyses, among countless other uses. The agency, seeing an influx of AI applications, is grappling with how to...

In data summarization we want to choose k prototypes in order to summarize a data set. We study a setting where the data set comprises several demographic groups and we are restricted to choose k_i prototypes belonging to group i. A common approach to the problem without the fairness constraint is to optimize a centroid-based clustering objective such as k-center. A natural extension then is to incorporate the fairness constraint into the clustering objective. Existing algorithms for doing so run in time super-quadratic in the size of the data set. This is in contrast to the standard k-center objective that can be approximately optimized in linear time. In this paper, we resolve this gap by providing a simple approximation algorithm for the k-center problem under the fairness constraint with running time linear in the size of the data set and k. If the number of demographic groups is small, the approximation guarantee of our algorithm only incurs a constant-factor overhead. We demonstrate the applicability of our algorithm on both synthetic and real data sets.

Fairness by decision-makers is believed to be auditable by third parties. In this study, we show that this is not always true. We consider the following scenario. Imagine a decision-maker who discloses a subset of his dataset with decisions to make his decisions auditable. If he is corrupt, and he deliberately selects a subset that looks fair even though the overall decision is unfair, can we identify this decision-maker's fraud? We answer this question negatively. We first propose a sampling method that produces a subset whose distribution is biased from the original (to pretend to be fair); however, its differentiation from uniform sampling is difficult. We call such a sampling method as stealthily biased sampling, which is formulated as a Wasserstein distance minimization problem, and is solved through a minimum-cost flow computation. We proved that the stealthily biased sampling minimizes an upper-bound of the indistinguishability. We conducted experiments to see that the stealthily biased sampling is, in fact, difficult to detect.

Much discussion of the future of work suggests it can only be a dystopian, robotic world. But the report of an ILO commission shows how humans, not algorithms, can be in charge. When the International Labour Organization (ILO) was founded 100 years ago in the aftermath of the first world war, governments, employers and workers came together convinced that lasting peace and stability depended on social justice. This is still true and, given the dramatic changes we are seeing, should encourage us to take bold and timely action. The constitution of the ILO of 1919, reinforced by the Philadelphia Declaration of 1944, remains the most ambitious global social contract in history.