Abstract
Conversational artificial intelligence (ConvAI) systems have attracted much academic and commercial attention recently, making significant progress on both fronts. However, little existing work discusses how these systems can be developed and deployed for social good in real-world applications, with comprehensive case studies and analyses of pros and cons. In this paper, we briefly review the progress the community has made towards better ConvAI systems and reflect on how existing technologies can help advance social good initiatives from various angles that are unique for ConvAI, or not yet become common knowledge in the community. We further discuss about the challenges ahead for ConvAI systems to better help us achieve these goals and highlight the risks involved in their development and deployment in the real world.
Introduction
While Conversational AI (ConvAI) technologies like virtual assistants and customer service bots have become ubiquitous, their potential for systematic social impact remains largely untapped. Most "AI for Social Good" (AI4SG) research has historically focused on computer vision or predictive analytics. However, the Natural Language User Interface (LUI) offered by ConvAI presents a unique opportunity to bridge the accessibility gap. By operating over basic telephone lines or SMS, ConvAI can deliver essential services—such as healthcare, education, and governance—to underserved populations who may lack high-speed internet or digital literacy.
The Growth and Maturity of ConvAI
The field of ConvAI has seen an explosion in research interest over the last few decades. Data from the ACL Anthology reveals a sharp rise in publications related to "dialogue," "chat," and "conversational" systems, particularly in the last five years. This surge indicates that the technology has reached a level of maturity where real-world deployment for social good is increasingly feasible.
Framework for Social Impact
To effectively apply ConvAI to the United Nations’ Sustainable Development Goals (SDGs), the authors categorize systems based on their "initiative" and "format":
- System-initiative: Ideal for structured tasks like pandemic surveys or government feedback collection.
- User-initiative: Suited for knowledge-based assistants, such as answering public health FAQs.
- Mixed-initiative: Necessary for complex scenarios like one-on-one tutoring or mental health support, where a fluid exchange of information is required.
Case Studies: ConvAI in Action
ConvAI can be a powerful tool in sectors like health and governance. For instance, during a pandemic, automated systems can conduct surveys over the phone to monitor disease spread or provide emotional support to isolated individuals.
In healthcare (SDG #3), ConvAI can act as a "first responder" by detecting distress signals in user interactions. In education (SDG #4), it provides scalable, low-pressure tutoring. In governance (SDG #10), it enables anonymous feedback, reducing the social stigma often associated with reporting local issues.
Ethical Challenges and Risks
Despite its potential, ConvAI deployment carries significant risks:
- The Gender Trap: Defaulting to female personas for assistants can reinforce harmful gender stereotypes.
- Digital Deception: As systems become more human-like, the risk of "deepfake" conversations and automated scams increases, necessitating "transparency-by-design."
- Technical Robustness: Current systems often struggle with non-standard accents or low-bandwidth audio, which may inadvertently exclude the very populations they are meant to help.
Conclusion
Bridging the gap between state-of-the-art research and social good requires a concerted effort to address technical robustness, ethical biases, and data-sharing barriers. By focusing on equitable access and transparent design, ConvAI can become a cornerstone of global social good initiatives.
Summary updated