Hybrid Intelligence (HI) is an emerging paradigm in which artificial intelligence (AI) augments human intelligence. The current literature lacks systematic models that guide the design and evaluation of HI systems. Further, discussions around HI primarily focus on technology, neglecting the holistic human-AI ensemble. In this paper, we take the initial steps toward the development of a quality model for characterizing and evaluating HI systems from a human-AI teams perspective. We first conducted a study investigating the adequacy of properties commonly associated with effective human teams to describe HI. The study features the insights of 50 HI researchers, and shows that various human team properties, including boundedness, interdependence, competency, purposefulness, initiative, normativity, and effectiveness, are important for HI systems. Based on these results, we developed a quality model for HI teams composed of seven high-level quality attributes, further refined into 16 specific ones. To evaluate the relevance and understanding of the proposed attributes, we conducted a second empirical investigation by staging competitions in which participants used the quality model to develop and analyze HI usage scenarios. Our analysis of 48 collected scenarios, which we openly release, confirms the proposed attributes' relevance and highlights insights that emerge when designers consider the quality model in HI system design.

Addressing climate change requires coordinated policy efforts of nations worldwide. These efforts are informed by scientific reports, which rely in part on Integrated Assessment Models (IAMs), prominent tools used to assess the economic impacts of climate policies. However, traditional IAMs optimize policies based on a single objective, limiting their ability to capture the trade-offs among economic growth, temperature goals, and climate justice. As a result, policy recommendations have been criticized for perpetuating inequalities, fueling disagreements during policy negotiations. We introduce JUSTICE, the first framework integrating IAM with Multi-Objective Multi-Agent Reinforcement Learning (MOMARL). By incorporating multiple objectives, JUSTICE generates policy recommendations that shed light on equity while balancing climate and economic goals. Further, using multiple agents can provide a realistic representation of the interactions among the diverse policy actors. We identify equitable Pareto-optimal policies using our framework, which facilitates deliberative decision-making by presenting policymakers with the inherent trade-offs in climate and economic policy.

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.

Michiel van der Meer, Piek Vossen, Catholijn M. Jonker, Pradeep K. Murukannaiah. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Multi-objective reinforcement learning (MORL) is used to solve problems involving multiple objectives. An MORL agent must make decisions based on the diverse signals provided by distinct reward functions. Training an MORL agent yields a set of solutions (policies), each presenting distinct trade-offs among the objectives (expected returns). MORL enhances explainability by enabling fine-grained comparisons of policies in the solution set based on their trade-offs as opposed to having a single policy. However, the solution set is typically large and multi-dimensional, where each policy (e.g., a neural network) is represented by its objective values. We propose an approach for clustering the solution set generated by MORL. By considering both policy behavior and objective values, our clustering method can reveal the relationship between policy behaviors and regions in the objective space. This approach can enable decision makers (DMs) to identify overarching trends and insights in the solution set rather than examining each policy individually. We tested our method in four multi-objective environments and found it outperformed traditional k-medoids clustering. Additionally, we include a case study that demonstrates its real-world application.

Active Learning (AL) addresses the high costs of collecting human annotations by strategically annotating the most informative samples.However, for subjective NLP tasks, incorporating a wide range of perspectives in the annotation process is crucial to capture the variability in human judgments.We introduce Annotator-Centric Active Learning (ACAL), which incorporates an annotator selection strategy following data sampling.Our objective is two-fold:(1) to efficiently approximate the full diversity of human judgments, and (2) to assess model performance using annotator-centric metrics, which value minority and majority perspectives equally.We experiment with multiple annotator selection strategies across seven subjective NLP tasks, employing both traditional and novel, human-centered evaluation metrics.Our findings indicate that ACAL improves data efficiency and excels in annotator-centric performance evaluations.However, its success depends on the availability of a sufficiently large and diverse pool of annotators to sample from.

Abstract The expressive power and effectiveness of large language models (LLMs) is going to increasingly push intelligent agents towards sub-symbolic models for natural language processing (NLP) tasks in human–agent interaction. However, LLMs are characterised by a performance vs. transparency trade-off that hinders their applicability to such sensitive scenarios. This is the main reason behind many approaches focusing on local post-hoc explanations, recently proposed by the XAI community in the NLP realm. However, to the best of our knowledge, a thorough comparison among available explainability techniques is currently missing, as well as approaches for constructing global post-hoc explanations leveraging the local information. This is why we propose a novel framework for comparing state-of-the-art local post-hoc explanation mechanisms and for extracting logic programs surrogating LLMs. Our experiments—over a wide variety of text classification tasks—show how most local post-hoc explainers are loosely correlated, highlighting substantial discrepancies in their results. By relying on the proposed novel framework, we also show how it is possible to extract faithful and efficient global explanations for the original LLM over multiple tasks, enabling explainable and resource-friendly AI techniques.

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.

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 ℓp-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.

Enrico Liscio, Oscar Araque, Lorenzo Gatti, Ionut Constantinescu, Catholijn Jonker, Kyriaki Kalimeri, Pradeep Kumar Murukannaiah. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Disagreements are common in online discussions.Disagreement may foster collaboration and improve the quality of a discussion under some conditions.Although there exist methods for recognizing disagreement, a deeper understanding of factors that influence disagreement is lacking in the literature.We investigate a hypothesis that differences in personal values are indicative of disagreement in online discussions.We show how state-of-the-art models can be used for estimating values in online discussions and how the estimated values can be aggregated into value profiles.We evaluate the estimated value profiles based on human-annotated agreement labels.We find that the dissimilarity of value profiles correlates with disagreement in specific cases.We also find that including value information in agreement prediction improves performance.

We present a review that unifies decision-support methods for exploring the solutions produced by multi-objective optimization (MOO) algorithms. As MOO is applied to solve diverse problems, approaches for analyzing the trade-offs offered by these algorithms are scattered across fields. We provide an overview of the current advances on this topic, including methods for visualization, mining the solution set, and uncertainty exploration as well as emerging research directions, including interactivity, explainability, and support on ethical aspects. We synthesize these methods drawing from different fields of research to enable building a unified approach, independent of the application. Our goals are to reduce the entry barrier for researchers and practitioners on using MOO algorithms and to provide novel research directions.

The pursuit of values drives human behavior and promotes cooperation. Existing research is focused on general values (e.g., Schwartz) that transcend contexts. However, context-specific values are necessary to (1) understand human decisions, and (2) engineer intelligent agents that can elicit and align with human values. We propose Axies, a hybrid (human and AI) methodology to identify context-specific values. Axies simplifies the abstract task of value identification as a guided value annotation process involving human annotators. Axies exploits the growing availability of value-laden text corpora and Natural Language Processing to assist the annotators in systematically identifying context-specific values. We evaluate Axies in a user study involving 80 human subjects. In our study, six annotators generate value lists for two timely and important contexts: Covid-19 measures and sustainable Energy. We employ two policy experts and 72 crowd workers to evaluate Axies value lists and compare them to a list of general (Schwartz) values. We find that Axies yields values that are (1) more context-specific than general values, (2) more suitable for value annotation than general values, and (3) independent of the people applying the methodology. Supplementary Information: The online version contains supplementary material available at 10.1007/s10458-022-09550-0.

The bargaining power of prosumers in a market can vary significantly. Participants can range from industrial participants to powerful and less powerful citizens. Existing trade mechanisms in such markets, e.g., in rural India’s energy trade market, show occurrences of discrimination, exclusion, and unfairness. We study how discrimination affects market access, efficiency, and demand satisfaction for the discriminating and discriminated groups via an agent-based simulation, incorporating the available real data. We introduce a mechanism for such markets that is designed for the values of inclusion and equal opportunities. The crux of our mechanism is that goods are divided into smaller units, as determined by the market participants’ surplus and demands, and traded anonymously via agents representing the prosumers. We evaluate six hypotheses in a case study about energy trade in rural India, where members of a caste known as Dalits are discriminated by Others. We show that anonymization contributes to the value of inclusion, and the combination of anonymization and inclusion contributes to equal opportunities with respect to market access for both Dalits and Others.

Background: As smart and automated applications pervade our lives, an increasing number of software developers are required to incorporate machine learning (ML) techniques into application development. However, acquiring the ML skill set can be nontrivial for software developers owing to both the breadth and depth of the ML domain. Aims: We seek to understand the challenges developers face in the process of ML application development and offer insights to simplify the process. Despite its importance, there has been little research on this topic. A few existing studies on development challenges with ML are outdated, small scale, or they do no involve a representative set of developers. Method: We conduct an empirical study of ML-related developer posts on Stack Overflow. We perform in-depth quantitative and qualitative analyses focusing on a series of research questions related to the challenges of developing ML applications and the directions to address them. Results: Our findings include: (1) ML questions suffer from a much higher percentage of unanswered questions on Stack Overflow than other domains; (2) there is a lack of ML experts in the Stack Overflow QA community; (3) the data preprocessing and model deployment phases are where most of the challenges lay; and (4) addressing most of these challenges require more ML implementation knowledge than ML conceptual knowledge. Conclusions: Our findings suggest that most challenges are under the data preparation and model deployment phases, i.e., early and late stages. Also, the implementation aspect of ML shows much higher difficulty level among developers than the conceptual aspect.

Norms describe the social architecture of a society and govern the interactions of its member agents. It may be appropriate for an agent to deviate from a norm; the deviation being indicative of a specialized norm applying under a specific context. Existing approaches for norm emergence assume simplified interactions wherein deviations are negatively sanctioned. We investigate via simulation the benefits of enriched interactions where deviating agents share selected elements of their contexts. We find that as a result (1) the norms are learned better with fewer sanctions, indicating improved social cohesion; and (2) the agents are better able to satisfy their individual goals. These results are robust under societies of varying sizes and characteristics reflecting pragmatic, considerate, and selfish agents.

Automated app review analysis is an important avenue for extracting a variety of requirements-related information. Typically, a first step toward performing such analysis is preparing a training dataset, where developers (experts) identify a set of reviews and, manually, annotate them according to a given task. Having sufficiently large training data is important for both achieving a high prediction accuracy and avoiding overfitting. Given millions of reviews, preparing a training set is laborious. We propose to incorporate active learning, a machine learning paradigm, in order to reduce the human effort involved in app review analysis. Our app review classification framework exploits three active learning strategies based on uncertainty sampling. We apply these strategies to an existing dataset of 4,400 app reviews for classifying app reviews as features, bugs, rating, and user experience. We find that active learning, compared to a training dataset chosen randomly, yields a significantly higher prediction accuracy under multiple scenarios.

In an increasingly digital urban setting, connected & concerned Citizens typically voice their opinions on various civic topics via social media. Efficient and scalable analysis of these citizen voices on social media to derive actionable insights is essential to the development of smart cities. The very nature of the data: heterogeneity and dynamism, the scarcity of gold standard annotated corpora, and the need for multi-dimensional analysis across space, time and semantics, makes urban social media analytics challenging. This workshop is dedicated to the theme of social media analytics for smart cities, with the aim of focusing the interest of CIKM research community on the challenges in mining social media data for urban informatics. The workshop hopes to foster collaboration between researchers working in information retrieval, social media analytics, linguistics; social scientists, and civic authorities, to develop scalable and practical systems for capturing and acting upon real world issues of cities as voiced by their citizens in social media. The aim of this workshop is to encourage researchers to develop techniques for urban analytics of social media data, with specific focus on applying these techniques to practical urban informatics applications for smart cities.

Online discussions about software applications generate a large amount of requirements-related information. This information can potentially be usefully applied in requirements engineering; however currently, there are few systematic approaches for extracting such information. To address this gap, we propose Canary, an approach for extracting and querying requirements-related information in online discussions. The highlight of our approach is a high-level query language that combines aspects of both requirements and discussion in online forums. We give the semantics of the query language in terms of relational databases and SQL. We demonstrate the usefulness of the language using examples on real data extracted from online discussions. Our approach relies on human annotations of online discussions. We highlight the subtleties involved in interpreting the content in online discussions and the assumptions and choices we made to effectively address them. We demonstrate the feasibility of generating high-quality annotations by obtaining them from lay Amazon Mechanical Turk users.

Interactions among stakeholders and engineers is key to Requirements engineering (RE). Increasingly, such interactions take place online, producing large quantities of qualitative (natural language) and quantitative (e.g., votes) data. Although a rich source of requirements-related information, extracting such information from online forums can be nontrivial.We propose Canary, a tool-assisted approach, to facilitate systematic extraction of requirements-related information from online forums via high-level queries. Canary (1) adds structure to natural language content on online forums using an annotation schema combining requirements and argumentation ontologies, (2) stores the structured data in a relational database, and (3) compiles high-level queries in Canary syntax to SQL queries that can be run on the relational database.We demonstrate key steps in Canary workflow, including (1) extracting raw data from online forums, (2) applying annotations to the raw data, and (3) compiling and running interesting Canary queries that leverage the social aspect of the data.

Existing work on identifying security requirements relies on training binary classification models using domain-specific data sets to achieve a high accuracy. Considering that domain-specific data sets are often not readily available, we propose a domain-independent model for classifying security requirements based on two key ideas. First, we train our model on the description of weaknesses from the Common Weakness Enumeration (CWE) data set. Although CWE does not describe requirements, it describes security weaknesses that are manifestations of unrealized security requirements. Second, we exploit a one-class classification model that relies only on positive samples (description of weaknesses in CWE), eliminating the need for negative samples, collecting which can be nontrivial.We evaluated our model on three industrial requirements documents from different domains. We found that a One-Class Support Vector Machine trained with domain-independent CWE data set outperforms a model from prior literature by identifying security requirements with an average precision, recall and F-score of 67.35%, 70.48% and 67.68%, respectively. Further, considering data sets from prior literature (consisting of both positive and negative examples), we found that one-class classifiers trained with only positive examples outperformed binary classifiers trained with both positive and negative examples in two out of three evaluation data sets, demonstrating the potential value of one-class classification for security requirements identification.

Crowd RE is an emerging avenue for engaging the general public or the so called crowd in variety of requirements engineering tasks. Crowd RE scales RE by involving, potentially, millions of users. Although humans are at the center of Crowd RE, automated techniques are necessary (1) to derive useful insights from large amounts of raw data the crowd can produce; and (2) to drive the Crowd RE process, itself, by facilitating novel workflows combining crowd and machine intelligence.To facilitate automated techniques for Crowd RE, first, we showcase a crowd-acquired dataset, consisting of requirements and their ratings on multiple dimensions for the smart homes application domain. Our dataset is unique in that it contains not only requirements, but also the characteristics of the crowd workers who produced those requirements including their demographics, personality traits, and creative potential. Understanding the crowd characteristics is essential to developing effective Crowd RE processes. Second, we outline key challenges involved in automating Crowd RE and describe, how our dataset can serve as a foundation for developing such automated techniques.

Social network services (SNSs) enable users to conveniently share personal information. Often, the information shared concerns other people, especially other members of the SNS. In such situations, two or more people can have conflicting privacy preferences; thus, an appropriate sharing policy may not be apparent. We identify such situations as multiuser privacy scenarios . Current approaches propose finding a sharing policy through preference aggregation. However, studies suggest that users feel more confident in their decisions regarding sharing when they know the reasons behind each other’s preferences. The goals of this paper are (1) understanding how people decide the appropriate sharing policy in multiuser scenarios where arguments are employed, and (2) developing a computational model to predict an appropriate sharing policy for a given scenario. We report on a study that involved a survey of 988 Amazon Mechanical Turk (MTurk) users about a variety of multiuser scenarios and the optimal sharing policy for each scenario. Our evaluation of the participants’ responses reveals that contextual factors, user preferences, and arguments influence the optimal sharing policy in a multiuser scenario. We develop and evaluate an inference model that predicts the optimal sharing policy given the three types of features. We analyze the predictions of our inference model to uncover potential scenario types that lead to incorrect predictions, and to enhance our understanding of when multiuser scenarios are more or less prone to dispute.

User location is crucial in understanding the dynamics of user activities, especially in relating their online and offline aspects. However, users' social media activities, such as tweets sent, do not always reveal their location. We consider the problem of estimating geo-tags for tweets and develop a comprehensive approach that incorporates textual content, the user's personalized behavior, and the user's social relationships. Our approach, Percimo, considers the two major kinds of communal attachment, which have distinct computational ramifications. We evaluate Percimo via three geo-social graphs based on the mutual-follow relationships of Twitter users, their geographical distance (computed from their geotagged tweets), and their preferences for location categories (collected from Foursquare). We find that Percimo yields a smaller prediction error than the two state-of-the-art approaches we compare with.

As a creative discipline, Requirements Engineering (RE), lends importance to understanding the associated human factors. Crowd RE, the approach of acquiring requirements from members of the public-the so-called crowd-emphasizes human factors further. We investigate how human personality and creative potential influence a requirement acquisition task. These factors are of specific importance to Crowd RE because (1) crowd workers are generally not trained in RE, and (2) a key motivation in engaging them is to benefit from their creativity. We propose a sequential Crowd RE process, where workers in one stage review requirements from the previous stage and produce additional requirements. To reduce potential information overload in this process, we propose strategies for selecting requirements from one stage to expose to workers in later stages. We conducted a study on Amazon Mechanical Turk tasking 300 workers with creating requirements via the above sequential process (in the domain of smart home applications for concreteness) and tasking an additional 300 workers to rate the creativity (novelty and usefulness) of those requirements. Our findings offer insights on how to carry out Crowd RE effectively. First, we find that a crowd worker's (1) creative potential, and personality traits of openness and conscientiousness have significant positive influence on the novelty of the worker's ideas, and (2) personality traits of agreeableness and conscientiousness have significant positive influence, but extraversion has significant negative influence on the usefulness of the worker's ideas. Second, we find that exposing a worker to ideas from previous workers cognitively stimulates the worker to produce creative ideas. Third, we identify effective strategies based on personality traits and creative potential for selecting a few requirements from a pool of previous requirements to stimulate a worker.

A stakeholder's beliefs influence his or her goals. However, a stakeholder's beliefs may not be consistent with the goals of all stakeholders of a system being constructed. Such belief-goal inconsistencies could manifest themselves as conflicting goals of the system to be. We propose Arg-ACH, a novel approach for capturing inconsistencies between stakeholders' goals and beliefs, and resolving goal conflicts. Arg-ACH employs a hybrid of (1) the analysis of competing hypotheses (ACH), a structured analytic technique, for systematically eliciting stakeholders' goals and beliefs, and (2) rational argumentation for determining belief-goal inconsistencies to resolve conflicts. Arg-ACH treats conflicting goals as hypotheses that compete with each other and the winning hypothesis as a goal of the system to be. Arg-ACH systematically captures the trail of a requirements engineer's thought process in resolving conflicts. We evaluated Arg-ACH via a study in which 20 subjects applied Arg-ACH or ACH to resolve goal conflicts in a sociotechnical system concerning national security. We found that Arg-ACH is superior to ACH with respect to completeness and coverage of belief search; length of belief chaining; ease of use; explicitness of the assumptions made; and repeatability of conclusions across subjects. Not surprisingly, Arg-ACH required more time than ACH: although this is justified by improvements in quality, the gap could be reduced through better tooling.

We introduce a high-level abstraction of location called place . A place derives its meaning from a user's physical space, activities, or social context. In this manner, place can facilitate improved user experience compared to the traditional representation of location, which is spatial coordinates. We propose the Platys framework as a way to address the special challenges of place-aware application development. The core of Platys is a middleware that (1) learns a model of places specific to each user via active learning , a machine learning paradigm that seeks to reduce the user-effort required for training the middleware, and (2) exposes the learned user-specific model of places to applications at run time, insulating application developers from dealing with both low-level sensors and user idiosyncrasies in perceiving places. We evaluated Platys via two studies. First, we collected place labels and Android phone sensor readings from 10 users. We applied Platys' active learning approach to learn each user's places and found that Platys (1) requires fewer place labels to learn a user's places with a desired accuracy than do two traditional supervised approaches, and (2) learns places with higher accuracy than two unsupervised approaches. Second, we conducted a developer study to evaluate Platys' efficiency in assisting developers and its effectiveness in enabling usable applications. In this study, 46 developers employed either Platys or the Android location API to develop a place-aware application. Our results indicate that application developers employing Platys, when compared to those employing the Android API, (1) develop a place-aware application faster and perceive reduced difficulty and (2) produce applications that are easier to understand (for developers) and potentially more usable and privacy preserving (for application users).

Link prediction on social media is an important problem for recommendation systems. Understanding the interplay of users' sentiments and social relationships can be potentially valuable. Specifically, we study how to exploit sentiment homophily for link prediction. We evaluate our approach on a dataset gathered fro Twitter that consists of tweets sent in one month during U.S. 2012 political campaign along with the "follows" relationship between users. Our first contribution is defining a set of sentiment-based features that help predict the likelihood of two users becoming "friends" (i.e., mutually mentioning or following each other) based on their sentiments toward topics of mutual interest. Our evaluation in a supervised learning framework demonstrates the benefits of sentiment-based features in link prediction. We find that Adamic-Adar and Euclidean distance measures are the best predictors. Our second contribution is proposing a factor graph model that incorporates a sentiment-based variant of cognitive balance theory. Our evaluation shows that, when tie strength is not too weak, our model is more effective in link prediction than traditional machine learning techniques.