What does generative AI mean for qualitative research?
Generative AI inspired boatloads of hype and scaremongering in 2023.
Some say we about to enter a new era of productivity. The new tools at our disposal mean we can all do more with less.
Others say technology will end up providing “good enough” answers to client questions, rendering our unique skillset obsolete.
Keeping up with it all has been like drinking from a firehose. It reminds me of the ‘big data’ debate 10 years ago. Then, as now, our job is to evaluate which market developments and associated tools are fads and which are keepers – then get using them.
The outcome? A view on the tasks AI and humans are each uniquely good at.
An industry hackathon
Last week TMc team got our hands dirty in a hackathon organised by the AQR alongside 60 colleagues from across the industry. The goal? To participate, learn, and reflect together on what generative AI means for our professional community.
We got a feel for different platforms, trying them out on a range of tasks along the research cycle.
What was particularly useful was the collaborative feedback from the teams afterwards.
- Writing tools such as ChatGPT remove the blank page and allow you to make faster progress. Yet many felt the outputs could be generic and superficial, overly rational and requiring considerable skill and iteration to get right.
- One task related to NPD in the household cleaning category. It threw up a valuable example: typically this is undertaken by men in China – contrary to Western norms. This was a potentially misleading blindspot for Chat CGPT’s outputs.
- The outputs from free imagery tools were poorly received: unimaginative and clichéd. Paid versions fared better.
We then looked at three qual specific platforms. Each has the potential to provide significant time savings across the project cycle.
- Online community moderation and video clipping using Qualzy. The instant translation of moderator prompts into other languages, and the fast analysis of participant video was the highlight. Ready-made prompt suggestions for moderators were also invaluable.
- Summarisation using Quillit. Users thought this could save a significant amount of time creating a first draft of a report using transcripts. Getting to an answer faster.
- Thematic analysis and reporting using CoLoop. Users described this as a “very effective research admin assistant.” Several imagined a world where your groups from last night were available in the morning in a readymade analysis grid – leaving you to do the clever thinking.
Bringing the strands together
Ultimately, qual is all about nuance. Seeking nuance, seizing on nuance and reporting on nuance. Generative AI tools do the opposite: they flatten the jagged edges of human experience to a flat plane. Terabytes of computing power do not and will not remove the need for expert human judgement.
The verdict? High quality research cannot be automated.