Potential for GenAI in the Public Domain: A Review of Transportation, Healthcare, Agriculture, and Law
Divya Dwivedi, Rahul De’Synopsis
Generative Artificial Intelligence (GenAI) tools are becoming quite popular for a variety of operations. One such tool, ChatGPT, is rapidly permeating into people’s daily lives and is considered to have significant potential to reshape our society. While private organizations are spending huge amounts of money on ChatGPT, its usage in the public domain is still driven by its open access and simple functionality.
This paper draws on the key concepts of ‘Effective Use’ theory: Transparent Interaction, Representational Fidelity, Informed Action, and Learning and Adaptation to examine ChatGPT’s current state of diffusion in the public sector. ‘Effective Use’ theory builds on prior theories of technology acceptance to focus on technology adoption and use. Transparent Interaction examines how easy it is to interact with GenAI tools for the domain of concern, and whether the interfaces are user-friendly and easy to access. Representational Fidelity is concerned with whether the outputs of the system are correct? Are the answers reliable and the insights relatable to the reality of the domain? (Or, is there a problem of “hallucination”?). Informed Action indicates whether the GenAI outputs are being used for informed decision-making and actions, and if users can make better decisions with GenAI. Lastly, Learning and Adaptation points to the availability of training programs to help users learn how to use these systems; and if there is a mechanism for them to raise issues and concerns to better fit the systems for their use?
The authors attempt to understand these key concepts in the context of GenAI’s use in four public domains: transportation, healthcare, agriculture, and law. They find transparent interaction is better in transportation, agriculture, and law than healthcare; representational fidelity presents a complex picture whereas informed action is positive across domains; and learning and adaptation is an ongoing need. They conclude with various suggestions related to research and policy toward boosting GenAI’s adoption, i.e., public domain models should be trained on languages and text obtained from different cultures and regions; and tools should be developed with transparency, and released as “public goods” – accessible by all. They suggest that governments invest resources and develop new regulatory frameworks considering the specific context and use cases for leveraging the enormous potential of GenAI tools in the public domain.
