Interpretable and explainable machine learning has seen a recent surge of interest. We focus on safety as a key motivation behind the surge and make the relationship between interpretability and safety more quantitative. Toward assessing safety, we introduce the concept of maximum deviation via an optimization problem to find the largest deviation of a supervised learning model from a reference model regarded as safe. We then show how interpretability facilitates this safety assessment. For models including decision trees, generalized linear and additive models, the maximum deviation can be computed exactly and efficiently. For tree ensembles, which are not regarded as interpretable, discrete optimization techniques can still provide informative bounds. For a broader class of piecewise Lipschitz functions, we leverage the multi-armed bandit literature to show that interpretability produces tighter (regret) bounds on the maximum deviation. We present case studies, including one on mortgage approval, to illustrate our methods and the insights about models that may be obtained from deviation maximization.

To act safely and ethically in the real world, agents must be able to reason about harm and avoid harmful actions. However, to date there is no statistical method for measuring harm and factoring it into algorithmic decisions. In this paper we propose the first formal definition of harm and benefit using causal models. We show that any factual definition of harm must violate basic intuitions in certain scenarios, and show that standard machine learning algorithms that cannot perform counterfactual reasoning are guaranteed to pursue harmful policies following distributional shifts. We use our definition of harm to devise a framework for harm-averse decision making using counterfactual objective functions. We demonstrate this framework on the problem of identifying optimal drug doses using a dose-response model learned from randomized control trial data. We find that the standard method of selecting doses using treatment effects results in unnecessarily harmful doses, while our counterfactual approach allows us to identify doses that are significantly less harmful without sacrificing efficacy.

It's never been more important for companies to ensure that their AI systems function safely, especially as new laws to hold them accountable kick in. The responsible AI teams they set up to do that are supposed to be a priority, but investment in it is still lagging behind. People working in the field suffer as a result, as I found in my latest piece. Organizations place huge pressure on individuals to fix big, systemic problems without proper support, while they often face a near-constant barrage of aggressive criticism online. The problem also feels very personal--AI systems often reflect and exacerbate the worst aspects of our societies, such as racism and sexism. The problematic technologies range from facial recognition systems that classify Black people as gorillas to deepfake software used to make porn videos of women who have not consented.

Jean-Luc Godard once claimed, regarding cinema, "When I die, it will be the end." Godard passed away last month; film perseveres. Yet artificial intelligence has raised a kindred specter: that humans may go obsolete long before their artistic mediums do. Novels scribed by GPT-3; art conjured by DALL·E--machines could be making art long after people are gone. As deepfakes evolve, fears are mounting that future films, TV shows, and commercials may not need them at all.

The full power of human language-based communication cannot be realized without negation. All human languages have some form of negation. Despite this, negation remains a challenging phenomenon for current natural language understanding systems. To facilitate the future development of models that can process negation effectively, we present CONDAQA, the first English reading comprehension dataset which requires reasoning about the implications of negated statements in paragraphs. We collect paragraphs with diverse negation cues, then have crowdworkers ask questions about the implications of the negated statement in the passage. We also have workers make three kinds of edits to the passage -- paraphrasing the negated statement, changing the scope of the negation, and reversing the negation -- resulting in clusters of question-answer pairs that are difficult for models to answer with spurious shortcuts. CONDAQA features 14,182 question-answer pairs with over 200 unique negation cues and is challenging for current state-of-the-art models. The best performing model on CONDAQA (UnifiedQA-v2-3b) achieves only 42% on our consistency metric, well below human performance which is 81%. We release our dataset, along with fully-finetuned, few-shot, and zero-shot evaluations, to facilitate the development of future NLP methods that work on negated language.

In a semi-realistic market simulator, independent reinforcement learning algorithms may facilitate market makers to maintain wide spreads even without communication. This unexpected outcome challenges the current antitrust law framework. We study the effectiveness of maker-taker fee models in preventing cooperation via algorithms. After modeling market making as a repeated general-sum game, we experimentally show that the relation between net transaction costs and maker rebates is not necessarily monotone. Besides an upper bound on taker fees, we may also need a lower bound on maker rebates to destabilize the cooperation. We also consider the taker-maker model and the effects of mid-price volatility, inventory risk, and the number of agents.

In this paper, we introduce the task of modeling interpersonal relationships for story generation. For addressing this task, we propose Relationships as Latent Variables for Story Generation, (ReLiSt). ReLiSt generates stories sentence by sentence and has two major components - a relationship selector and a story continuer. The relationship selector specifies a latent variable to pick the relationship to exhibit in the next sentence and the story continuer generates the next sentence while expressing the selected relationship in a coherent way. Our automatic and human evaluations demonstrate that ReLiSt is able to generate stories with relationships that are more faithful to desired relationships while maintaining the content quality. The relationship assignments to sentences during inference bring interpretability to ReLiSt.

The research field of Legal Natural Language Processing (NLP) has been very active recently, with Legal Judgment Prediction (LJP) becoming one of the most extensively studied tasks. To date, most publicly released LJP datasets originate from countries with civil law. In this work, we release, for the first time, a challenging LJP dataset focused on class action cases in the US. It is the first dataset in the common law system that focuses on the harder and more realistic task involving the complaints as input instead of the often used facts summary written by the court. Additionally, we study the difficulty of the task by collecting expert human predictions, showing that even human experts can only reach 53% accuracy on this dataset. Our Longformer model clearly outperforms the human baseline (63%), despite only considering the first 2,048 tokens. Furthermore, we perform a detailed error analysis and find that the Longformer model is significantly better calibrated than the human experts. Finally, we publicly release the dataset and the code used for the experiments.

From healthcare and education to finance and policing, artificial intelligence (AI) is becoming increasingly embedded in people's daily lives. Despite being posited by advocates as a dispassionate and fairer means of making decisions, free from the influence of human prejudice, the rapid development and deployment of AI has prompted concern over how the technology can be used and abused. These concerns include how it affects people's employment opportunities, its potential to enable mass surveillance, and its role in facilitating access to basic goods and services, among others. In response, the organisations that design, develop and deploy AI technologies – often with limited input from those most affected by its operation – have attempted to quell people's fears by setting out how they are approaching AI in a fair and ethical manner. Since around 2018, this has led to a deluge of ethical AI principles, guidelines, frameworks and declarations being published by both private organisations and government agencies around the world.

Companies are always looking for practical AI tools to make recruitment easier and more efficient. With increasing numbers of applications to screen, recruiters often struggle to streamline a growing workload, making AI recruiting tools an appealing solution. More companies are turning to automated video interview software to minimize hire time, help recruiters screen more applicants fairly and consistently, and reduce inherent bias in the interview process. No technology can replace the role of a recruiter, however. It's essential to evaluate the effectiveness of AI tools and how they're best used to ensure they'll benefit your hiring practices.