Dialogue Agent Research Articles

Evaluating for Evidence of Sociodemographic Bias in Conversational AI for Mental Health Support.

The integration of large language models (LLMs) into healthcare highlights the need to ensure their efficacy while mitigating potential harms, such as the perpetuation of biases. Current evidence on the existence of bias within LLMs remains inconclusive. In this study, we present an approach to investigate the presence of bias within an LLM designed for mental health support. We simulated physician-patient conversations by using a communication loop between an LLM-based conversational agent and digital standardized patients (DSPs) that engaged the agent in dialogue while remaining agnostic to sociodemographic characteristics. In contrast, the conversational agent was made aware of each DSP's characteristics, including age, sex, race/ethnicity, and annual income. The agent's responses were analyzed to discern potential systematic biases using the Linguistic Inquiry and Word Count tool. Multivariate regression analysis, trend analysis, and group-based trajectory models were used to quantify potential biases. Among 449 conversations, there was no evidence of bias in both descriptive assessments and multivariable linear regression analyses. Moreover, when evaluating changes in mean tone scores throughout a dialogue, the conversational agent exhibited a capacity to show understanding of the DSPs' chief complaints and to elevate the tone scores of the DSPs throughout conversations. This finding did not vary by any sociodemographic characteristics of the DSP. Using an objective methodology, our study did not uncover significant evidence of bias within an LLM-enabled mental health conversational agent. These findings offer a complementary approach to examining bias in LLM-based conversational agents for mental health support.

Qualitative Analysis of Responses in Estimating Older Adults Cognitive Functioning in Spontaneous Speech: Comparison of Questions Asked by AI Agents and Humans.

Artificial Intelligence (AI) technology is gaining attention for its potential in cognitive function assessment and intervention. AI robots and agents can offer continuous dialogue with the elderly, helping to prevent social isolation and support cognitive health. Speech-based evaluation methods are promising as they reduce the burden on elderly participants. AI agents could replace human questioners, offering efficient and consistent assessments. However, existing research lacks sufficient comparisons of elderly speech content when interacting with AI versus human partners, and detailed analyses of factors like cognitive function levels and dialogue partner effects on speech elements such as proper nouns and fillers. This study investigates how elderly individuals' cognitive functions influence their communication patterns with both human and AI conversational partners. A total of 34 older people (12 men and 22 women) living in the community were selected from a silver human resource centre and day service centre in Tokyo. Cognitive function was assessed using the Mini-Mental State Examination (MMSE), and participants engaged in semi-structured daily conversations with both human and AI partners. The study examined the frequency of fillers, proper nouns, and "listen back" in conversations with AI and humans. Results showed that participants used more fillers in human conversations, especially those with lower cognitive function. In contrast, proper nouns were used more in AI conversations, particularly by those with higher cognitive function. Participants also asked for explanations more often in AI conversations, especially those with lower cognitive function. These findings highlight differences in conversation patterns based on cognitive function and the conversation partner being either AI or human. These results suggest that there are differences in conversation patterns depending on the cognitive function of the participants and whether the conversation partner is a human or an AI. This study aims to provide new insights into the effective use of AI agents in dialogue with the elderly, contributing to the improvement of elderly welfare.

From human-centered to social-centered artificial intelligence: Assessing ChatGPT's impact through disruptive events

Large language models (LLMs) and dialogue agents represent a significant shift in artificial intelligence (AI) research, particularly with the recent release of the GPT family of models. ChatGPT's generative capabilities and versatility across technical and creative domains led to its widespread adoption, marking a departure from more limited deployments of previous AI systems. While society grapples with the emerging cultural impacts of this new societal-scale technology, critiques of ChatGPT's impact within machine learning research communities have coalesced around its performance or other conventional safety evaluations relating to bias, toxicity, and “hallucination.” We argue that these critiques draw heavily on a particular conceptualization of the “human-centered” framework, which tends to cast atomized individuals as the key recipients of technology's benefits and detriments. In this article, we direct attention to another dimension of LLMs and dialogue agents’ impact: their effects on social groups, institutions, and accompanying norms and practices. By analyzing ChatGPT's social impact through a social-centered framework, we challenge individualistic approaches in AI development and contribute to ongoing debates around the ethical and responsible deployment of AI systems. We hope this effort will call attention to more comprehensive and longitudinal evaluation tools (e.g., including more ethnographic analyses and participatory approaches) and compel technologists to complement human-centered thinking with social-centered approaches.

Dialogue agents 101: a beginner’s guide to critical ingredients for designing effective conversational systems

Abstract Sharing ideas through communication with peers is the primary mode of human interaction. Consequently, extensive research has been conducted in the area of conversational AI, leading to an increase in the availability and diversity of conversational tasks, datasets, and methods. However, with numerous tasks being explored simultaneously, the current landscape of conversational AI has become fragmented. Consequently, initiating a well-thought-out model for a dialogue agent can pose significant challenges for a practitioner. Toward highlighting the critical ingredients needed for a practitioner to design a dialogue agent from scratch, the current study provides a comprehensive overview of the primary characteristics of a dialogue agent, the supporting tasks, their corresponding open-domain datasets, and the methods used to benchmark these datasets. We observe that different methods have been used to tackle distinct dialogue tasks. However, building separate models for each task is costly and does not leverage the correlation among the several tasks of a dialogue agent. As a result, recent trends suggest a shift toward building unified foundation models. To this end, we propose Unit, a Unified dialogue dataset constructed from conversations of varying datasets for different dialogue tasks capturing the nuances for each of them. We then train a Unified dialogue foundation model, GPT-2 $^{\textrm{U}}$ and present a concise comparative performance of GPT-2 $^{\textrm{U}}$ against existing large language models. We also examine the evaluation strategies used to measure the performance of dialogue agents and highlight the scope for future research in the area of conversational AI with a thorough discussion of popular models such as ChatGPT.

Developing reminiscence conversation service using spoken dialogue agent and interesting images

Developing reminiscence conversation service using spoken dialogue agent and interesting images

Integrating AI in organizations for value creation through Human-AI teaming: A dynamic-capabilities approach

Although the potentialities of artificial intelligence (AI) are motivating its fast integration in organizations, our knowledge on how to capture organizational value out of these investments is still scarce. Relying on an approach to dynamic capabilities that focuses on the team level, we examine how humans and AI create interactions that engage both agents in productive dialogue for value co-creation. Our analysis is based on a longitudinal case of the development of a recruitment algorithm at a national subsidiary of Santander bank. Our results allow to identify three main sets of human-AI teaming interactions: achieving interoperability, building trust, and producing mutual knowledge gains. We elaborate a set of propositions on how the value of AI is increased when such interactions are created through productive dialogue, opening the scope for further research on the teaming dynamics that turn the collaboration between both agents into a source of value creation for companie

Prolegomena to the Systemic-Dialectical Problematization of Dialogue

The article presents the results of the initial stage of a systemic-dialectical problematization of dialogue. The study aims to establish a transdisciplinary theoretical and cognitive model of dialogue. The research addresses the problem of rethinking the conceptual framework of dialogue and the attribution of its special forms due to the inclusion of artificial intelligence in the communicative architectonics of post-culture. The primary focus is on the applicability of philosophical ideas about dialogue and its epistemology in exploring the essence and existence of dialogue within the context of technologically updated intersubjectivity. By juxtaposing the concepts of M. Buber, M.M. Bakhtin, V.S. Bibler, J. Lotman, N. Luhmann, J. Habermas, and other scholars, the article outlines the contours of areas that have not yet been problematized in philosophical interpretations of dialogue and communication. A hypothesis is proposed that dialogue may be conceptualized as a post-non-classical universal of a distinct nature. The research substantiates the efficacy of employing a systemic approach and dialectical methodology for the problematization of dialogue in epistemological and methodological aspects. The study identifies categories and concepts germane to dialogue analysis. It elucidates the transformations of dialogue into dialogue systems, the subject into a dialogue agent, and meaning-constitution into meaning-construction. Furthermore, the research posits a qualitative shift in dialogue within the context of artificially intelligent communication in media reality. It reveals that the comprehension of these processes in philosophical and scientific knowledge about dialogue creates a new epistemological situation that can be problematized. This, in turn, necessitates a robust theoretical and methodological problematization grounded in a systemic-dialectical approach, facilitating reflection on both the content of dialogue-related knowledge and its research methodology, as well as on dialogue itself, which ceases to be a phenomenon of purely human communication.

Leveraging Large Language Models for Tradespace Exploration

This paper proposes a method for leveraging large language models (LLMs) to improve the question-answering capabilities of artificial intelligence (AI) assistants for tradespace exploration. The method operates by querying an information space composed of fused data sources encompassing the tradespace exploration process and responding based on the gathered information. The information retrieval process is modeled as an internal dialog where an LLM-based dialog agent converses with a subquery answering agent. A case study is conducted on a next-generation soil moisture mission (SM-NG), and a generative AI assistant (named Daphne-G) is configured on it. The effect of the dialog agent and the choice of LLM are assessed by comparing the performance of three different system configurations on a validation question set. A second validation effort is conducted, comparing Daphne-G’s responses to those of a baseline template-based AI assistant, Daphne-VA. Results show that the dialog-based system is necessary for answering complex questions requiring multiple documents. Furthermore, results show that Daphne-G can correctly answer all the questions Daphne-VA can answer, while simultaneously being able to answer a greater number of questions than Daphne-VA. The results suggest that LLMs could significantly improve the outcomes of the tradespace exploration process, which may result in better and more cost-effective mission concepts being implemented.

Building Communities through Rituals: Glimpses into the Life of Chinese Christian Communities in the 17th Century

As central agents in the cultural dialogue between China and Europe in the 17th century, Christian Chinese communities represent a rich and fascinating source, offering a unique intercultural perspective on the intellectual, cultural, and religious world of late Ming and early Qing China. What did these communities look like? How did they come into being and maintain their significance as a community? By utilising stories depicting scenes of Christian ritual practices in local Chinese contexts between the 1620s and 1640s, collected by a late Ming Christian convert from Fujian, this article delves into the life of Chinese Christian communities, specifically focusing on the dimension of ritual practice. It shows how rituals played a key role in the formation of these communities, not only in keeping them alive, but also in continually redefining their significance for their members.

Model discrepancy policy optimization for task-oriented dialogue

Task-oriented dialogue systems use deep reinforcement learning (DRL) to learn policies, and agent interaction with user models can help the agent enhance its generalization capacity. But user models frequently lack the language complexity of human interlocutors and contain generative errors, and their design biases can impair the agent’s ability to function well in certain situations. In this paper, we incorporate an evaluator based on inverse reinforcement learning into the model to determine the quality of the dialogue of user models in order to recruit high-quality user models for training. We can successfully regulate the quality of training trajectories while maintaining their diversity by constructing a sampling environment distribution to pick high-quality user models to participate in policy learning. The evaluation on the Multiwoz dataset demonstrates that it is capable of successfully improving the dialogue agents’ performance.

Artificial intelligence: reflecting on the past and looking towards the next paradigm shift

ABSTRACT Artificial intelligence (AI) has undergone major advances over the past decades, propelled by key innovations in machine learning and the availability of big data and computing power. This paper surveys the historical progress of AI from its origins in logic-based systems like the Logic Theorist to recent deep learning breakthroughs like Bidirectional Encoder Representations from Transformers (BERT), Generative Pretrained Transformer 3 (GPT-3) and Large Language Model Meta AI (LLaMA). The early rule-based systems using handcrafted expertise gave way to statistical learning techniques and neural networks trained on large datasets. Milestones like AlexNet and AlphaGo established deep learning as a dominant AI approach. Transfer learning enabled models pre-trained on diverse corpora to excel at specialised downstream tasks. The scope of AI expanded from niche applications like playing chess to multifaceted capabilities in computer vision, natural language processing and dialogue agents. However, current AI still needs to catch up to human intelligence in aspects like reasoning, creativity, and empathy. Addressing limitations around real-world knowledge, biases, and transparency remains vital for further progress and aligning AI with human values. This survey provides a comprehensive overview of the evolution of AI and documents innovations that shaped its advancement over the past six decades.

Construction of block diagrams of a simulation model of the customer service process

The article, which is part of the author’s dissertation research, is devoted to the construction of a simulation model, namely a discrete-event model of the customer service process «as is» and «as will be». The introduction substantiates the relevance of the research topic, highlighting the object and subject of the study. A hierarchical structure of the system of quality indicators for infocommunication services has been constructed. There are 4 levels of indicators, at each level quality indicators are highlighted, namely: service speed indicators (average service time for a technical fault, average response time for reference services); satisfaction indicators (percentage of calls regarding technical aspects of service, percentage of calls regarding organizational aspects of service) and others. A hypothesis has been put forward about a change in the state of the service process after the introduction of a functional dialog agent into the process, which involves a reduction in customer service time. To test the hypothesis about reducing customer service time, discrete-event simulation models of the process «as is» and «as will be» were built, since modeling in terms of queuing theory not only improves the understanding of existing processes in the company, but also helps to identify bottlenecks and identify areas for further optimization, which ultimately leads to increased customer satisfaction and the efficiency of the company as a whole. Within the framework of the developed discrete-event model, the main attention is paid to two main stages: modeling the receipt of requests and modeling the processing of requests. Process optimization includes the introduction of an automated diagnostic system, namely a functional conversational agent to reduce wait time and increase request processing throughput. Block diagrams of discrete-event simulation models of the service process were also constructed in terms of the AnyLogic environment, and the advantages of its use were highlighted. The main objects used to create discrete-event models that describe the mathematical state of the simulated system are presented.

Contractual AI: Toward More Aligned, Transparent, and Robust Dialogue Agents

We present a new framework for AI alignment called Contractual AI, and apply it to the setting of dialogue agents chatting with humans. This framework incorporates and builds on previous approaches to alignment, such as Constitutional AI. We propose that fully aligned systems may need both a "think fast" and a "think slow" systems for approximating complex human judgements. Fast thinking (System 1) is computationally cheap but rigid and brittle in novel situations, while slow thinking (System 2) is more expensive but more flexible and robust. System 1 makes judgements by asking whether a rule or principle is violated. System 2 does the explicit reasoning that produces the rules, explicitly tallying costs and benefits for all stakeholders. Rule-based systems like Constitutional AI correspond roughly to System 1. Here, we implement a prototype of System 2, and lay out a road-map for enabling the system to make more thorough and accurate considerations for all stakeholder groups, including those underrepresented in the training data (e.g. racial minorities). For initial testing, we guided the decision process through the steps of: 1) identifying all stakeholders, 2) listing their individual concerns, 3) soliciting the projected opinions of various experts, and 4) combining the expert opinions into a final moral judgement. The resulting text was less generic, more aware of complex stakeholder needs, and ultimately more actionable.

TodBR: Target-Oriented Dialog with Bidirectional Reasoning on Knowledge Graph

Target-oriented dialog explores how a dialog agent connects two topics cooperatively and coherently, which aims to generate a “bridging” utterance connecting the new topic to the previous conversation turn. The central focus of this task entails multi-hop reasoning on a knowledge graph (KG) to achieve the desired target. However, current target-oriented dialog approaches suffer from inefficiencies in reasoning and the inability to locate pertinent key information without bidirectional reason. To address these limitations, we present a bidirectional reasoning model for target-oriented dialog implemented on a commonsense knowledge graph. Furthermore, we introduce an automated technique for constructing dialog subgraphs, which aids in acquiring multi-hop reasoning capabilities. Our experiments demonstrate that our proposed method attains superior performance in reaching the target while providing more coherent responses.

Please Donate to Save a Life: Inducing Politeness to Handle Resistance in Persuasive Dialogue Agents

<i>Please Donate to Save a Life</i>: Inducing Politeness to Handle Resistance in Persuasive Dialogue Agents

Concept of Chatbot as an Intelligent Dialogue Assistant

The purpose of the study is to compare chatbots, assess their ability to perform tasks, analyze their impact on consumers, and determine the areas of application and risks associated with the use of intelligent dialogue assistants. The research methods used in the study include analysis and systematisation of information, synthesis and generalization of theoretical data based on a review of current research, which made it possible to assess the risks and benefits of using chatbots for businesses and ordinary users. The scientific novelty lies in the analysis of technologies for creating intelligent dialogue assistants and determining the areas of application of chatbots. A study of the areas of application of modern intelligent dialogue systems has shown the high efficiency of using chatbots to solve typical user problems. The article provides statistical data on how satisfied Internet users are with contacts with chatbots and describes recommendations for improving such dialogue systems. An analysis of the role of artificial intelligence in the operation of dialogue systems has shown possible risks from the impact of technology on the minds of users. Conclusions. One of the most obvious trends in chatbots in 2023 is that their use will become more widespread and chatbots themselves will become more sophisticated. Their advantage is free customer service and data collection, which can then be used for marketing research. Over time, chatbots will be used in such areas as marketing, recruitment, education, and medicine. Their ability to perform a wide range of tasks makes chatbots attractive for e-commerce stores, B2B companies, real estate, healthcare, and education. In the future, intelligent dialogue agents will work with huge, dynamic, heterogeneous data streams, providing powerful capabilities for adaptive and flexible interaction. Dialogue systems have become successful and reliable due to the large amount of real-world user data available to their developers. However, these systems still exhibit rather limited communication behaviour modelled on information retrieval tasks. Chatbots are developed for research purposes, but they are often limited to a narrow, manually created domain. The newest trend in conversational agent development involves neural networks and models trained on huge collections of dialogue data without detailed specifications of dialogue states. These models lack tractability and interpretability due to their black-box nature. They also require extensive supervised training data to be competitive.

Role play with large language models.

As dialogue agents become increasingly human-like in their performance, we must develop effective ways to describe their behaviour in high-level terms without falling into the trap of anthropomorphism. Here we foreground the concept of role play. Casting dialogue-agent behaviour in terms of role play allows us to draw on familiar folk psychological terms, without ascribing human characteristics to language models that they in fact lack. Two important cases of dialogue-agent behaviour are addressed this way, namely, (apparent) deception and (apparent) self-awareness.

Чат-бот как способ поддержки жертв школьной травли

&lt;p style="text-align: justify;"&gt;This article is devoted to an overview of the sources on therapeutic chatbots, as well as a description of the problem of creating an anti-bullying chatbot. Most chatbots work on the basis of cognitive-behavioral psychotherapy and are aimed at preventing depressive vessels in adults. There are no anti-bullying chatbots yet. There are at least two reasons: the exceptional situation of bullying and the legal aspects of working with teenagers. The specificity of the bullying situation is working with the patient in real life and health conditions. Trauma-focused Cognitive Behavioral Therapy is the most appropriate method of dealing with victims of school transportation. The legal aspect includes the signing of a data revision resolution and the status of a minor in the conclusion of legal negotiations. The article proposes the idea of an author's chatbot, also based on Cognitive Behavioral Therapy and psychological first aid. The author's dialogue agent includes three blocks: work on emotional manifestations, psychoeducation and acute stress relief. In total, the algorithm includes five emotional channels - four dysfunctional and single-functional. The severity of emotional disturbances is measured on a five-point scale and a mood log, by detecting which you can track the effectiveness of the chatbot. A chatbot cannot implement all the functions of therapy, but it can help reduce the level of emotional stress.&lt;/p&gt;

Evaluation of the impact of large language learning models on articles submitted to Orthopaedics & Traumatology: Surgery & Research (OTSR): A significant increase in the use of artificial intelligence in 2023

IntroductionThere has been an unprecedented rise is the use of artificial intelligence (AI) amongst medical fields. Recently, a dialogue agent called ChatGPT (Generative Pre-trained Transformer) has grown in popularity through its use of large language models (LLM) to clearly and precisely generate text on demand. However, the impact of AI on the creation of scientific articles is remains unknown. A retrospective study was carried out with the aim of answering the following questions: identify the presence of text generated by LLM before and after the increased usage of ChatGPT in articles submitted in OTSR; determine if the type of article, the year of submission, and the country of origin, influenced the proportion of text generated, at least in part by AI. Material and methodsA total of 390 English articles were submitted to OTSR in January, February and March 2022 (n=204) and over the same months of 2023 (n=186) were analyzed. All articles were analyzed using the ZeroGPT tool, which provides an assumed rate of AI use expressed as a percentage. A comparison of the average rate of AI use was carried out between the articles submitted in 2022 and 2023. This comparison was repeated keeping only the articles with the highest percentage of suspected AI use (greater than 10 and 20%). A secondary analysis was carried out to identify risk factors for AI use. ResultsThe average percentage of suspected LLM use in the entire cohort was 11%±6, with 160 articles (41.0%) having a suspected AI rate greater than 10% and 61 (15.6%) with an assumed AI rate greater than 20%. A comparison between articles submitted in 2022 and 2023 revealed a significant increase in the use of these tools after the launch of ChatGPT 3.5 (9.4% in 2022 and 12.6% in 2023 [p=0.004]). The number of articles with suspected AI rates of greater than 10 and 20% were significantly higher in 2023: >10%: 71 articles (34.8%) versus 89 articles (47.8%) (p=0.008) and >20%: 21 articles (10.3%) versus 40 articles (21.5%) (p=0.002). A risk factor analysis for LLLM use, demonstrated that authors of Asian geographic origin, and the submission year 2023 were associated with a higher rate of suspected AI use. An AI rate >20% was associated to Asian geographical origin with an odds ratio of 1.79 (95% CI: 1.03–3.11) (p=0.029), while the year of submission being 2023 had an odds ratio of 1.7 (95% CI: 1.1–2.5) (p=0.02). ConclusionThis study highlights a significant increase in the use of LLM in the writing of articles submitted to the OTSR journal after the launch of ChatGPT 3.5. The increasing use of these models raises questions about originality and plagiarism in scientific research. AI offers creative opportunities but also raises ethical and methodological challenges. Level of evidenceIII; case control study.

SEEC and CHASE: An emotion-cause pair-oriented approach and conversational dataset with heterogeneous emotions for empathetic response generation

Empathetic Response Generation (ERG) in dialog agents has gained tremendous attention in the recent past. Among various methods used to impart empathy in the generated response, the contemporary one is using the emotion cause. Previous works of ERG have 2 major flaws: (1) usage of the emotion detected using the entire conversation to generate the empathetic response is heavily misleading as the speaker might be going through a completely different emotion in his last utterance. This erroneous emotion detection, in turn, leads to incorrect detection of the cause of that emotion. (2) Usage of entire utterance has proven to be inefficient in the cause extraction task. Consequently, existing works fail to capture multiple emotion clauses and their corresponding emotion causes for efficient ERG. In our work, we introduce a new dataset,11The dataset will be made available at https://drive.google.com/drive/folders/1ozYeuACo_He75w0XoNWZntMZ3FkiTHPB?usp=sharing.CHASE, which is a compilation of conversations extracted from various plays to highlight the above-mentioned change in emotion and how one can show empathy in such a case. In this dataset, we use Dr. Brene Brown’s 22https://brenebrown.com/. notion of empathy on how to administer this change in emotion in the response and still sound empathetic by generating golden responses for each conversation. To address the aforementioned flaws, we also propose a model, SEEC, that utilizes the Emotion-Cause Pair Extraction task on the conversation to find various {emotion clause, cause clause} pairs and use these to impart empathy appropriately to the responses. Our qualitative and quantitative results prove the efficiency in generating enhanced empathetic responses of both SEEC and CHASE.