4/5 Sentence, "Full of statistical neurons floating, weaving, intersecting daily."
This film/artwork explores the differences between an empathetic chatbot (Replika) and a research-trained AI assistant. Through visual and auditory comparisons, it reveals the strengths and limitations of each model in interpreting and responding to emotional prompts.
Visual outcomes
Development
I was starting to feel pretty isolated and frustrated, working on this massive project by myself. It was getting a bit too overwhelming, and I realized I needed help—especially from someone with a science background. The scope of it was beyond me, and I had to be realistic: I couldn't do it all alone. So, I decided to explore potential collaborations.
The previous year, I'd noticed an opportunity to apply for arts and science-based projects at King's College. I thought, why not give it a shot? I attended some talks, met a few people, but it didn't seem to click. Then, out of the blue, I was contacted by Arunav Das, who said he loved the project and wanted to collaborate. I remember feeling so excited. Finally, I wasn't in this alone. Even better, there was someone with more expertise than me to help bring this to life. Arunav was doing a postdoctoral research project in the Informatics Department at King's College and had a solid understanding of language models and chatbot interactions.
This was the turning point for the project. As I mentioned earlier, it was at this moment that I realized the interactions were the most crucial element. If you can convince a person they're having a conversation with someone who feels real—someone who could be human—then the interaction becomes meaningful. It's in these moments of connection where we feel things, where we relate to others and to the world around us, and ultimately, how we comprehend each other emotionally.
Film 1: Sentence 4/5 Visualising and interpreting AI empathy
Film 2: Analysis with Interactive Touch-board
Concluded hypothesis : Heidi Stokes & Arunav Das
In this analysis, we examined the patterns that emerge when comparing the response of an Empathetic trained chatbot to that of an assistant in shape sound and hand gesture analysis.
This Analysis showed that the research assistant consistently fulfilled specific tasks when responding to emotional prompts. In contrast, the empathetic AI chatbot often generated its own suggestions instead of completing assigned tasks. While this tendency could be frustrating, it showcased the chatbot's independent thinking capabilities. Replika generated novel outcomes based on provided information. However, the research assistant outperformed the empathetic AI in more complex tasks, such as providing emotive prompts for hand gestures. This highlights the assistant's proficiency in handling