Combating widespread misinformation requires scalable and reliable fact-checking methods. Fact-checking involves several steps, including question generation, evidence retrieval, and veracity prediction. Importantly, fact-checking is well-suited to exploit hybrid intelligence since it requires both human expertise and AI’s large-scale information processing abilities. Thus, constructing an effective fact-checking pipeline requires a systematic understanding of the relative strengths and weaknesses of humans and AI in different steps of the fact-checking process. We investigate the ability of LLMs to perform the first step of the process, i.e., to generate pertinent questions for analyzing a claim. To evaluate the quality of the LLM-generated questions, we crowdsource a dataset in which 150 claims are annotated with questions (1) a novice fact-checker would ask and (2) a professional fact-checker would ask when fact-checking those claims. We study the effects of the human- and LLM-generated questions on evidence retrieval and veracity prediction. We find that LLMs are able to generate nuanced questions to verify a complex claim, but the final label prediction depends on the quality of the evidence corpus. However, the evidence collected by automated methods yields lower accuracy in the veracity prediction task than the evidence curated by experts.

In today's society, where Artificial Intelligence (AI) has gained a vital role, concerns regarding user's trust have garnered significant attention. The use of AI systems in high-risk domains have often led users to either under-trust it, potentially causing inadequate reliance or over-trust it, resulting in over-compliance. Therefore, users must maintain an appropriate level of trust. Past research has indicated that explanations provided by AI systems can enhance user understanding of when to trust or not trust the system. However, the utility of presentation of different explanations forms still remains to be explored especially in high-risk domains. Therefore, this study explores the impact of different explanation types (text, visual, and hybrid) and user expertise (retired police officers and lay users) on establishing appropriate trust in AI-based predictive policing. While we observed that the hybrid form of explanations increased the subjective trust in AI for expert users, it did not led to better decision-making. Furthermore, no form of explanations helped build appropriate trust. The findings of our study emphasize the importance of re-evaluating the use of explanations to build [appropriate] trust in AI based systems especially when the system's use is questionable. Finally, we synthesize potential challenges and policy recommendations based on our results to design for appropriate trust in high-risk based AI-based systems.

Mutual trust between humans and interactive artificial agents is crucial for effective human-agent teamwork. This involves not only the human appropriately trusting the artificial teammate, but also the artificial teammate assessing the human’s trustworthiness for different tasks (i.e., artificial trust in human partners). Literature indicated that transparency and explainability is generally beneficial for human-agent collaboration. However, communicating artificial trust potentially affects human trust and satisfaction, which impact team dynamics. Towards studying these effects, we developed an artificial trust model and implemented five distinct communication approaches which varied in modality (visual/graphical and/or text), level (communication and/or explanation), and timing (real-time or occasional). We evaluated the effects of the different communication styles through a user study (N=120) in a 2D grid-world Search and Rescue scenario. Our results show that all our artificial trust explanations improved human trust and satisfaction, but the mere graphical communication of it did not. These results are bound to the specific scenario and context in which this study was run and require further exploration. As such, this work presents a first step towards understanding the consequences of communicating and explaining to a human teammate their assessed trustworthiness.

The roles of humans and AI as the labor force of organizations need continuous re-evaluation with the advancement of AI. While automation has replaced some tasks, knowledge-intensive work environments rely on human intelligence, as those work practices transcend canonical procedures. We propose a hybrid intelligence methodology for organizations to address knowledge erosion. We contextualize this methodology in an example case study from the Legal Desk in the Netherlands, following the six principles of designing intelligent organizations [1], i.e., addition, relevance, substitution, diversity, collaboration, and explanation. We found that adhering to these six basic principles appeared to be a balancing act on two axes: contribution of AI versus human intelligence towards the tasks, and the way of working of human and artificial agents over time. We propose two additional principles. The first is human oversight, which highlights the importance of human control in organizational decision-making. The second principle is collaborative reflection which emphasizes the need to actively manage organizational intelligence. We also discuss the challenges to enable our methodology in the organizational context. This paper aims to inspire researchers and practitioners to pursue new initiatives towards achieving hybrid intelligence for learning organizations.

Understanding citizens’ values in participatory systems is crucial for citizen-centric policy-making. We envision a hybrid participatory system where participants make choices and provide motivations for those choices, and AI agents estimate their value preferences by interacting with them. We focus on situations where a conflict is detected between participants’ choices and motivations, and propose methods for estimating value preferences while addressing detected inconsistencies by interacting with the participants. We operationalize the philosophical stance that “valuing is deliberatively consequential.” That is, if a participant’s choice is based on a deliberation of value preferences, the value preferences can be observed in the motivation the participant provides for the choice. Thus, we propose and compare value preferences estimation methods that prioritize the values estimated from motivations over the values estimated from choices alone. Then, we introduce a disambiguation strategy that combines Natural Language Processing and Active Learning to address the detected inconsistencies between choices and motivations. We evaluate the proposed methods on a dataset of a large-scale survey on energy transition. The results show that explicitly addressing inconsistencies between choices and motivations improves the estimation of an individual’s value preferences. The disambiguation strategy does not show substantial improvements when compared to similar baselines—however, we discuss how the novelty of the approach can open new research avenues and propose improvements to address the current limitations.

A lack of self-awareness of communicative behaviours can lead to disadvantages in important interactions. Video recordings as a tool for self-observation have been widely adopted to initiate behaviour change and reflection. Seeing oneself in a recording can lead to negative affect. Forcing an external perspective can lead to cognitive dissonance. Avatars and virtual agents have the advantage that they can copy a human's behaviour while potentially avoiding this dissonance. To explore the design space of mimicking agents, we set up a user study where a video baseline is compared to agent-mediated conditions ranging from idle non-verbal behaviour to complete mimicry of the voice and face. We show that participants gain increased self-awareness from seeing themselves mediated through the virtual agent. We further discuss qualitative observations for the future design of systems that aid in self-reflection, and particularly note that partial mimicry seems to be less appreciated than full mimicry.

Effective support from personal assistive technologies relies on accurate user models that capture user values, preferences, and context. Knowledge-based techniques model these relationships, enabling support agents to align their actions with user values. However, understanding values in a single context is insufficient due to the dynamic nature of behaviour. This study explores the use of dialogue strategies to update user models. Participants were randomly assigned to different strategies and they discussed one randomly chosen non-adherence situation with the agent. Then, their emotions, acquired information accuracy, completeness, and dialogue experience were rated. Our findings suggest that multiple-choice dialogues may limit response depth, reducing the perceived completeness of behaviour reasons. In contrast, open-ended questions allow more detailed input but require more time and effort, potentially worsening the dialogue experience. Through inductive coding, we identified key topics, such as individual challenges, priorities, tangible outcomes, and values, essential for constructing personalised user models. We also analyzed conversation paths to improve dialogue-based user model updates in support agents. Further research is needed to refine the relationship between dialogue strategies and self-conscious emotions, considering diverse backgrounds and health goals, while enhancing dialogue design.

Aggregating multiple annotations into a single ground truth label may hide valuable insights into annotator disagreement, particularly in tasks where subjectivity plays a crucial role. In this work, we explore methods for identifying subjectivity in recognizing the human values that motivate arguments. We evaluate two main approaches: inferring subjectivity through value prediction vs. directly identifying subjectivity. Our experiments show that direct subjectivity identification significantly improves the model performance of flagging subjective arguments. Furthermore, combining contrastive loss with binary cross-entropy loss does not improve performance but reduces the dependency on per-label subjectivity. Our proposed methods can help identify arguments that individuals may interpret differently, fostering a more nuanced annotation process.

Understanding behavior of human drivers in interactions with automated vehicles (AV) can aid the development of future AVs. Existing investigations of such behavior have predominantly focused on situations in which an AV a priori needs to take action because the human has the right of way. However, future AVs might need to proactively manage interactions even if they have the right of way over humans, e.g., a human driver taking a left turn in front of the approaching AV. Yet it remains unclear how AVs could behave in such interactions and how humans would react to them. To address this issue, here we investigated behavior of human drivers (N = 19) when interacting with an oncoming AV during unprotected left turns in a driving simulator experiment. We measured the outcomes (Go or Stay) and timing of participants’ decisions when interacting with an AV which performed subtle longitudinal nudging maneuvers, e.g. briefly decelerating and then accelerating back to its original speed. We found that participants’ behavior was sensitive to deceleration nudges but not acceleration nudges. We compared the obtained data to predictions of several variants of a drift-diffusion model of human decision making. The most parsimonious model that captured the data hypothesized noisy integration of dynamic information on time-to-arrival and distance to a fixed decision boundary, with an initial accumulation bias towards the Go decision. Our model not only accounts for the observed behavior but can also flexibly generate predictions of human responses to arbitrary longitudinal AV maneuvers, and can be used for both informing future studies of human behavior and incorporating insights from such studies into computational frameworks for AV interaction planning.

Disagreements are common in online societal deliberation and may be crucial for effective collaboration, for instance in helping users understand opposing viewpoints. Although there exist automated methods for recognizing disagreement, a deeper understanding of factors that influence disagreement is currently missing. We investigate a hypothesis that differences in personal values influence disagreement in online discussions. Using Large Language Models (LLMs) for estimating both profiles of personal values and disagreement, we conduct a large-scale experiment involving 11.4M user comments. We find that the dissimilarity of value profiles correlates with disagreement only in specific cases, but that incorporating self-reported value profiles changes these results to be more undecided.

Purpose: The objective of this paper is to develop a redesigned commissioning process for social care services that fosters integrated care, encourages collaboration and balances professional expertise with client engagement. Design/methodology/approach: This study employs a two-pronged approach: a case study of a municipality’s use of subsidy tables and a literature scoping review on integrated care research. Findings: The paper introduces a new framework for the study of the new “subsidy tables.” A well-defined and extensive consultation process involving both social care providers (suppliers), the Service Triad, and client representation adds to the existing research on supplier consultation, and on how to define the outcomes for clients via client engagement. Research limitations/implications: While aspects are clearly relevant to the Netherlands, the design of the commissioning process of social care has international relevance as well: finding definitions, formulating outcomes and incentives, designing a more collaborative instead of competitive process, stakeholder engagement and consultation. Practical implications: Several Dutch municipalities started using the “subsidy tables” method for commissioning integrated social care. This paper offers clear improvements that benefit the commissioners, the social care providers and their clients. Social implications: Improving the commissioning process of integrated social care will lead to better fitting care for people who need social care. Originality/value: This paper is one of the first to do a thorough analysis of the “subsidy tables” method for commissioning integrated social care.

Presenting high-level arguments is a crucial task for fostering participation in online societal discussions. Current argument summarization approaches miss an important facet of this task-capturing diversity-which is important for accommodating multiple perspectives. We introduce three aspects of diversity: those of opinions, annotators, and sources. We evaluate approaches to a popular argument summarization task called Key Point Analysis, which shows how these approaches struggle to (1) represent arguments shared by few people, (2) deal with data from various sources, and (3) align with subjectivity in human-provided annotations. We find that both general-purpose LLMs and dedicated KPA models exhibit this behavior, but have complementary strengths. Further, we observe that diversification of training data may ameliorate generalization. Addressing diversity in argument summarization requires a mix of strategies to deal with subjectivity.

Epistemic logic can be used to reason about statements such as ‘I know that you know that I know that φ ’. In this logic, and its extensions, it is commonly assumed that agents can reason about epistemic statements of arbitrary nesting depth. In contrast, empirical findings on Theory of Mind, the ability to (recursively) reason about mental states of others, show that human recursive reasoning capability has an upper bound. In the present paper we work towards resolving this disparity by proposing some elements of a logic of bounded Theory of Mind, built on Public Announcement Logic. Using this logic, and a statistical method called Random-Effects Bayesian Model Selection, we estimate the distribution of Theory of Mind levels in the participant population of a previous behavioral experiment. Despite not modeling stochastic behavior, we find that approximately three-quarters of participants’ decisions can be described using Theory of Mind. In contrast to previous empirical research, our models estimate the majority of participants to be second-order Theory of Mind users.

Appropriate Trust in Artificial Intelligence (AI) systems has rapidly become an important area of focus for both researchers and practitioners. Various approaches have been used to achieve it, such as confidence scores, explanations, trustworthiness cues, or uncertainty communication. However, a comprehensive understanding of the field is lacking due to the diversity of perspectives arising from various backgrounds that influence it and the lack of a single definition for appropriate trust. To investigate this topic, this paper presents a systematic review to identify current practices in building appropriate trust, different ways to measure it, types of tasks used, and potential challenges associated with it. We also propose a Belief, Intentions, and Actions (BIA) mapping to study commonalities and differences in the concepts related to appropriate trust by (a) describing the existing disagreements on defining appropriate trust, and (b) providing an overview of the concepts and definitions related to appropriate trust in AI from the existing literature. Finally, the challenges identified in studying appropriate trust are discussed, and observations are summarized as current trends, potential gaps, and research opportunities for future work. Overall, the paper provides insights into the complex concept of appropriate trust in human-AI interaction and presents research opportunities to advance our understanding on this topic.

Large-scale survey tools enable the collection of citizen feedback in opinion corpora. Extracting the key arguments from a large and noisy set of opinions helps in understanding the opinions quickly and accurately. Fully automated methods can extract arguments but (1) require large labeled datasets that induce large annotation costs and (2) work well for known viewpoints, but not for novel points of view. We propose HyEnA, a hybrid (human + AI) method for extracting arguments from opinionated texts, combining the speed of automated processing with the understanding and reasoning capabilities of humans. We evaluate HyEnA on three citizen feedback corpora. We find that, on the one hand, HyEnA achieves higher coverage and precision than a state-of-The-Art automated method when compared to a common set of diverse opinions, justifying the need for human insight. On the other hand, HyEnA requires less human effort and does not compromise quality compared to (fully manual) expert analysis, demonstrating the benefit of combining human and artificial intelligence.

Appropriate trust is an important component of the interaction between people and AI systems, in that "inappropriate"trust can cause disuse, misuse, or abuse of AI. To foster appropriate trust in AI, we need to understand how AI systems can elicit appropriate levels of trust from their users. Out of the aspects that influence trust, this article focuses on the effect of showing integrity. In particular, this article presents a study of how different integrity-based explanations made by an AI agent affect the appropriateness of trust of a human in that agent. To explore this, (1) we provide a formal definition to measure appropriate trust, (2) present a between-subject user study with 160 participants who collaborated with an AI agent in such a task. In the study, the AI agent assisted its human partner in estimating calories on a food plate by expressing its integrity through explanations focusing on either honesty, transparency, or fairness. Our results show that (a) an agent who displays its integrity by being explicit about potential biases in data or algorithms achieved appropriate trust more often compared to being honest about capability or transparent about the decision-making process, and (b) subjective trust builds up and recovers better with honesty-like integrity explanations. Our results contribute to the design of agent-based AI systems that guide humans to appropriately trust them, a formal method to measure appropriate trust, and how to support humans in calibrating their trust in AI.

Relevancy is a prevalent term in value alignment. We either need to keep track of the relevant moral reasons, we need to embed the relevant values, or we need to learn from the relevant behaviour. What relevancy entails in particular cases, however, is often ill-defined. The reasons for this are obvious, it is hard to define relevancy in a way that is both general and concrete enough to give direction towards a specific implementation. In this paper, we describe the inherent difficulty that comes along with defining what is relevant to a particular situation. Simply due to design and the way an AI system functions, we need to state or learn particular goals and circumstances under which that goal is completed. However, because of both the changing nature of the world and the varied wielders and users of such implements, misalignment occurs, especially after a longer amount of time. We propose a way to counteract this by putting contestability front and centre throughout the lifecycle of an AI system, as it can provide insight into what is actually relevant at a particular instance. This allows designers to update the applications in such a manner that they can account for oversight during design.

As human-machine teams become a more common scenario, we need to ensure mutual trust between humans and machines. More important than having trust, we need all teammates to trust each other appropriately. This means that they should not overtrust or undertrust each other, avoiding risks and inefficiencies, respectively. We usually think of natural trust, that is, humans trusting machines, but we should also consider artificial trust, that is, artificial agents trusting humans. Appropriate artificial trust allows the agents to interpret human behavior and predict their behavior in a certain context. In this chapter, we explore how we can define this context in terms of task and team characteristics. We present a taxonomy that shows how trust is context-dependent. In fact, we propose that no trust model presented in the literature fits all contexts and argue that our taxonomy facilitates the choice of the trust model that better fits a certain context. The taxonomy helps to understand which internal characteristics of the teammate (krypta) are important to consider and how they will show in behavior cues (manifesta). This taxonomy can also be used to help human-machine teams’ researchers in the problem definition and process of experimental design as it allows a detailed characterization of the task and team configuration. Furthermore, we propose a formalization of the belief of trust as context-dependent trustworthiness, and show how beliefs of trust can be used to reach appropriate trust. Our work provides a starting point to implement mutual appropriate trust in human-machine teams.

We adopt an emerging and prominent vision of human-centred Artificial Intelligence that requires building trustworthy intelligent systems. Such systems should be capable of dealing with the challenges of an interconnected, globalised world by handling plurality and by abiding by human values. Within this vision, pluralistic value alignment is a core problem for AI– that is, the challenge of creating AI systems that align with a set of diverse individual value systems. So far, most literature on value alignment has considered alignment to a single value system. To address this research gap, we propose a novel method for estimating and aggregating multiple individual value systems. We rely on recent results in the social choice literature and formalise the value system aggregation problem as an optimisation problem. We then cast this problem as an ℓ_p-regression problem. Doing so provides a principled and general theoretical framework to model and solve the aggregation problem. Our aggregation method allows us to consider a range of ethical principles, from utilitarian (maximum utility) to egalitarian (maximum fairness). We illustrate the aggregation of value systems by considering real-world data from two case studies: the Participatory Value Evaluation process and the European Values Study. Our experimental evaluation shows how different consensus value systems can be obtained depending on the ethical principle of choice, leading to practical insights for a decision-maker on how to perform value system aggregation.

With the growing capabilities and pervasiveness of AI systems, societies must collectively choose between reduced human autonomy, endangered democracies and limited human rights, and AI that is aligned to human and social values, nurturing collaboration, resilience, knowledge and ethical behaviour. In this chapter, we introduce the notion of self-reflective AI systems for meaningful human control over AI systems. Focusing on decision support systems, we propose a framework that integrates knowledge from psychology and philosophy with formal reasoning methods and machine learning approaches to create AI systems responsive to human values and social norms. We also propose a possible research approach to design and develop self-reflective capability in AI systems. Finally, we argue that self-reflective AI systems can lead to self-reflective hybrid systems (human + AI), thus increasing meaningful human control and empowering human moral reasoning by providing comprehensible information and insights on possible human moral blind spots.