Many scientific conferences employ a two-phase paper review process, where some papers are assigned additional reviewers after the initial reviews are submitted. Many conferences also design and run experiments on their paper review process, where some papers are assigned reviewers who provide reviews under an experimental condition. In this paper, we consider the question: how should reviewers be divided between phases or conditions in order to maximize total assignment similarity? We make several contributions towards answering this question. First, we prove that when the set of papers requiring additional review is unknown, a simplified variant of this problem is NP-hard. Second, we empirically show that across several datasets pertaining to real conference data, dividing reviewers between phases/conditions uniformly at random allows an assignment that is nearly as good as the oracle optimal assignment. This uniformly random choice is practical for both the two-phase and conference experiment design settings. Third, we provide explanations of this phenomenon by providing theoretical bounds on the suboptimality of this random strategy under certain natural conditions. From these easily-interpretable conditions, we provide actionable insights to conference program chairs about whether a random reviewer split is suitable for their conference.
Citizen science projects that rely on human computation can attempt to solicit volunteers or use paid microwork platforms such as Amazon Mechanical Turk. To better understand these approaches, this paper analyzes crowdsourced image label data sourced from an environmental justice project looking at wetland loss off the coast of Louisiana. This retrospective analysis identifies key differences between the two populations: while Mechanical Turk workers are accessible, cost-efficient, and rate more images than volunteers (on average), their labels are of lower quality, whereas volunteers can achieve high accuracy with comparably few votes. Volunteer organizations can also interface with the educational or outreach goals of an organization in ways that the limited context of microwork prevents.
Model explanations are generated by XAI (explainable AI) methods to help people understand and interpret machine learning models. To study XAI methods from the human perspective, we propose a human-based benchmark dataset, i.e., human saliency benchmark (HSB), for evaluating saliency-based XAI methods. Different from existing human saliency annotations where class-related features are manually and subjectively labeled, this benchmark collects more objective human attention on vision information with a precise eye-tracking device and a novel crowdsourcing experiment. Taking the labor cost of human experiment into consideration, we further explore the potential of utilizing a prediction model trained on HSB to mimic saliency annotating by humans. Hence, a dense prediction problem is formulated, and we propose an encoder-decoder architecture which combines multi-modal and multi-scale features to produce the human saliency maps. Accordingly, a pretraining-finetuning method is designed to address the model training problem. Finally, we arrive at a model trained on HSB named human saliency imitator (HSI). We show, through an extensive evaluation, that HSI can successfully predict human saliency on our HSB dataset, and the HSI-generated human saliency dataset on ImageNet showcases the ability of benchmarking XAI methods both qualitatively and quantitatively.
Predictive models are increasingly used to make various consequential decisions in high-stakes domains such as healthcare, finance, and policy. It becomes critical to ensure that these models make accurate predictions, are robust to shifts in the data, do not rely on spurious features, and do not unduly discriminate against minority groups. To this end, several approaches spanning various areas such as explainability, fairness, and robustness have been proposed in recent literature. Such approaches need to be human-centered as they cater to the understanding of the models to their users. However, there is little to no research on understanding the needs and challenges in monitoring deployed machine learning (ML) models from a human-centric perspective. To address this gap, we conducted semi-structured interviews with 13 practitioners who are experienced with deploying ML models and engaging with customers spanning domains such as financial services, healthcare, hiring, online retail, computational advertising, and conversational assistants. We identified various human-centric challenges and requirements for model monitoring in real-world applications. Specifically, we found that relevant stakeholders would want model monitoring systems to provide clear, unambiguous, and easy-to-understand insights that are readily actionable. Furthermore, our study also revealed that stakeholders desire customization of model monitoring systems to cater to domain-specific use cases.
Different input modalities have been proposed and employed in technological landscapes like microtask crowdsourcing. However, sign language remains an input modality that has received little attention. Despite the fact that thousands of people around the world primarily use sign language, very little has been done to include them in such technological landscapes. We aim to address this gap and take a step towards the inclusion of deaf and mute people in microtask crowdsourcing. We first identify various microtasks which can be adapted to use sign language as input, while elucidating the challenges it introduces. We built a system called ‘SignUpCrowd’ that can be used to support sign language input for microtask crowdsourcing. We carried out a between-subjects study (N=240) to understand the effectiveness of sign language as an input modality for microtask crowdsourcing in comparison to prevalent textual and click input modalities. We explored this through the lens of visual question answering and sentiment analysis tasks by recruiting workers from the Prolific crowdsourcing platform. Our results indicate that sign language as an input modality in microtask crowdsourcing is comparable to the prevalent standards of using text and click input. Although people with no knowledge of sign language found it difficult to use, this input modality has the potential to broaden participation in crowd work. We highlight evidence suggesting the scope for sign language as a viable input type for microtask crowdsourcing. Our findings pave the way for further research to introduce sign language in real-world applications and create an inclusive technological landscape that more people can benefit from.
Many AI system designers grapple with how best to collect human input for different types of training data. Online crowds provide a cheap on-demand source of intelligence, but they often lack the expertise required in many domains. Experts offer tacit knowledge and more nuanced input, but they are harder to recruit. To explore this trade off, we compared novices and experts in terms of performance and perceptions on human intelligence tasks in the context of designing a text-based conversational agent. We developed a preliminary chatbot that simulates conversations with someone seeking mental health advice to help educate volunteer listeners at 7cups.com. We then recruited experienced listeners (domain experts) and MTurk novice workers (crowd workers) to conduct tasks to improve the chatbot with different levels of complexity. Novice crowds perform comparably to experts on tasks that only require natural language understanding, such as correcting how the system classifies a user statement. For more generative tasks, like creating new lines of chatbot dialogue, the experts demonstrated higher quality, novelty, and emotion. We also uncovered a motivational gap: crowd workers enjoyed the interactive tasks, while experts found the work to be tedious and repetitive. We offer design considerations for allocating crowd workers and experts on input tasks for AI systems, and for better motivating experts to participate in low-level data work for AI.
Clear instructions are a necessity for obtaining accurate results from crowd workers. Even small ambiguities can force workers to choose an interpretation arbitrarily, resulting in errors and inconsistency. Crisp instructions require significant time to design, test, and iterate. Recent approaches have engaged workers to detect and correct ambiguities. However, this process increases the time and money required to obtain accurate, consistent results. We present TaskLint, a system to automatically detect problems with task instructions. Leveraging a diverse set of existing NLP tools, TaskLint identifies words and sentences that might foretell worker confusion. This is analogous to static analysis tools for code ("linters"), which detect possible features in code that might indicate the presence of bugs. Our evaluation of TaskLint using task instructions created by novices confirms the potential for static tools to improve task clarity and the accuracy of results, while also highlighting several challenges.
Gerrymandering is the manipulation of redistricting to influence the results of a set of elections for local representatives. Gerrymandering has the potential to drastically swing power in legislative bodies even with no change in a population’s political views. Identifying gerrymandering and measuring fairness using metrics of proposed district plans is a topic of current research, but there is less work on how such plans will be perceived by voters. Gathering data on such perceptions presents several challenges such as the ambiguous definitions of ‘fair’ and the complexity of real world geography and district plans. We present a dataset collected from an online crowdsourcing platform on a survey asking respondents to mark which of two maps of equal population distribution but different districts appear more ‘fair’ and the reasoning for their decision. We performed preliminary analysis on this data and identified which of several commonly suggested metrics are most predictive of the responses. We found that the maximum perimeter of any district was the most predictive metric, especially with participants who reported that they made their decision based on the shape of the districts.
With recent advances in explainable artificial intelligence (XAI), researchers have started to pay attention to concept-level explanations, which explain model predictions with a high level of abstraction. However, such explanations may be difficult to digest for laypeople due to the potential knowledge gap and the concomitant cognitive load. Inspired by recent work, we argue that analogy-based explanations composed of commonsense knowledge may be a potential solution to tackle this issue. In this paper, we propose analogical inference as a bridge to help end-users leverage their commonsense knowledge to better understand the concept-level explanations. Specifically, we design an effective analogy-based explanation generation method and collect 600 analogy-based explanations from 100 crowd workers. Furthermore, we propose a set of structured dimensions for the qualitative assessment of analogy-based explanations and conduct an empirical evaluation of the generated analogies with experts. Our findings reveal significant positive correlations between the qualitative dimensions of analogies and the perceived helpfulness of analogy-based explanations. These insights can inform the design of future methods for the generation of effective analogy-based explanations. We also find that the understanding of commonsense explanations varies with the experience of the recipient user, which points out the need for further work on personalization when leveraging commonsense explanations.
Human input is pivotal in building reliable and robust artificial intelligence systems. By providing a means to gather diverse, high-quality, representative, and cost-effective human input on demand, microtask crowdsourcing marketplaces have thrived. Despite the unmistakable benefits available from online crowd work, the lack of health provisions and safeguards, along with existing work practices threatens the sustainability of this paradigm. Prior work has investigated worker engagement and mental health, yet no such investigations into the effects of crowd work on the physical health of workers have been undertaken. Crowd workers complete their work in various sub-optimal work environments, often using a conventional input modality of a mouse and keyboard. The repetitive nature of microtask crowdsourcing can lead to stress-related injuries, such as the well-documented carpal tunnel syndrome. It is known that stretching exercises can help reduce injuries and discomfort in office workers. Gestures, the act of using the body intentionally to affect the behavior of an intelligent system, can serve as both stretches and an alternative form of input for microtasks. To better understand the usefulness of the dual-purpose input modality of ergonomically-informed gestures across different crowdsourced microtasks, we carried out a controlled 2 x 3 between-subjects study (N=294). Considering the potential benefits of gestures as an input modality, our results suggest a real trade-off between worker accuracy in exchange for potential short to long-term health benefits.
Explaining the behaviour of Artificial Intelligence models has become a necessity. Their opaqueness and fragility are not tolerable in high-stakes domains especially. Although considerable progress is being made in the field of Explainable Artificial Intelligence, scholars have demonstrated limits and flaws of existing approaches: explanations requiring further interpretation, non-standardised explanatory format, and overall fragility. In light of this fragmentation, we turn to the field of philosophy of science to understand what constitutes a good explanation, that is, a generalisation that covers both the actual outcome and, possibly multiple, counterfactual outcomes. Inspired by this, we propose CHIME: a human-in-the-loop, post-hoc approach focused on creating such explanations by establishing the causal features in the input. We first elicit people's cognitive abilities to understand what parts of the input the model might be attending to. Then, through Causal Discovery we uncover the underlying causal graph relating the different concepts. Finally, with such a structure, we compute the causal effects different concepts have towards a model's outcome. We evaluate the Fidelity, Coherence, and Accuracy of the explanations obtained with CHIME with respect to two state-of-the-art Computer Vision models trained on real-world image data sets. We found evidence that the explanations reflect the causal concepts tied to a model's prediction, both in terms of causal strength and accuracy.
A wealth of evidence across several domains indicates that goal setting improves performance and learning by enabling individuals to commit their thoughts and actions to goal achievement. Recently, researchers have begun studying the effects of goal setting in paid crowdsourcing to improve the quality and quantity of contributions, increase learning gains, and hold participants accountable for contributing more effectively. However, there is a lack of research addressing crowd workers' goal-setting practices, how they are currently pursuing them, and the challenges that they face. This information is essential for researchers and developers to create tools that assist crowd workers in pursuing their goals more effectively, thereby improving the quality of their contributions. This paper addresses these gaps by conducting mixed-method research in which we surveyed 205 workers from two crowdsourcing platforms -- Amazon Mechanical Turk (MTurk) and Prolific -- about their goal-setting practices. Through a 14-item survey, we asked workers regarding the types of goals they create, their goal achievement strategies, potential barriers that impede goal attainment, and their use of software tools for effective goal management. We discovered that (a) workers actively create intrinsic and extrinsic goals; (b) use a combination of tools for goal management; (c) medical issues and a busy lifestyle are some obstacles to their goal achievement; and (d) we gathered novel features for future goal management tools. Our findings shed light on the broader implications of developing goal management tools to improve workers' well-being.
Many applications such as hiring and university admissions involve evaluation and selection of applicants. These tasks are fundamentally difficult, and require combining evidence from multiple different aspects (what we term "attributes"). In these applications, the number of applicants is often large, and a common practice is to assign the task to multiple evaluators in a distributed fashion. Specifically, in the often-used holistic allocation, each evaluator is assigned a subset of the applicants, and is asked to assess all relevant information for their assigned applicants. However, such an evaluation process is subject to issues such as miscalibration (evaluators see only a small fraction of the applicants and may not get a good sense of relative quality), and discrimination (evaluators are influenced by irrelevant information about the applicants). We identify that such attribute-based evaluation allows alternative allocation schemes. Specifically, we consider assigning each evaluator more applicants but fewer attributes per applicant, termed segmented allocation. We compare segmented allocation to holistic allocation on several dimensions via theoretical and experimental methods. We establish various tradeoffs between these two approaches, and identify conditions under which one approach results in more accurate evaluation than the other.
Machine learning algorithms are increasingly used to assist human decision-making. When the goal of machine assistance is to improve the accuracy of human decisions, it might seem appealing to design ML algorithms that complement human knowledge. While neither the algorithm nor the human are perfectly accurate, one could expect that their complementary expertise might lead to improved outcomes. In this study, we demonstrate that in practice decision aids that are not complementary, but make errors similar to human ones may have their own benefits. In a series of human-subject experiments with a total of 901 participants, we study how the similarity of human and machine errors influences human perceptions of and interactions with algorithmic decision aids. We find that (i) people perceive more similar decision aids as more useful, accurate, and predictable, and that (ii) people are more likely to take opposing advice from more similar decision aids, while (iii) decision aids that are less similar to humans have more opportunities to provide opposing advice, resulting in a higher influence on people’s decisions overall.
An increased awareness concerning risks of algorithmic bias has driven a surge of efforts around bias mitigation strategies. A vast majority of the proposed approaches fall under one of two categories: (1) imposing algorithmic fairness constraints on predictive models, and (2) collecting additional training samples. Most recently and at the intersection of these two categories, methods that propose active learning under fairness constraints have been developed. However, proposed bias mitigation strategies typically overlook the bias presented in the observed labels. In this work, we study fairness considerations of active data collection strategies in the presence of label bias. We first present an overview of different types of label bias in the context of supervised learning systems. We then empirically show that, when overlooking label bias, collecting more data can aggravate bias, and imposing fairness constraints that rely on the observed labels in the data collection process may not address the problem. Our results illustrate the unintended consequences of deploying a model that attempts to mitigate a single type of bias while neglecting others, emphasizing the importance of explicitly differentiating between the types of bias that fairness-aware algorithms aim to address, and highlighting the risks of neglecting label bias during data collection.
The labels used to train machine learning (ML) models are of paramount importance. Typically for ML classification tasks, datasets contain hard labels, yet learning using soft labels has been shown to yield benefits for model generalization, robustness, and calibration. Earlier work found success in forming soft labels from multiple annotators' hard labels; however, this approach may not converge to the best labels and necessitates many annotators, which can be expensive and inefficient. We focus on efficiently eliciting soft labels from individual annotators. We collect and release a dataset of soft labels (which we call CIFAR-10S) over the CIFAR-10 test set via a crowdsourcing study (N=248). We demonstrate that learning with our labels achieves comparable model performance to prior approaches while requiring far fewer annotators -- albeit with significant temporal costs per elicitation. Our elicitation methodology therefore shows nuanced promise in enabling practitioners to enjoy the benefits of improved model performance and reliability with fewer annotators, and serves as a guide for future dataset curators on the benefits of leveraging richer information, such as categorical uncertainty, from individual annotators.
Several election districts in the US have recently moved to ranked-choice voting (RCV) to decide the results of local elections. RCV allows voters to rank their choices, and the results are computed in rounds, eliminating one candidate at a time. RCV ensures fairer elections and has been shown to increase elected representation of women and people of color. A main drawback of RCV is that the round-by-round process requires all the ballots to be tallied before the results of an election can be calculated. With increasingly large portions of ballots coming from absentee voters, RCV election outcomes are not always apparent on election night, and can take several weeks to be published, leading to a loss of trust in the electoral process from the public. In this paper, we present an algorithm for efficiently computing possible winners of RCV elections from partially known ballots and evaluate it on data from the recent New York City Primary elections. We show that our techniques allow to significantly narrow down the field of possible election winners, and in some case identify the winner as soon as election night despite a number of yet-unaccounted absentee ballots, providing more transparency in the electoral process.
Recent years have seen a surge of interest in the field of explainable AI (XAI), with a plethora of algorithms proposed in the literature. However, a lack of consensus on how to evaluate XAI hinders the advancement of the field. We highlight that XAI is not a monolithic set of technologies---researchers and practitioners have begun to leverage XAI algorithms to build XAI systems that serve different usage contexts, such as model debugging and decision-support. Algorithmic research of XAI, however, often does not account for these diverse downstream usage contexts, resulting in limited effectiveness or even unintended consequences for actual users, as well as difficulties for practitioners to make technical choices. We argue that one way to close the gap is to develop evaluation methods that account for different user requirements in these usage contexts. Towards this goal, we introduce a perspective of contextualized XAI evaluation by considering the relative importance of XAI evaluation criteria for prototypical usage contexts of XAI. To explore the context dependency of XAI evaluation criteria, we conduct two survey studies, one with XAI topical experts and another with crowd workers. Our results urge for responsible AI research with usage-informed evaluation practices, and provide a nuanced understanding of user requirements for XAI in different usage contexts.
Many researchers and policymakers have expressed excitement about algorithmic explanations enabling more fair and responsible decision-making. However, recent experimental studies have found that explanations do not always improve human use of algorithmic advice. In this study, we shed light on how people interpret and respond to counterfactual explanations (CFEs) -- explanations that show how a model's output would change with marginal changes to its input(s) -- in the context of pretrial risk assessment instruments (PRAIs). We ran think-aloud trials with eight sitting U.S. state court judges, providing them with recommendations from a PRAI that includes CFEs. We found that the CFEs did not alter the judges' decisions. At first, judges misinterpreted the counterfactuals as real -- rather than hypothetical -- changes to defendants. Once judges understood what the counterfactuals meant, they ignored them, stating their role is only to make decisions regarding the actual defendant in question. The judges also expressed a mix of reasons for ignoring or following the advice of the PRAI without CFEs. These results add to the literature detailing the unexpected ways in which people respond to algorithms and explanations. They also highlight new challenges associated with improving human-algorithm collaborations through explanations.
Strategic behavior is a fundamental problem in a variety of real-world applications that require some form of peer assessment, such as peer grading of homeworks, grant proposal review, conference peer review of scientific papers, and peer assessment of employees in organizations. Since an individual’s own work is in competition with the submissions they are evaluating, they may provide dishonest evaluations to increase the relative standing of their own submission. This issue is typically addressed by partitioning the individuals and assigning them to evaluate the work of only those from different subsets. Although this method ensures strategyproofness, each submission may require a different type of expertise for effective evaluation. In this paper, we focus on finding an assignment of evaluators to submissions that maximizes assigned evaluators’ expertise subject to the constraint of strategyproofness. We analyze the price of strategyproofness: that is, the amount of compromise on the assigned evaluators’ expertise required in order to get strategyproofness. We establish several polynomial-time algorithms for strategyproof assignment along with assignment-quality guarantees. Finally, we evaluate the methods on a dataset from conference peer review.
