At IntraHealth, where our mission is to support health workers and improve the conditions in which they work, AI-assisted training could be a game changer.
Artificial Intelligence, and especially Generative AI, is performing as well or better than humans in reading comprehension and language understanding. It could reduce training expenses and free up nurse time for providing patient care – ultimately improving health outcomes.
Let’s explore what is known today and how we incorporated AI into our software development workflows.
ChatGPT in Software Development
Early studies show that large language models like ChatGPT can support college educated professionals and developers to reduce the time people spend on tasks (like drafting documents) and improve the quality of the work for those who have less-mature writing skills.
GitHub found similar results for software developers doing “boilerplate” tasks like setting up a web server in JavaScript. Using GitHub’s CoPilot tool improved productivity by over 50%!
We set out to see if GitHub’s Copilot and Open AI’s ChatGPT 3.5 could assist our software developers to write documentation, comment code, write unit tests, and find and fix bugs in iHRIS, IntraHealth’s open source software that helps countries around the world track and manage their health workforce data to improve access to health services.
We asked ChatGPT 3.5 if it knew iHRIS, and it gave the response below.
“As of my last knowledge update in January 2022, IHRIS (Integrated Human Resource Information System) is a free, open-source software specifically designed for managing human resources in the health sector. It is often used by organizations and governments to keep track of health workforce data, including information about health workers, their skills, training, and deployment.
However, please note that software information can change over time, and there may have been updates or changes to IHRIS since my last update. I recommend checking the official website or other reliable sources for the latest information on IHRIS and its features as of your current date in 2024.” – Open AI’s ChatGPT 3.5
It helpfully gave the last date it was updated, gave an accurate description of iHRIS and added a caveat about going to the official website for information. We confirmed that Open AI’s ChatGPT 3.5 could write basic installation instructions and challenged it to make the instructions as fun as we think health workforce management is. The answer did not disappoint.
“In the digital kingdom, where data flows like rivers, a legendary tool called iHRIS awaits brave souls seeking to conquer the challenges of workforce management. Prepare yourself for a thrilling quest to install iHRIS and harness its mighty capabilities.” – Open AI’s ChatGPT 3.5
Our developer team members located in Tanzania, Ethiopia, and Uganda tried ChatGPT 3.5 out for iHRIS and found it was easily able to add comments to existing code chunks. It could also provide step by step instructions in “black screen” mode so that a developer could get a reproducible example and put it into the software.
When we tried GitHub’s Co-Pilot to assist with software development, it could create repetitive code from scratch – like unit tests. But they required tweaking by developers who were skilled in writing code in Node.js and Vue.js (the codebase upon which iHRIS is written).
Future ChatGPT Use in Software Development
So, what did our iHRIS software development team say about ChatGPT?
- It helped with repetitive tasks that software developers don’t often enjoy, like writing documentation, especially when the logic is well-known.
- It helped reduce the time required to write the boilerplate and mock objects for tests since those are repetitive.
- However, it can generate code out of context or miss a single part which leads to more time spent on de-bugging.
- If GitHub Co-Pilot was free, there would be more uptake.
In the coming months, we will release a public user guide for incorporating LLMs and AI into the workflow for developers working with global public goods software.
Have you tried using Chat GPT or GitHub’s CoPilot for software development support? Start a conversation in the comments!
By: Amy Finnegan, Deputy Director of Data Science at IntraHealth International. The image is ChatGPT’s impression of African software developers working as wizards, using magic in their thrilling quest to blend modern technology with health data in a digital kingdom.
Interesting use case you are proposing. However, I would have appreciated to see some comments about the risks. For instance what you can find well explained here https://www.techtarget.com/searchSecurity/news/366571117/GitHub-Copilot-replicating-vulnerabilities-insecure-code. The ease of use of copilot means that people risk to not pay attention to what is proposed and either include dangerous bugs or even worse include poisoned code.
Also, I wanted to comment on your intro. You say “Artificial Intelligence, and especially Generative AI, is performing as well or better than humans in reading comprehension and language understanding. ” I trust this is a shortcut because the biggest risk for the community is precisely to believe that LLM understand. They do not. They are generating text to produce new content which is likely to be plausible based on the input and various other contextual parameters they can get hold of (e.g. the environment, the user history, etc). There is no understanding of the context or of the task at hand per say.
Hi Vincent – Thanks for sharing those points! ChatGPT definitely still needs “developers in the loop” to adjust the code. As to the intro, I was referring to this article on Our World in Data. https://ourworldindata.org/brief-history-of-ai Seems like the link got lost from an earlier version. Thanks again for raising these points.