In case you missed it (which seems unlikely), ChatGPT, the Artificial Intelligence model trained for conversation interactions has been making waves in the last few months. But once you’ve finished asking it to write songs about your cats in the style of Dolly Parton (guilty — and it wasn’t exactly the new Jolene) or explain Amazon Web Services (AWS) Fargate to you like you’re 5 (also guilty — and it involved cake, in case you’re interested), how might you leverage ChatGPT in novel ways in the development of tools and features for users of an insight platform?
Rather than asking ChatGPT itself, I tapped into another highly intelligent hive mind that I have at my disposal to ask silly questions of — the FlexMR Development team.
The team has already been experimenting with several of the applications below, leveraging several of the characteristics that make models such as ChatGPT useful in terms of the ability to: generate dialogue and converse convincingly, describe and summarise data and iterate over repetitive tasks. But here are five ideas for applying current-state AI to insight platforms.
1. Summarising Data
An insight platform will produce a very high volume of text-based answers and content from participants that can take a long time to find, read through, synthesise and summarise. What if we could plumb in AI to help parse some of those responses and find patterns and themes, especially in relation to user profiling data from databases and survey responses?
Take a 1.5-hour text or video focus group, for example — and then multiply that by the 3 or 4 you might run across a single project, and that’s a lot of time spent in reading transcripts, understanding and summarising. Although AI integration won’t replace the ability of a skilled researcher to draw out and synthesise insights, it will certainly give them ideas for promising themes to explore.
2. Conversational Search
Finding relevant participants on a user account summary page, or drilling down into a large number of video clips can be a dry business. But what if, rather than stacking up a series of filters, we could literally ask questions, for example: “Show me all the users who are between the ages of 18 and 35 and are in the North West of England”… “Now show only those who were not invited to a survey last week already”.
You would need to take care to ensure that search results don’t ‘fall through the gaps’ of what you’re specifying when a question is expressed in a way that might include filter items that are non-exclusive or not easily evaluated, but for exploratory searches, this may feel more intuitive than current standard filtering.
3. Generating Survey Questions
Many survey tools offer a bank of survey questions which researchers can draw from in order to build a survey, which is especially useful for very standard types of demographic questions that are asked across multiple surveys. If AI can be used to populate not only questions but also answer choices too, this makes for a much faster way to generate questions. It also makes it easier to write questions that conform to specific regions’ guidelines and standards when conducting multi-country studies.
Going one step further, AI could be leveraged to take charge not only of question content but question logic, to control what follow-up questions are shown based on responses in a more organic way than conventional question routing. This does need careful consideration when it comes to quota management and response analysis, though, and is therefore more applicable to exploratory studies.
This particular application has just been the subject of a recent experiment at FlexMR, with our Content Manager, Emily James exploring the current impact of ChatGPT’s survey writing capabilities in a separate blog.
4. Training and Familiarisation
Many online research tools take a little bit of practice to be used most effectively, and when these involve real-time research methods, convening colleagues, family and friends to help you try out and practice running an online focus group, for example, can be time-consuming in both the setup and the implementation.
Creating a virtual group of users, though, would help you to learn how to use and test out a new tool or feature without the need to use multiple colleagues’ time or learn on the go in a real session with real participants. It’s also far less scary!
5. Generating Status-Based Avatars
So far we have looked at research and analysis tools, but not dwelt upon the participant experience or engagement. As well as the desire to provide feedback, prizes and cold, hard cash, gamification and personalisation are the other strongly motivating factors in getting — and keeping — participants involved in online research communities.
We have already implemented the ability to set default participant avatars based on bands derived from a user’s participation in community activities, but image generation AI tools now are specific enough to allow multiple permutations of user avatars, which could be fed specific characteristics about the way the participant identifies themselves plus their community status to create an avatar that grows with them over time and thus increases recognition from other users.
In conclusion, these are just give novel ways that ChatGPT and other AI models might be leveraged in the development of tools and features for users of an insight platform.
Although AI integration won’t replace skilled researchers and community managers, it can certainly save time and provide promising themes to explore and interactive tools. There are several of these avenues which we might look to pursue in the development of our InsightHub over the next few years, so watch this space!
With thanks to Andrew Smith, Nana Akwah & Mark Brewster from the FlexMR development team for allowing me to shamelessly crib their creative ideas.
This article was originally published on the FlexMR Insights Blog and can be accessed here.