NOTE: This is an excerpt of Planet Money's newsletter. You can sign up here. If you shop online, there's a good chance the price you pay for stuff is determined by a computer algorithm. As of 2015, over one third of the 1,600 best-selling items sold on Amazon came from sellers who used algorithms to set their price. Algorithms are spreading like crazy, but are they giving companies too much power over consumers?

As algorithms are increasingly used to make important decisions pertaining to individuals, algorithmic discrimination is becoming a prominent concern. The seminal work of Dwork et al. [ITCS 2012] introduced the notion of individual fairness (IF): given a task-specific similarity metric, every pair of similar individuals should receive similar outcomes. In this work, we study fairness when individuals have diverse preferences over the possible outcomes. We show that in such settings, individual fairness can be too restrictive: requiring individual fairness can lead to less-preferred outcomes for the very individuals that IF aims to protect (e.g. a protected minority group). We introduce and study a new notion of preference-informed individual fairness (PIIF), a relaxation of individual fairness that allows for outcomes that deviate from IF, provided the deviations are in line with individuals' preferences. We show that PIIF can allow for solutions that are considerably more beneficial to individuals than the best IF solution. We further show how to efficiently optimize any convex objective over the outcomes subject to PIIF, for a rich class of individual preferences. Motivated by fairness concerns in targeted advertising, we apply this new fairness notion to the multiple-task setting introduced by Dwork and Ilvento [ITCS 2019]. We show that, in this setting too, PIIF can allow for considerably more beneficial solutions, and we extend our efficient optimization algorithm to this setting.

Understanding of the effect of a particular treatment or a policy pertains to many areas of interest -- ranging from political economics, marketing to health-care and personalized treatment studies. In this paper, we develop a non-parametric, model-free test for detecting the effects of treatment over time that extends widely used Synthetic Control tests. The test is built on counterfactual predictions arising from many learning algorithms. In the Neyman-Rubin potential outcome framework with possible carry-over effects, we show that the proposed test is asymptotically consistent for stationary, beta mixing processes. We do not assume that class of learners captures the correct model necessarily. We also discuss estimates of the average treatment effect, and we provide regret bounds on the predictive performance. To the best of our knowledge, this is the first set of results that allow for example any Random Forest to be useful for provably valid statistical inference in the Synthetic Control setting. In experiments, we show that our Synthetic Learner is substantially more powerful than classical methods based on Synthetic Control or Difference-in-Differences, especially in the presence of non-linear outcome models.

Missing data are a concern in many real world data sets and imputation methods are often needed to estimate the values of missing data, but data sets with excessive missingness and high dimensionality challenge most approaches to imputation. Here we show that appropriate feature selection can be an effective preprocessing step for imputation, allowing for more accurate imputation and subsequent model predictions. The key feature of this preprocessing is that it incorporates uncertainty: by accounting for uncertainty due to missingness when selecting features we can reduce the degree of missingness while also limiting the number of uninformative features being used to make predictive models. We introduce a method to perform uncertainty-aware feature selection (UAFS), provide a theoretical motivation, and test UAFS on both real and synthetic problems, demonstrating that across a variety of data sets and levels of missingness we can improve the accuracy of imputations. Improved imputation due to UAFS also results in improved prediction accuracy when performing supervised learning using these imputed data sets. Our UAFS method is general and can be fruitfully coupled with a variety of imputation methods.

Most of the time, machine learning does not touch on particularly sensitive social, moral, or ethical issues. Someone gives us a data set and asks us to predict house prices based on given attributes, classifying pictures into different categories, or teaching a computer the best way to play PAC-MAN -- what do we do when we are asked to base predictions of protected attributes according to anti-discrimination laws? How do we ensure that we do not embed racist, sexist, or other potential biases into our algorithms, be it explicitly or implicitly? It may not surprise you that there have been several important lawsuits in the United States on this topic, possibly the most notably one involving Northpointe's controversial COMPAS -- Correctional Offender Management Profiling for Alternative Sanctions -- software, which predicts the risk that a defendant will commit another crime. The proprietary algorithm considers some of the answers from a 137-item questionnaire to predict this risk.

Learning portable neural networks is very essential for computer vision for the purpose that pre-trained heavy deep models can be well applied on edge devices such as mobile phones and micro sensors. Most existing deep neural network compression and speed-up methods are very effective for training compact deep models, when we can directly access the training dataset. However, training data for the given deep network are often unavailable due to some practice problems (e.g. privacy, legal issue, and transmission), and the architecture of the given network are also unknown except some interfaces. To this end, we propose a novel framework for training efficient deep neural networks by exploiting generative adversarial networks (GANs). To be specific, the pre-trained teacher networks are regarded as a fixed discriminator and the generator is utilized for derivating training samples which can obtain the maximum response on the discriminator. Then, an efficient network with smaller model size and computational complexity is trained using the generated data and the teacher network, simultaneously. Efficient student networks learned using the proposed Data-Free Learning (DFL) method achieve 92.22% and 74.47% accuracies without any training data on the CIFAR-10 and CIFAR-100 datasets, respectively. Meanwhile, our student network obtains an 80.56% accuracy on the CelebA benchmark.

The UK government is among a group of countries that are attempting to thwart plans to formulate and impose a pre-emptive ban on killer robots. Delegates have been meeting at the UN in Geneva all week to discuss potential restrictions under international law to so-called lethal autonomous weapons systems, which use artificial intelligence to help decide when and who to kill. Most states taking part – and particularly those from the global south – support either a total ban or strict legal regulation governing their development and deployment, a position backed by the UN secretary general, António Guterres, who has described machines empowered to kill as "morally repugnant". But the UK is among a group of states – including Australia, Israel, Russia and the US – speaking forcefully against legal regulation. As discussions operate on a consensus basis, their objections are preventing any progress on regulation.

The next day at the Slack office, people were quite literally sobbing in the cafeteria. I was mostly keeping my shit together until my parents called from Canada. I went into one of the little phone booths and just sobbed on the phone. It took a bit of time to grieve, but then you also have to act. The space that Maciej1 created in Tech Solidarity was incredibly important. To show up at that first meeting at the Stripe offices and see hundreds of other people who are figuring out what the hell to do next was incredibly gratifying. "Oh, Joe who works over at the security team at a text-editor company actually cares about the fate of Muslim people in America." There were lots of pleasant surprises like that. I think one of the things that Tech Solidarity got really right was: "Don't show up at these organizations offering to make an app for them that you're going to abandon. Show up and help them fix their printer. Show up and just give them money. You made a lot of money on the IPO or whatever. Just give them your money." It was after the first meeting that I thought about the pledge.

When Stanford announced a new artificial intelligence institute, the university said the "designers of AI must be broadly representative of humanity" and unveiled 120 faculty and tech leaders partnering on the initiative. Some were quick to notice that not a single member of this "representative" group appeared to be black. The backlash was swift, sparking discussion on the severe lack of diversity across the AI field. But the problems surrounding representation extend far beyond exclusion and prejudice in academia. Major tech corporations have launched AI "ethics" boards that not only lack diversity, but sometimes include powerful people with interests that don't align with the ethics mission.

BEIJING - People.cn, the online unit of China's influential People's Daily, is boosting its numbers of human internet censors backed by artificial intelligence to help firms vet content on apps and adverts, capitalizing on its unmatched Communist Party lineage. Demand for online censoring services provided by the Shanghai-listed People.cn has soared since last year after China tightened its already strict online censorship rules. As a unit of the People's Daily -- the ruling Communist Party's mouthpiece -- it is seen by clients as the go-to online censor. Investors concur, lifting shares in People.cn "The biggest advantage of People.cn is its precise grasp of policy trends," said An Fushuang, an independent analyst based in Shenzhen.