Using AI for research
July 2, 2024
At Rangahau Aotearoa we have a keen interest in AI generative programs. Our specific use for these programs is to automate the thematic coding of the vast amount of consumer/stakeholder feedback we receive by way of the open-ended, free-text questions in our surveys. The key benefit we see of using AI generative programs for this important task are the standardisation of this process leading to more insightful, consistent and accurate analysis, with significant savings in both time and money for our clients.
ChatGPT gave us an initial ‘feel’ for the potential of AI generative programs, and although we had developed a process to ‘protect’ the data we were submitting, like many others, we’ve now reached the point where we’ve discontinued trialling this particular program. Our major concern was that, given ChatGPT is an OpenAI service, the data disappeared overseas and we had no idea of what it was being used for after we had submitted it. The only safety feature we’d been able to add was to thoroughly ‘clean’ and de-personalise the data (i.e., anonymising it) before submitting it.
An alternative we are currently considering is Microsoft 365 Copilot. The two key features which make this of interest are:
1. Data used for Copilot - including the information in the user instructions/prompts, the data retrieved and the responses generated - remain within the Microsoft 365 service boundary.
2. The service used for processing Copilot data is called ‘Azure OpenAI’ and this sits within the wider Microsoft Azure environment. This means that it is fully controlled by Microsoft and, therefore, not available to OpenAI.