Empathy in AI: When Conversational AI Turns into Agents of Empathy

In Apple TV’s dark comedy, mystery series, Sunny, Rashida Jones plays Suzie, an ex-pat mourning her missing (and presumed dead) family. She’s assigned an emotional support robot named Sunny, with eternally quizzical and concerned facial features, to help her carry on in the face of her loss. That and dealing with her dominating and disdainful mother-in-law. But can Sunny really be the authentic empathic companion who (that) understands what Suzie is actually feeling?

This resonates, doesn’t it? It mirrors our own doubts about AI. The idea of AI becoming “sentient” – thinking, feeling, and perceiving the world like a human – is a source of great debate and concern. Can they really act on their own? Not quite yet. 

Instead, what is possible today is to train Generative AI to “carry on coherent conversations in which they convey feelings, musings, opinions and other “reflections of consciousness” (Ellen Glover, link). Indeed, AI that can have conversations have spawned an explosion of Conversational AI solutions in the last two years in the market research industry.

But Not All AI is Created Equal

Some simulate the emotional empathy of in-person qualitative moderators better than others. So, can AI truly be empathic? Based on our experience, the answer is a resounding yes. 

Just like in traditional research, there are great moderators and not-so-great ones. The same goes for Conversational AI. The best AI models are trained to follow key research principles – avoid leading questions, don’t ask double-barreled questions, keep it simple, stay on topic, and perhaps most importantly, probe with empathy.

Training a model on best practice research principles is indeed important, but there is more. By allowing “human in the loop” feedback to steer the model’s performance, we can make it even better. Where researchers add context and objectives, it becomes possible to ensure the AI maintains empathy, while also meeting specific research objectives. This collaboration between human expertise and AI capabilities enhances the quality of interactions and the depth of insights gathered.

MMR recently conducted an experiment, using Nexxt Intelligence’s inca research-trained Conversational AI solution with the researcher in the loop against a different approach.

The difference in results was marked. Looking first at the less good Conversational AI:

• The 1st probing question could have been asked without any intelligence. It doesn’t build on the participant’s response. As a result, it doesn’t deliver greater depth. It is not engaging, and so we see a very short, limited response.
• The 2nd probing question is too long and complicated, making it feel like a “big ask”, trying to cover too many areas in one question. This is not only overwhelming, but also not specific enough to support clarity in the response and does not deliver further depth. It also risks starting to feel repetitive and annoying to the participant.

Using the inca research trained solution with the researcher in the loop, our conversational bots became agents of empathy, yielding much better results.

When done right, research-relevant, trained AI can simulate the empathy that provides a participant experience that gives much better research results, because it facilitates the following:

• Personal connection: The research participant feels like they are speaking to someone they know; questions and probing feel relatable, familiar, comfortable.
• Freedom to express: The participant feels free to answer all questions; they see this as a safe environment, indicative of sensitive listening.
• Gentleness: The follow-on probes are seen as gentle; the experience feels considerate, sensitive to the participant, and alleviates anxiety.
• Desire for continued interaction: Participants report being more willing to chat again; post interaction, participants express positive reactions to the experience, a sense of connection, and emotional needs having been met.

We may not yet have cute anthropomorphic robots who look like Sunny, but for us there is no doubt that when trained correctly and guided by human expertise, Conversational AI can become true agents of empathy. In this fusion of technology with a human touch, we see the potential for rich, truly authentic insights at scale.