Table 1. Equitable health chatbot implementation roadmap.
DEVELOPMENT AND IMPLEMENTATION OF EQUITABLE CONVERSATIONAL AI IN HEALTHCARE |
---|
Co-design and co-development phase |
1. Conception and planning |
○ Identify existing health disparities and explain how conversational AI can address them ○ Define and set behavioural and health outcomes that the conversational AI is aiming to influence ○ Define conversational AI roles and tasks aimed at reducing health inequalities ○ Define intended users of conversational AI and their characteristics (take into account aspects of marginalisation and intersectionality) ○ Define the underlying behaviour change frameworks and potential mechanisms of action on which conversational AI operates |
2. Diversity and collaboration |
○ Ensure diversity in the design and implementation team throughout the project ○ Involve various stakeholders (end users and healthcare professionals) from diverse communities in the conceptualisation and development of the conversational AI tool ○ Involve community champions and peer support groups in public and patient engagement activities ○ Adopt a user-centred and culturally sensitive approach to reduce bias |
3. Preliminary research |
○ Review evidence base for conversational AI tools concerning their acceptability, uptake and reach ○ Review existing conversational AI tools or product designs that could be adopted ○ Review ethical guidelines on the use of conversational AI in healthcare for diverse communities ○ Review and identify high-quality AI training datasets from diverse communities ○ Review and identify accessible patient-facing IT systems for wider reach in underserved communities ○ Identify barriers to engagement with conversational AI specifically in minoritised communities |
4. Co-production |
○ Co-develop and optimise the content of the conversational AI tool (i.e. knowledge base and language) for comprehension, readability and interactivity ○ Ensure language understanding, response accuracy and impartiality through user testing activities ○ Enable multiple language translation for culturally and linguistically diverse communities ○ Ensure the ease of use for those with lower literacy or disability ○ Augment conversational AI with human-like responses and characteristics in a culturally sensitive manner ○ Ensure conversational AI transparency and its ability to present itself to people from diverse communities |
5. Safety measures |
○ Address ethical issues and prevent harm from inaccurate or unreliable responses from conversational AI ○ Develop safeguarding measures to protect vulnerable users ○ Enable a human contact pathway for further support and assistance ○ Protect user privacy when collecting sensitive personal information, considering data storage and access ○ Report and manage adverse events and unintended consequences, including complaint management |
6. Preliminary testing |
○ Proof-of-concept testing with diverse communities ○ Adress aspects related to user hesitancy or disengagement with conversational AI ○ Gather and incorporate user feedback ○ Measure the impact of conversational AI on behavioural and health outcomes within a diverse sample |
Healthcare system implementation phase |
7. Healthcare integration |
○ Understand the healthcare system and clinical pathways for the supplementary role of conversational AI ○ Define an adequate level of governance and gain regulatory requirements/approvals ○ Clarify outstanding legal and licencing matters (i.e. intellectual property and third-party involvement) ○ Define accountability for system failure, glitches and malfunctions ○ Define regional variations in healthcare delivery and user characteristics for AI integration and tailoring ○ Define scalability and generalisability of conversational AI across the healthcare system ○ Develop staff training materials and resources on how to administer and supervise conversational AI |
8. Evaluation and auditing |
○ Conduct a feasibility study exploring usage, satisfaction and confidence in conversational AI ○ Assess uptake and engagement in marginalised communities across healthcare ○ Set up regular auditing with multiple outcomes and various methodological designs ○ Assess the sustainability of conversational AI within healthcare and community settings (e.g. cost-effectiveness) |
9. Maintenance |
○ Implement regular system updates and technical improvements ○ Understand the impact of conversational AI on the healthcare environment, service delivery, and clinical team workloads ○ Raise awareness and promote conversational AI in diverse communities |
10. Termination |
○ Design termination procedure for conversational AI within the healthcare environment ○ Assess the impact of AI termination in marginalised communities ○ Ensure the continuation of patient care through the identification of alternative services ○ Monitor and evaluate the impact of AI termination on healthcare service delivery |