Dong Wang

Explainability has been a critical aspect in social intelligence applications that analyze human-centered data and can directly impact human decision-making and well-being. This chapter presents two graph-based AI-driven explanation approaches, HC-COVID and DExFC, that address several fundamental challenges in developing explainable social intelligence systems. These challenges include the varying knowledge fact quality contributed by humans with diverse expertise, lack of modality-level annotations, and diverse cross-modal explanations. Through extensive experiments on real-world social intelligence case studies, including COVID-19 news truth discovery and fauxtography detection, both frameworks demonstrate significant performance gains in both prediction accuracy and explanation quality compared to state-of-the-art baselines.

Fairness and bias are critical concerns in modern AI and social intelligence systems. This chapter first introduces the fundamental issues of demographic bias in data-driven social intelligence applications, such as facial analysis and educational assessment. We present two novel frameworks, FairCrowd and DebiasEdu to address these critical concerns. In particular, FairCrowd is a fair crowdsourcing-based data sampling framework that leverages crowd intelligence to infer demographic labels and achieve balanced dataset representation without requiring extensive manual annotations. DebiasEdu is a crowd-AI collaborative framework that combines gradient-based bias identification with crowd-guided bias calibration to achieve fair and accurate student performance prediction. Through comprehensive case studies on human face data sampling and student performance prediction, we demonstrate the effectiveness of these approaches to address fairness and bias issues in social intelligence and show the potential of integrating human intelligence with AI systems to create more equitable and effective social intelligence applications.

Social intelligence systems often face two fundamental limitations, data sparsity and model generality, when the systems need to adapt to new domains or scenarios where training data is limited or unavailable. This is particularly important in critical social intelligence applications such as health truth discovery and disaster damage assessment. This chapter introduces two human-AI hybrid approaches, CrowdAdapt and CollabGeneral, that explore human intelligence and domain expertise to effectively bridge the knowledge gap between data-rich source domains and emergent target domains. Two fundamental challenges exist in developing such hybrid approaches, including: (1) the domain discrepancy where knowledge transfer is degraded by irrelevant or inapplicable information from source domains, and (2) the optimal trade-off between model generality and domain specificity that ensures a social intelligence model can perform well across different domains while maintaining high accuracy for domain-specific characteristics and patterns. Two real-world case studies demonstrate the superiority and great potential of CrowdAdapt and CollabGeneral in addressing these challenges and advancing the development of adaptive social intelligence systems.

Human-AI collaboration is a transformative approach in advancing AI-driven applications, combining artificial intelligence and human expertise to address complex problems. This chapter presents collaborative frameworks that enable the integration of human intelligence from crowdsourcing systems into the AI process, overcoming critical challenges such as large design search spaces and managing imperfect crowd-sourced data. Specifically, we present two case studies: (1) CrowdNAS, a crowd-guided neural architecture search framework for disaster damage assessment, which uses crowd inputs to identify optimal network architectures for accurate damage severity estimation, and (2) CrowdOptim, a crowd-driven hyperparameter optimization framework for AI-based smart urban sensing applications, which leverages crowdsourced feedback to enhance model performance in assessing urban environments. These frameworks and case studies demonstrate the power of collaborative frameworks in guiding AI optimization and neural architecture discovery, showing how crowd-AI systems can achieve high accuracy while reducing computational demands, paving the way for robust human-centered AI applications across dynamic and data-intensive environments.

Humans have traditionally been an integral part of artificial intelligence systems as a means of generating labeled training data [5, 26, 42, 48, 54, 64, 73]. Such a paradigm has been proven to be effective in supervised learning tasks such as image classification [13], speech recognition [20], autonomous driving [65], social media mining [83], and virtual reality [61]. However, it also suffers from two key limitations. First, some applications (e.g., disaster response and damage assessment, online truth discovery) may require a large amount of training data to achieve reasonable performance, which could be impractical due to the labor cost [22, 37]. Second, the AI models are often black-box systems and it is difficult to diagnose in the event of failure or unsatisfactory performance. To address these limitations, a few human-AI hybrid frameworks have been developed in recent years. For example, Holzinger et al. proposed the notion of interactive human machine learning (“iML”), where humans directly interact with AI by identifying useful features that could be incorporated into the AI algorithms [23]. Branson et al. invented a human-in-the-loop visual recognition system to accurately classify the objects in the picture based on the descriptions of the picture from humans [7]. More recently, researchers have been interested in diagnosing the black-box AI algorithms to provide accountability. For example, Nushi et al. developed an accountable human-AI system that leverages workers on Amazon Mechanical Turk (AMT) to identify the limitations of the AI algorithms [49] and provide suggestions to improve them. However, the above solutions largely ignored the innate limitations of the AI algorithms that cannot be simply improved by retraining the model with more data. Human-AI interaction is a trending research area that aims at harnessing the power of human intelligence and AI to optimize the human-AI systems and improve human experience with AI [11, 40, 52, 56, 71]. Such a paradigm has been applied in many domains, including image classification [50, 60], natural language translation [6, 86], medical diagnosis [14, 39], and autonomous driving [74, 84]. More recently, a few human-AI interaction systems have been developed to explore the human intelligence of crowd workers through interactive crowdsourcing tasks [41, 46]. For example, Nguyen et al. designed a human-AI interactive news article fact-checking algorithm that checks the truthfulness of textual news and claims by assigning crowd workers to correct identification errors by AI models [46]. Mandel et al. developed a game-based crowdsourcing interface to incorporate a crowd of non-AI experts to reason the dynamics of AI misbehavior and improve AI performance in online advertisement recommendations [41]. However, these solutions either assume the human workers have sufficient domain knowledge or require them to be well-trained on domain-specific crowdsourcing tasks. Such approaches often suffer from noisy crowdsourcing results since the ordinary crowd workers usually do not have the essential domain knowledge for the domain-specific tasks or are not interested in those training tasks [29, 76]. Future works in the direction of human-AI collaboration systems are expected to address some of these limitations.

In this chapter, we discuss the root of social intelligence from the perspective of human intelligence (HI), artificial intelligence (AI), and their integration. The interdisciplinary nature of social intelligence is also highlighted with a detailed discussion on closely related domains, such as natural language processing and text mining, computer vision and image processing, social computing and human-computer interaction, estimation theory, and statistical learning. Additionally, we present a few emerging social intelligence applications in real-world scenarios to further demonstrate the necessity of this new intelligence paradigm. Examples of these applications include social media misbehavior identification and mitigation, multimodal truth discovery, explainable AI and machine learning, disaster response and damage assessment, AI and crowdsourcing for education, and social sensing in smart city applications.

In this chapter, we summarize the techniques, theories, models, and solutions reviewed in previous chapters. We also discuss a few remaining challenges and exciting directions for future research in the field of social intelligence. We expect interest in social intelligence from different research communities (e.g., AI, machine learning, NLP, computer vision, social computing, human-computer interaction, estimation theory and statistical learning, fairness, and privacy) will keep on increasing and more fundamental and interesting research work will be carried out in the future.

In this chapter, we introduce the new paradigm of Social Intelligence (SI) where the goal is to explore the collective intelligence of both humans and machines by understanding their complementary strengths and interactions in the social space. We highlight the uniqueness of the social intelligence paradigm in the context of related literature. We further discuss the motivation of SI from both the challenge and application perspectives. Examples of some key challenges in SI include data heterogeneity, model generality, explainability, fairness and bias, privacy, and hybrid intelligence integration. Finally, we conclude the chapter by summarizing the contributions of this book and presenting the structure for the rest of the book.

This chapter discusses critical privacy challenges in social intelligence systems when the collective social intelligence data meets privacy constraints. We examine how the collection and integration of sensitive personal information across platforms raise significant privacy concerns which potentially limit the effectiveness of social intelligence solutions. To address these challenges, we present two novel solutions: CoviDKG and FaceCrowd. CoviDKG is a distributed knowledge graph framework that constructs a set of knowledge graphs from individual sources/platforms and exchanges the privacy-aware information across different sources/platforms to effectively detect online false information. FaceCrowd is a web crowdsourcing-based face partition approach that aims to improve the performance of current face recognition models in social intelligence by designing a novel crowdsourced partial face graph generated from privacy-preserved social media face images. Through extensive experiments on real-world datasets and user studies, we demonstrate that both frameworks successfully balance privacy protection with superior performance.

The chapter outlines the mathematical aspects of social intelligence and covers major estimation theory techniques such as Maximum Likelihood Estimation (MLE), Expectation-Maximization (EM), Hidden Markov Models (HMM), Bayesian Estimation, and Subjective Logic. With our understanding of their mathematics and application, we seek to shed light on their utility in parameter inference and uncertainty-based decision-making in social intelligence settings. Furthermore, the chapter transitions into an analysis of deep learning models such as Multilayer Perceptrons (MLP), Convolutional Neural Networks (CNN), Graph Neural Networks (GNN), and Transformers, emphasizing their architectural innovations and application-specific optimizations. This comprehensive synthesis provides a unified framework for understanding and leveraging advanced computational methods in Social Intelligence, setting the stage for future research and applications in this interdisciplinary domain.

Data heterogeneity is a fundamental issue in social intelligence, where the data is often obtained from diverse sources, modalities, and contexts, due to the varying nature of human interactions and behaviors. Addressing the data heterogeneity problem in social intelligence encounters several unique challenges, such as cross-modal information inconsistency, sparse multimodal data annotation, and heterogeneous feature fusion. To overcome these challenges, this chapter reviews state-of-the-art multimodal solutions that address the data heterogeneity challenge in social intelligence applications. In particular, we present two case studies: one on generative learning based multimodal truth discovery and another on contrastive learning based fauxtography detection. These case studies demonstrate the superiority and potential of advanced deep learning techniques in addressing data heterogeneity issues in social intelligence tasks, paving the way for more accurate and reliable analysis of diverse data modalities.

Given the rise of AI and the advent of online collaboration opportunities (e.g., social media, crowdsourcing), emerging research has started to investigate the integration of AI and human intelligence, especially in a collaborative social context. This creates unprecedented challenges and opportunities in the field of Social Intelligence (SI), where the goal is to explore the collective intelligence of both humans and machines by understanding their complementary strengths and interactions in the social space. In this book, a set of novel human-centered AI techniques are presented to address the challenges of social intelligence applications, including multimodal approaches, robust and generalizable frameworks, and socially empowered explainable AI designs. The book then presents several human-AI collaborative learning frameworks that jointly integrate the strengths of crowd wisdom and AI to address the limitations inherent in standalone solutions. The book also emphasizes pressing societal issues in the realm of social intelligence, such as fairness, bias, and privacy. Real-world case studies from different applications in social intelligence are presented to demonstrate the effectiveness of the proposed solutions in achieving substantial performance gains in various aspects, such as prediction accuracy, model generalizability and explainability, algorithmic fairness, and system robustness.

Calligraphy and painting aesthetics not only involves artistic creation and aesthetic experience, but also contains profound philosophical thinking, reflecting the unique oriental art view and world view. This paper adopts the method of philosophical analysis and interdisciplinary integration to re-examine the essence of calligraphy and painting aesthetics, and analyze its core value as a philosophical aesthetic practice, aiming to explore the interdisciplinary integration path of calligraphy and painting aesthetics and philosophy, and enrich and develop the theoretical system of calligraphy and painting aesthetics. This paper proposes that the essence of calligraphy and painting aesthetics is a philosophical aesthetic practice, the core of which is to construct an artistic ontology that transcends form with the dialectical unity of “skill” and “Tao”, thus opening up a unique oriental path for the cross-cultural dialogue and universal construction of the art theory system. The interdisciplinary research paradigm that combines calligraphy and painting aesthetics with philosophy can deepen the understanding of calligraphy and painting aesthetics and provide direction for the innovation of art theory under the background of globalization.

Building on the literature on optimal distinctiveness, this study explores the differential impacts of normative versus strategic corporate social responsibility (CSR) practices on social evaluations. We argue that normative CSR, which entails conformity in the scope of CSR, mitigates negative social evaluations, whereas strategic CSR, which emphasizes differentiation in CSR practices, enhances positive social evaluations. Additionally, we examine the moderating factors influencing these relationships. Using the case of corporate social activities of Chinese listed firms during the period from 2006 to 2019, our results show that the mitigating effect of normative CSR on negative evaluations is more pronounced in firms with higher reputational risk. Conversely, the relationship between strategic CSR and positive social evaluations is stronger among firms with lower levels of peers' strategic CSR. Furthermore, we demonstrate that negative and positive social evaluations serve as mechanisms through which normative and strategic CSR, respectively, influence market outcomes. These findings contribute to the optimal distinctiveness theory and advance research on CSR strategy.

This study seeks to shed light on the effect of star CEOs on the environmental, social, and governance (ESG) performance of Chinese firms. Relying on the theoretical perspective of role identity and role constraints, we analyze data from 1222 Chinese firms listed on the Shanghai and Shenzhen Stock Exchanges from 2006 to 2019. The results analyzed using the ordinary least squares estimate method reveal a positive effect of star CEOs' extreme confidence and legitimacy pressure mechanisms on ESG performance. We also find that this positive relationship is negatively moderated by higher organizational slack but strengthened by a lower level of regional marketization, as the degree of firms' dependence on external stakeholders' resources differs significantly. Additionally, this study finds that ESG activities are less likely to be associated with firm performance if star CEOs pursue them to satisfy their need for star status. These findings supplement the early tests of the influence mechanism of star CEOs on organization-level results and expand the context discussion of role identity and role constraints theory in the Chinese institutional environment.

Both Peirce and Dewey explained the concept of inquiry. From an epistemological point of view, Peirce regarded inquiry as the process of belief changing from an uncertain state to a certain one. Dewey developed the concept of inquiry and imbued it with an existential meaning. Inquiry is a functional product generated in the process of dealing with the real world. Specifically, habits are stable patterns of temporarily fixed behaviors; impulses are the primary drive for inquiry; and intelligence is an effective way to deal with new problems and make experience more controllable.

Purpose To investigate the reliability and accuracy of two pseudo-continuous arterial spin labeling sequences, using two-dimensional gradient-echo echo planar imaging and 3D gradient and spin echo as the readout, respectively. Materials and Methods Each sequence was performed twice 4 weeks apart on six normal control subjects, six elderly subjects with mild cognitive impairment, and one participant with Alzheimer's disease. Eight of these subjects also underwent H 215O positron emission tomography scans on the same day or proximal to their second MRI scan. The longitudinal repeatability of EPI and GRASE pCASL were evaluated with the intraclass correlation coefficient and within-subject coefficient of variation. Results The ICCs of global perfusion measurements were 0.697 and 0.413 for GRASE and EPI based pCASL respectively. GRASE pCASL also demonstrated a higher longitudinal repeatability for regional perfusion measurements across 24 regions-of-interests compared with EPI pCASL. When compared with PET, EPI pCASL showed a higher degree of spatial correlation with PET than GRASE pCASL, although the difference was not statistically significant. Conclusion The 3D GRASE pCASL offers better repeatability than 2D EPI pCASL, thereby may provide a reliable imaging marker for the evaluation of disease progression and treatment effects in MCI and early AD subjects. © 2013 Wiley Periodicals, Inc.

The notion of “agent bookkeeping” was proposed when the “Accounting Law of the People’s Republic of China” was updated in 1993. Since their business is specialized in serving small and micro-enterprises, this has created the industry characteristic of generally small in the size of company and low in the salary of employees in Chinese agent bookkeeping companies. Such characteristic results in a series of problems including negative work attitude of employees in the development process, which seriously limit the development of Chinese agent bookkeeping companies. However, issues occurred in the development of agent bookkeeping, such as negative employee attitudes, which severely hampered the growth of Chinese agent bookkeeping companies. Therefore, a model of three dimensions of work attitude has been set up in this paper to demonstrate the working attitude of the staff of the agent bookkeeping companies in China’s Guangdong-Hong Kong-Macao Greater Bay Area, including emotional dimension, behavioral dimension, and cognitive dimension which consists of eight factors: work environment, career satisfaction, interpersonal relationship, role engagement, work vitality, responsibility attitude, emotional identity, and retention attitude. The result indicates that female employees outnumber male ones in agent bookkeeping companies, most employees have a low sense of belonging to the company, limited career development affects employees’ enthusiasm for work, work attitudes are influenced by mediocre work performance among employees with a graduate degree or higher.

ObjectiveThe psychological condition of healthcare workers since the COVID-19 pandemic has attracted the attention of many studies. However, few have reported on psychosocial problems of primary healthcare workers in the COVID-19 pandemic. This study aimed to examine the mediating roles of social support and resilience in COVID-19-related work stress and symptoms of anxiety and depression.MethodsA total of 840 primary healthcare workers in 17 community health centers in Guangzhou, China, were recruited from May to July 2021. Data on demographic characteristics, COVID-19-related work stress, social support, resilience, anxiety and depression were collected. A structural equation model was used for mediation analysis.ResultsMore than half of participants reported mild or more severe symptoms of anxiety and depression. Social support and resilience mediate the association between COVID-19-related work stress and symptoms of anxiety and depression, respectively. Furthermore, the association between work stress and symptoms of anxiety and depression was also mediated by an accumulation of social support and resilience. The indirect effect of COVID-19-related work stress on anxiety and depression through resilience was much greater than other indirect effects.ConclusionAnxiety and depression were prevalent among primary healthcare workers. This study highlights the psychological impact of the COVID-19-related psychosocial work environment on primary healthcare workers. There is an urgent need to improve working conditions for primary healthcare workers in the COVID-19 and to implement intervention strategies aimed at increasing individual resilience alongside the establishment of external supportive work environments.

The adaptive object model method is an effective way to develop dynamic and configurable adaptive software. It has the characteristics of metamodel, description drive, and runtime reflection. First, the core idea of the adaptive object model is explained; then, the five modes of establishing the metamodel in the adaptive object model architecture, the model engine, and supporting tools are analyzed; and the basketball tracking algorithm of the adaptive object model is discussed. Secondly, a two-dimensional joint information strategy is proposed to improve the tracking effect. When the basketball is in a very complex environment, there will always be some color information in the background that is the same as the target, which affects the effect of basketball tracking. Therefore, this paper proposes a Camshift tracking method based on the significance of histograms, through real time. The basketball movement is compared with the background histogram to continuously adjust the basketball movement tracking model. These two methods can better establish the tracking model of the basketball adaptive object, reduce the interference of background information, and achieve the effect of stable tracking of the target. The simulation experiment results show that the method proposed in this paper can effectively improve the accuracy of the basketball goal model and achieve stable tracking of the goal.

Objective: The objective of this study was to evaluate the psychometric properties of the Chinese version of the decision regret scale. Methods: The data of 704 patients who completed the DRSc were used for the analyses. We evaluated the construct, convergent/discriminant, and known-group validity; internal consistency and test–retest reliability; and the item invariance of the DRSc. A receiver operating characteristic curve was employed to confirm the optimal cutoff point of the scale. Results: A confirmatory factor analysis indicated that a one-factor model fits the data. The internal consistency and test–retest reliability [intraclass correlation coefficient = 0.71] of the DRSc were acceptable. The DRSc demonstrated unidimensionality and invariance for use across the sexes. It was confirmed that an optimal cutoff point of 25 could discriminate between patients with high and low decisional regret during clinical practice. Conclusion: The DRSc is a parsimonious instrument that can be used to measure the uncertainty inherent in medical decisions. It can be employed to provide knowledge, offer support, and elicit patient preferences in an attempt to promote shared decision-making.