The field of multi-agent system (MAS) is an active area of research within artificial intelligence, with an increasingly important impact in industrial and other real-world applications. In a MAS, autonomous agents interact to pursue personal interests and/or to achieve common objectives. Distributed Constraint Optimization Problems (DCOPs) have emerged as a prominent agent model to govern the agents' autonomous behavior, where both algorithms and communication models are driven by the structure of the specific problem. During the last decade, several extensions to the DCOP model have been proposed to enable support of MAS in complex, real-time, and uncertain environments. This survey provides an overview of the DCOP model, offering a classification of its multiple extensions and addressing both resolution methods and applications that find a natural mapping within each class of DCOPs. The proposed classification suggests several future perspectives for DCOP extensions and identifies challenges in the design of efficient resolution algorithms, possibly through the adaptation of strategies from different areas.

Future Tense is a partnership of Slate, New America, and Arizona State University that examines emerging technologies, public policy, and society. This week we're excited to share our latest installment of Future Tense Fiction, "Domestic Violence," a story by futurist and science-fiction writer Madeline Ashby. The story raises questions about the relationship between abuse and technology, something that you can learn more about in a response to the story by the National Network to End Domestic Violence's Ian Harris. The ways our technology use can be weaponized against our best interests continued to dominate the news this week. April Glaser explains what might come of the new Federal Trade Commission investigation into Facebook's privacy practices in the wake of the Cambridge Analytica debacle.

Friday's Supreme Court decision legalizing gay marriage was a historic moment for civil rights in America, and for the first time ever, Facebook released a tool that encouraged people express solidarity with a rainbow profile picture. I really resent that they are using something like this to glean information that they can use later on. I try to stay away from giving Facebook any information that I don't have to. Things like this are why. They are taking advantage of people's good intentions.

ARTIFICIAL intelligence (AI) is barging its way into business. As our special report this week explains, firms of all types are harnessing AI to forecast demand, hire workers and deal with customers. In 2017 companies spent around $22bn on AI-related mergers and acquisitions, about 26 times more than in 2015. The McKinsey Global Institute, a think-tank within a consultancy, reckons that just applying AI to marketing, sales and supply chains could create economic value, including profits and efficiencies, of $2.7trn over the next 20 years. Google's boss has gone so far as to declare that AI will do more for humanity than fire or electricity.

Legal trials historically rely on a plethora of different steps before a conclusion is made. But for many cases heard in the US, it's not just juries, judges and magistrates who contribute to the final decision. COMPAS (Correctional Offender Management Profiling for Alternative Sanctions) is an artificially intelligent tool used as part of the sentencing process. It uses 137 different features to predict if a defendant will reoffend, and therefore influences the type and length of punishment. It makes sense to assume that this software had undergone strict trials itself.

Originally published in French as Phénoménologie de la Perception (1962/1945).

Artificial Intelligence is on the verge of penetrating every major industry from healthcare to advertising, transportation, finance, legal, education, and now inside the workplace. Many of us may have already interacted with a chatbot (defined as an automated, yet personalized, conversation between software and human users) whether it's on Facebook Messenger to book a hotel room or ordering flowers through 1-800 flowers. According to Facebook Vice President, David Marcus, there are now more than 100,000 chatbots on the Facebook Messenger platform, up from 33,000 in 2016. As we increase the usage of chatbots in our personal lives, we will expect to use them in the workplace to assist us with things like finding new jobs, answering frequently asked HR related questions or even receiving coaching and mentoring. Chatbots digitize HR processes and enable employees to access HR solutions from anywhere.

But tech shares were hit the hardest, dragging down the broader market in the final hour of trading. A series of recent developments pointed to more government oversight of the industry. Facebook Chief Executive Mark Zuckerberg is planning to testify before Congress about the social-media company's privacy and data-use standards, according to people familiar with the matter. The company's shares fell 4.9% on Tuesday and are down 15% this month over concerns about its handling of user data, on track for its worst monthly decline since 2012. The Federal Trade Commission, in a statement Monday, signaled that it is conducting a broad probe of Facebook, while 37 state attorneys general are also demanding explanations for its practices.

Causal influence measures for machine learnt classifiers shed light on the reasons behind classification, and aid in identifying influential input features and revealing their biases. However, such analyses involve evaluating the classifier using datapoints that may be atypical of its training distribution. Standard methods for training classifiers that minimize empirical risk do not constrain the behavior of the classifier on such datapoints. As a result, training to minimize empirical risk does not distinguish among classifiers that agree on predictions in the training distribution but have wildly different causal influences. We term this problem covariate shift in causal testing and formally characterize conditions under which it arises. As a solution to this problem, we propose a novel active learning algorithm that constrains the influence measures of the trained model. We prove that any two predictors whose errors are close on both the original training distribution and the distribution of atypical points are guaranteed to have causal influences that are also close. Further, we empirically demonstrate with synthetic labelers that our algorithm trains models that (i) have similar causal influences as the labeler's model, and (ii) generalize better to out-of-distribution points while (iii) retaining their accuracy on in-distribution points.

It's no secret that much of the legal industry is build on reusable content. Most law firms have their own customized set of standard documents (like NDAs or Wills), but lawyers or associates still have to customize these documents by hand each time a client needs them drafted. Lawyaw, part of YC's Winter '18 class, is building software to automate this process by letting lawyers turn previously completed documents into smart templates. Here's how it works: Lawyers can drag an already customized world document into Lawyaw's platform and it will automatically use natural language processing to first figure out what sections need to be replaced, then actually fill in those sections with the correct personalized phrases and variables. For example, software will automatically detect and replace a client's name, contact information, location, and even more complicated things like scope of engagement. If a variable isn't automatically detected Lawyaw lets users manually select it, which the software will remember for future uses.