Artificial Intelligence (AI), Machine Learning (ML), and Large Language Models (LLM) hold vast opportunities to help us improve health information systems and their efficacy in supporting health services and management. After decades of cycles with hype and disappointment, there is emergent evidence that real application of these technologies are at a turning point. Most notably, the proliferation of consumer-oriented LLMs such as ChatGPT has drastically increased the availability and reach of such technologies. The purpose of this master project is to explore the potential uses and consequences of AI-based tools in the context of healthcare.
The HISP Centre, coordinating development and implementation of the DHIS2 platform, has already began exploring concrete applications of these technologies. The work is in an early phase and research is needed on the applicability, appropriateness, and ethics of using this in the health sector. Some of the work that as already been undertaken is:
- DHIS2 Documentation Chatbot: using LLM to provide relevant instructions on how to configure DHIS2, based primarily on existing documentation material
- Report builder: Application for taking a visualization from DHIS2, analyzing it with GPT, and writing up a report for its findings. Include a custom prompt to steer the AI into giving you a report in the format you want. Typical use cases involves finding key insights and actionable points from raw data.
- API Integration: Chatbot that integrates and allows the AI to fetch relevant and updated data from the API. If you ask it to get all event programs, it would run a function that fetches this data, and so on. This could be extended to include other functions.
- Research chatbot: a chatbot that synthesizes conclusions from a large amount of scientific papers. Initially it was pretrained on ∽150 documents focusing on AI. This can be extended to involve Climate Health, Logistics, EMIS and all the sectors DHIS2 is used in.
Master projects can engage in both design, development, and experimentation with AI-based tools as well study the use, effects, and consequences of these. Under this theme, there are several research topics available for Master theses.
Potential topics include;
- How can interactive chat bots, using LLMs such as ChatGPT, provide support for various users groups? What are the dilemmas in tuning such a chat bot to provide assistance to groups as diverse as health managers and system administrators? And related;
- What is affecting the use of chat bots or other AI-assisted support mechanisms? Here it is natural to think of factors such as trust and competency to critically apply such technologies.
- How can such technologies play a role in distributed development in a platform ecosystem? For example, can LLMs lower the bar for innovation, allowing less experienced developers/practitioners to solve problems with for instance new apps?
- What are ethical concerns of applying AI and LLM in the context of HISP, and how can and should they be met?
- Data synthetization. To make robust digital health technologies, they need rigorous quality assurance and testing with realistic data. Likewise, training should be done on realistic data that pose no privacy and confidentiality concerns. For the DHIS2 software, both these use cases could be solved with ways of producing large amounts of synthetic data that is realistic enough to expose system vulnerabilities. How can LLMs or associated technologies be applied to generate such data?
The research will be conducted in collaboration with the HISP Centre and in coordination with ongoing activities among developers, HISP groups, and countries implementing DHIS2.
Who is this for?
These topics are available for a group of students interested in the potential benefits and challenges of applying AI, machine learning, or Large Language Models to improve health information systems and health management. Theses can explore more technical issues requiring programming, or more exploratory or evaluative issues examining user experiences and organizational aspects. The project can accommodate up to 5 students.
Data collection can involve, if wanted and appropriate, field work in a context where DHIS2 is used.