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 slightly 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.

On Human Predictions with Explanations and Predictions of Machine Learning Models: A Case Study on Deception Detection

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 slightly 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.

Massive dissemination of fake news and its potential to erode democracy has increased the demand for accurate fake news detection. Recent advancements in this area have proposed novel techniques that aim to detect fake news by exploring how it propagates on social networks. Nevertheless, to detect fake news at an early stage, i.e., when it is published on a news outlet but not yet spread on social media, one cannot rely on news propagation information as it does not exist. Hence, there is a strong need to develop approaches that can detect fake news by focusing on news content. In this article, a theory-driven model is proposed for fake news detection. The method investigates news content at various levels: lexicon-level, syntax-level, semantic-level, and discourse-level. We represent news at each level, relying on well-established theories in social and forensic psychology. Fake news detection is then conducted within a supervised machine learning framework. As an interdisciplinary research, our work explores potential fake news patterns, enhances the interpretability in fake news feature engineering, and studies the relationships among fake news, deception/disinformation, and clickbaits. Experiments conducted on two real-world datasets indicate the proposed method can outperform the state-of-the-art and enable fake news early detection when there is limited content information.

Fake News Early Detection: A Theory-driven Model

A new era of Information Warfare has arrived. Various actors, including state-sponsored ones, are weaponizing information on Online Social Networks to run false-information campaigns with targeted manipulation of public opinion on specific topics. These false-information campaigns can have dire consequences to the public: mutating their opinions and actions, especially with respect to critical world events like major elections. Evidently, the problem of false information on the Web is a crucial one and needs increased public awareness as well as immediate attention from law enforcement agencies, public institutions, and in particular, the research community. In this article, we make a step in this direction by providing a typology of the Web’s false-information ecosystem, composed of various types of false-information, actors, and their motives. We report a comprehensive overview of existing research on the false-information ecosystem by identifying several lines of work: (1) how the public perceives false information; (2) understanding the propagation of false information; (3) detecting and containing false information on the Web; and (4) false information on the political stage. In this work, we pay particular attention to political false information as: (1) it can have dire consequences to the community (e.g., when election results are mutated) and (2) previous work shows that this type of false information propagates faster and further when compared to other types of false information. Finally, for each of these lines of work, we report several future research directions that can help us better understand and mitigate the emerging problem of false-information dissemination on the Web.

The Web of False Information: Rumors, Fake News, Hoaxes, Clickbait, and Various Other Shenanigans

There may be as much philosophical controversy about how to distinguish naturalism from non-naturalism as there is about which view is correct. In spite of this widespread disagreement about the content of naturalism and nonnaturalism there is considerable agreement about the status of certain historically influential philosophical accounts as non-naturalist. In particular, there is widespread agreement that G.E. Moore's account of goodness in Principia Ethica is a paradigmatically non-naturalist account. Indeed, if a representative sample of contemporary philosophers were asked to name a non-naturalist in meta-ethics then Moore's name almost certainly would predominate. For better or worse, Moore's discussion of non-naturalism profoundly shaped 20 century meta-ethics. Thomas Baldwin was not exaggerating much when he claimed that, “twentieth century British ethical theory is unintelligible without reference to Principia Ethica; its history until 1960 or so being, in brief, that although Moore was taken to have refuted ‘ethical naturalism’, Moore's own brand of ‘ethical non-naturalism’ was thought to make unacceptable metaphysical and epistemological demands; so the only recourse was to abandon belief in an objective moral reality and accept an emotivist, prescriptivist or otherwise anti-realist, account of ethical values.” (Baldwin 1990: 66) More recent defenders of non-naturalist theories (e.g. Shafer-Landau 2003, Enoch 2011, and Parfit 2011) often distinguish their accounts from Moore's in a number of ways, but few if any would dispute that Moore's account is a paradigmatically non-naturalist one.

The naturalistic fallacy

The proliferation of misinformation in online news and its amplification by platforms are a growing concern, leading to numerous efforts to improve the detection of and response to misinformation. Given the variety of approaches, collective agreement on the indicators that signify credible content could allow for greater collaboration and data-sharing across initiatives. In this paper, we present an initial set of indicators for article credibility defined by a diverse coalition of experts. These indicators originate from both within an article's text as well as from external sources or article metadata. As a proof-of-concept, we present a dataset of 40 articles of varying credibility annotated with our indicators by 6 trained annotators using specialized platforms. We discuss future steps including expanding annotation, broadening the set of indicators, and considering their use by platforms and the public, towards the development of interoperable standards for content credibility.

A Structured Response to Misinformation: Defining and Annotating Credibility Indicators in News Articles

We investigate how a shepherd should move in order to eﬀectively herd and guide a ﬂock of agents towards a target. Using a detailed agent-based model (ABM) for the members of the ﬂock, we pose and solve an optimization problem for the shepherd that has to simultaneously work to keep the ﬂock cohesive while coercing it towards a prescribed project. We ﬁnd that three distinct strategies emerge as potential solutions as a function of just two parameters: the ratio of herd size to shepherd repulsion length and the ratio of herd speed to shepherd speed. We term these as: (i) mustering, in which the shepherd circles the herd to ensure compactness, (ii) droving, in which the shepherd chases the herd in a desired direction, and (iii) driving, a hitherto unreported strategy where the ﬂock surrounds a shepherd that drives it from within. A minimal dynamical model for the size, shape and position of the herd captures the eﬀective behavior of the ABM, and further allows us to characterize the diﬀerent herding strategies in terms of the behavior of the shepherd that librates (mustering), oscillates (droving) or moves steadily (driving). All together, our study yields a simple and intuitive classiﬁcation of herding strategies that ought to be of general interest in the context of controlling the collective behavior of active matter.

Aditya Ranganathan

Papers

A Structured Response to Misinformation: Defining and Annotating Credibility Indicators in News Articles

Optimal shepherding and transport of a flock