Cai4sg Advances Social Good: Examining Roles, Challenges and Emerging Trends

Researchers are increasingly exploring how conversational AI can be harnessed to address pressing societal challenges, a field now known as Conversational AI for Social Good (CAI4SG). Yi-Chieh Lee, Junti Zhang, and Tianqi Song from the National University of Singapore’s School of Computing, together with Yugin Tan and colleagues, present a comprehensive overview of this rapidly evolving area by categorising systems according to their levels of autonomy and emotional intelligence. Their work goes beyond the technical development of conversational agents, focusing instead on how these systems operate within social good contexts, such as empathetic mental health support and accessibility assistance. By adopting a role-based framework, the study highlights critical challenges including algorithmic bias and data privacy, offering guidance for the more equitable and ethical design and deployment of CAI4SG applications.

AI Autonomy and Engagement in CAI4SG represent critical

Researchers are increasingly examining how conversational AI can be applied to pressing societal challenges, a field now known as Conversational AI for Social Good (CAI4SG). Yi-Chieh Lee, Junti Zhang, and Tianqi Song from the National University of Singapore’s School of Computing, together with Yugin Tan and colleagues, present a comprehensive overview of this rapidly evolving area. Their work moves beyond the technical construction of conversational agents and instead focuses on how such systems function within social good contexts, including mental health support and accessibility services. By introducing a role-based framework that categorises systems according to their levels of autonomy and emotional engagement, the study highlights critical challenges such as algorithmic bias, data privacy, and ethical responsibility, providing a structured foundation for more equitable and responsible CAI4SG development.

The research situates CAI4SG within the broader evolution of conversational systems, which have progressed from simple rule-based interfaces to sophisticated platforms offering scalability, continuous availability, and cost efficiency. These systems increasingly rival or complement human operators while enabling greater personalisation and, in some cases, enhanced privacy protection. As the global chatbot market expands, conversational AI has become a transformative technology across domains ranging from customer service to healthcare. Within this context, AI for Social Good (AI4SG) emphasises the use of AI to address social, environmental, and economic challenges, aligning technological innovation with goals such as equity, inclusion, fairness, and the United Nations Sustainable Development Goals.

At the intersection of conversational AI and social good, CAI4SG offers unique opportunities to address societal challenges at scale. Natural language interfaces reduce technological barriers, allowing deployment across diverse populations regardless of technical literacy. CAI4SG systems, including chatbots and virtual assistants, are increasingly used to promote public well-being, expand access to reliable information, and support underserved communities in areas such as mental health, public health, education, and humanitarian response. However, deploying CAI in these sensitive contexts raises distinct concerns about accountability, trust, and the nature of human–AI interaction.

To address these complexities, the role-based framework analyses CAI4SG applications along two key dimensions: AI autonomy and emotional engagement. Autonomy reflects the degree of independent decision-making assigned to the system, ranging from low-autonomy tools that provide information and guidance to high-autonomy systems that actively collaborate with users in problem-solving. Emotional engagement captures how systems recognise, interpret, and respond to users’ emotional states, from purely informational exchanges to emotionally intelligent interactions capable of building sustained relationships. These dimensions are particularly significant in social good applications, where higher autonomy and emotional engagement can increase impact but also introduce risks such as emotional dependency and accountability challenges. By framing CAI4SG through these roles, the study offers a nuanced understanding of how conversational AI can be designed and governed to maximise societal benefit while mitigating potential harms.

Role-based framework for analysing CAI4SG challenges

Researchers investigated Conversational AI for Social Good (CAI4SG) through a role-based framework that categorises systems according to their levels of autonomy and emotional engagement. This approach enabled the identification of role-specific challenges associated with different applications, such as empathetic support systems and accessibility assistants, and how these challenges vary with interaction intensity. The study systematically examined how algorithmic bias, data privacy risks, and broader socio-technical harms manifest differently depending on a conversational agent’s function. To address limitations in emotion recognition, the authors advocate for multimodal approaches that integrate textual, vocal, and contextual signals, noting that text-only sentiment analysis often fails to capture critical paralinguistic cues such as tone and speech dynamics. The framework further highlights the need for cultural adaptation mechanisms and rigorous evaluation across diverse demographic groups to ensure equitable performance in sensitive social good contexts.

The research also places strong emphasis on safeguarding user data, particularly in emotionally engaged applications such as health support, through robust encryption practices and transparent user interfaces that clearly communicate data usage. Privacy-preserving technologies are presented as essential for maintaining accountability and user trust without compromising system functionality. While personalisation is recognised as important for emotionally intensive CAI, the study warns against excessive tailoring that may lead to user overwhelm or sycophantic behaviour. To mitigate this risk, the authors propose principle-grounded reward shaping and calibrated disagreement objectives that prioritise honesty over agreement and discourage belief-congruent but misleading responses. For high-autonomy systems, the study underscores the importance of algorithmic audits, bias-aware user reporting mechanisms, and actionable transparency to address biases in training data. Finally, it stresses the value of calibrated uncertainty expression by CAI systems as a means of fostering trust, accountability, and responsible deployment.

Autonomy, Engagement and Ethical Challenges Identified

Scientists are increasingly focused on Conversational AI for Social Good (CAI4SG) and its potential to address global challenges, as highlighted by a new role-based framework examining recent advancements in the field. This research categorises conversational AI systems according to their levels of autonomy and emotional engagement, emphasising the importance of clearly defining the role a system plays in social good contexts. These roles range from empathetic mental health support to accessibility assistance, each introducing distinct technical, ethical, and societal considerations. By mapping the current CAI4SG landscape, the study identifies key application areas and reveals how challenges such as algorithmic bias, data privacy risks, and broader socio-technical harms vary depending on the system’s role and depth of engagement.

The findings show that strong ethical frameworks and collective governance mechanisms are essential for realising the positive potential of CAI4SG while minimising harm, ultimately supporting a sustainable, just, and healthy future. The researchers highlight that systems designed for empathetic interaction, particularly in mental health contexts, demand heightened ethical scrutiny due to risks such as emotional dependency and misuse of sensitive personal data. The study also reports partial funding from the National University of Singapore CSSH (24-1774-A0002), the NUS HSS Seed Fund CR (2024 24-1191-A0001), and a Google Research Gift, reflecting broad institutional support for this area of research.

Further analysis points to progress in related areas such as persuasive technologies, supported by evidence from a four-month field study involving persuasive social robots, which contributes to the development of more effective CAI4SG systems. Ongoing initiatives like OpenAGI demonstrate how multi-agent collaborative frameworks are expanding the scope of complex problem-solving using conversational AI. The study also references real-world deployments, including Woebot, which has been evaluated through randomised controlled trials delivering cognitive behavioural therapy to young adults, and CommunityBots, which enable public input elicitation and collaborative decision-making. At the same time, the authors stress the importance of privacy-preserving AI, particularly in healthcare, and caution against unintended harms, such as emotional reliance on social chatbots like Replika. Overall, the work provides a comprehensive view of both the opportunities and risks associated with CAI4SG, reinforcing the need for responsible, human-centred design and deployment.

CAI4SG Roles, Risks and Transformative Potential

Scientists are increasingly exploring Conversational AI for Social Good (CAI4SG), recognising its potential to address pressing global challenges. This paper introduces a role-based framework that categorises conversational AI systems according to their levels of autonomy and emotional engagement, highlighting the diverse functions these systems can serve. These roles range from empathetic mental health support and accessibility assistance to public service delivery and disaster risk reduction, each presenting distinct technical, ethical, and societal challenges. The research underscores the importance of explicitly considering these roles during system design and deployment, particularly in light of risks such as algorithmic bias, data privacy violations, and broader socio-technical harms. Overall, the findings demonstrate that CAI4SG holds transformative potential across multiple domains by enabling scalable, accessible, and responsive support mechanisms.

By leveraging varying degrees of autonomy and emotional engagement, CAI4SG systems can significantly expand access to services, reduce resource constraints, and address urgent societal needs that were previously difficult to meet at scale. However, the authors caution that realising this potential requires a strong commitment to ethical, human-centred design principles. Emphasis is placed on fairness, accountability, inclusivity, and transparency to ensure that these systems do not reinforce existing inequalities or marginalise vulnerable populations. The study highlights the necessity of policies and governance structures that prioritise human agency and well-being over purely commercial objectives. Future research is encouraged to focus on systematically embedding these ethical principles throughout the CAI4SG development lifecycle, ensuring responsible innovation and sustainable societal impact. Ultimately, the paper frames the responsible development of conversational AI not merely as a technical challenge, but as a moral imperative essential to fostering a more equitable, healthy, and sustainable future.