Could we learn to love AI?

Last month I chaired the morning session at the AQR’s Humanity Fights Back conference exploring how AI is impacting qualitative research. 

It’s a big topic with lots to get your head around. The easy part is understanding how the underlying technologies work and the pros and cons of generic and research specific tools. Understanding the second order effects and unintended consequences is much harder.  

Anna Parker from the BBC and Sherry Nugent from IPSOS delved into public perceptions of AI, exploring how they feel about its use by the BBC. Takeaways included:

• Scepticism and fear of the unknown are common.
• There is a tipping point when people see a tangible benefit or use case for AI.
• Transparency fosters trust: what is being used for what purpose? This is especially important for an organisation like the BBC.  

Heeru Sajnani and Charlotte Joel from The Ninth Seat discussed innovative ways to engage participants using AI stimulus. On a project exploring cleaning products they replaced the usual stock images with hyper-realistic AI generated imagery. They found:

• AI can provoke playful and fun responses, helping participants articulate themselves better.
• Participants were more likely to tell stories and give extended responses when responding to the AI imagery.
• Being able to generate pictures during focus groups enables fast iteration (e.g. “here’s an image of a new product with the attributes you just mentioned”). This can enhance the participant experience.

Richard Owen took us through some of Firefish’s “research on research” – focusing on the relative strengths of pure AI research versus AI as a tool:

• AI is valuable for speed and scale, particularly in scenarios where traditional qualitative methods would be impractical. For example, quali-quant methods allow for 100 mini interviews overnight across multiple languages with a report in 24 hours. 
• It is less valuable for exploratory or futuristic research where human insight is crucial.
• It’s importance to keep a human in the loop: clients pay for meaning and human understanding. 
• As standalone AI analyses become more well known, the role of human curation and interpretation becomes clearer. We could be entering a “golden age” for qual therefore. 

Luke Meddings from Trinity McQueen highlighted the unique and nuanced nature of human communication in a thought-provoking session:

• Human communication combines the verbal and the non-verbal, and is complex. AI relies on a context-free transcript which contains a fraction of the embedded meaning in a conversation.
• Qualitative research is ultimately about interaction, not extraction of data. The “dance” of communication involves reacting and responding in subtle ways that AI cannot fully replicate.
• Language is as a highly evolved social tool: if it were a newly invented product, it would be hailed as a revolutionary gateway to our private thoughts.

The future

The underlying question is will the technology replace the human? This is a question with several layers. But the short answer is no.  

At foundational level

It is reasonable to argue that there is no “intelligence” in the Large Language Models (LLMs) which Generative AI is built on. LLMs do not reason. At best they imitate reason “probabilistically” (predicting the next word in a sequence). This is why they “hallucinate” (where a query returns answers which are entirely incorrect). 

At a tool level  

Qual specific platforms add most value at the data collection (AI enabled chatbots enabling quali-quant at scale) and analysis stages of a project (auto transcription and summaries). The sheer volume of material these platforms can consume through the brute force of processing power is awe-inspiring. However, outputs still careful checking, interpretation and augmentation. 

At a business level

Working with clients, teasing out their needs and delivering business-changing recommendations isn’t easy. It relies on “soft skills” which are tacit, nuanced and require empathy. Your AI model isn’t going to take your client to lunch. 

The essence of qualitative 

As qualitative researchers, it is our responsibility to capture and convey the rich, multifaceted nature of human communication – even as we integrate AI tools into our practice.

Our data? Direct, unmediated interaction with people which captures the full context of their emotions and stories. 

Our process? It is about what people say (language), how they say it (emotion), what they don’t say (body language) and the context (where they are).

The outcome is nuanced. It is about meaning. It requires judgement. 

To deliver the best advice to clients we must “Be faithful to our own perceptions and transmit them as faithfully as we can.”