TL;DR
I’ve been assisting a PhD student with her project on factors that influence human-AI interaction and it’s quality. While I can’t share details just yet, I’ve been taking notes of all that I’m learning with this project and think they’re valid seeds of thought to be in the garden.
Major Tasks taken on over the course of last year:
- Kick-starting exploratory research by combing through social media channels to see what users have to say about their interactions with LLMs, especially when the results are inaccurate/ the LLM refuses to answer (i.e a guardrail is invoked).
- Literature reviews on guardrail mechanisms and what taxonomies exist - Looking at papers like the Meta Llama Guard paper.
- Identifying the aspects of this interaction that influences user perception and thus the UX
- Designing the basis of a controlled experiment for users to take part in as we measure perception.
Updates on Study - Spring 2025
- Focus on how rejections are perceived
- Aligning with the CocoNot taxonomy for contextual non-compliance
Lit Review: Folk-theories and how it shapes user interaction (Facebook Paper)
- People’s understanding of how the Facebook algorithm works how they interact with it, and thus the UX