@sunniesuhyoung@hci.social

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 18mo ago

Replying to

@jenn @qveraliao @PrincetonCS @hci @citp "Fostering Appropriate Reliance on LLMs" received an Honorable Mention at #CHI2025 This work is also the last chapter of my dissertation, so the recognition feels more special🏅🎓😊 🎉 to the team!!! https://programs.sigchi.org/chi/2025/program/content/188664

programs.sigchi.org

6

2

4

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 24mo ago

Super honored and happy to be selected as an EECS Rising Star and a Siebel Scholar🌟 I'll also be on the job market this year looking for positions where I can continue AI + HCI research on responsible AI!

https://risingstars-eecs.mit.edu/participants/?y=2024
https://www.siebelscholars.com/articles/siebel-scholars-foundation-announces-class-of-2025/

@hci@micro.blogs.princeton.edu

risingstars-eecs.mit.edu

6

0

2

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 28mo ago

📢 Recruiting for a paid study about Explainable AI and birds identification 🐦

We're looking for US-based participants for a 1.5 hour study on Zoom (payment: $25 Amazon gift card). Open to diverse levels of expertise in ML and birding.

If interested, please fill out 👇
https://forms.gle/F7bmWkVG7JyT6XA27

@parastoo@hci.social @princetonhci@bird.makeup

RT appreciated!

forms.gle

2

1

4

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 19mo ago

Replying to

First, through a think-aloud study (N=16) in which participants use ChatGPT to answer objective questions, we identify 3 features of LLM responses that shape users' reliance: #explanations (supporting details for answers), #inconsistencies in explanations, and #sources. 2/7

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 19mo ago

Replying to

Then, through a pre-registered, controlled experiment (N=308), we isolate and study the effects of these features on users' reliance, accuracy, and other measures, such as confidence, source clicking, task time, evaluation of LLM responses, and asking of follow-up questions. 3/7

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 19mo ago

Replying to

We find that the presence of #explanations increases reliance on both correct and incorrect LLM responses. This isn't surprising, given prior HCI/Psychology work on other types of explanations, but raises the question whether LLMs should provide explanations by default 🤔 4/7

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 19mo ago

Replying to

On a more optimistic note, we observe less overreliance on incorrect LLM responses when accurate and relevant #sources are provided or when explanations from LLMs exhibit #inconsistencies (i.e., sets of statements that cannot be true at the same time). 5/7

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 19mo ago

Replying to

This work is with @jenn, @qveraliao, Tania Lombrozo, and Olga Russakovsky, a fantastic collaboration between @PrincetonCS, @hci, @citp, Princeton Psychology, and Microsoft Research FATE 🤝 A few of us will be at #CHI2025 in person! 📄 https://arxiv.org/abs/2502.08554 7/7

arxiv.org

1

2

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 40mo ago

Replying to

This is another joint work with Elizabeth Anne Watkins, Olga Russakovsky, Ruth Fong, and @andresmh 🥳 Humans, AI, and Context: Understanding End-Users’ Trust in a Real-World Computer Vision Application 📄 https://arxiv.org/abs/2305.08598 🎥 https://youtu.be/2e1frsZ0NDw 🌐 https://sunniesuhyoung.github.io/XAI_Trust/ 5/6

3

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 44mo ago

Join the 2nd Explainable AI for CV (XAI4CV) workshop at #CVPR2023!
https://xai4cv.github.io/workshop_cvpr23

👥 The workshop will take place on June 19th in Vancouver, Canada. More to follow on in-person/hybrid setup

🎤 We have a fantastic line-up of speakers: @qveraliao@hci.social, Mohit Bansal, Marina M.-C. Höhne (née Vidovic), @arvind@vis.social, Alice Xiang, and @davidbau@sigmoid.social

📄 Submit papers and demos by March 14th (proceedings track) and May 19th (non-proceedings track)

hci.social

3

0

2

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 40mo ago

Replying to

(2) We also found trust in AI is influenced by many many factors. While this is not a complete set of factors that can influence trust in AI, but what we observed in our study, we hope it helps you anticipate what can influence trust in AI in your context of interest. 4/6

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 40mo ago

Replying to

We’re excited to present it at FAccT next week! 6/12 Monday 3:15-4:15pm CDT in W196BC Paper session 1: Decisions, Trust & Reliance Come say hi👋 Most of us are attending in person! 6/6

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 43mo ago

Replying to

1️⃣ XAI needs We found that participants' needs for system details varied depending on their AI background and domain interest. But everyone wanted practically useful information that can improve their collaboration with AI. 3/7

1

4

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 43mo ago

Replying to

💭 Here is a visual summary of the creator-consumer gap in XAI observed. Check out the paper for details! "Help Me Help the AI": Understanding How Explainability Can Support Human-AI Interaction 📄 https://arxiv.org/abs/2210.03735 🎥 https://youtu.be/IcVsi5-ON4Y 6/7

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 43mo ago

Replying to

❤️ I loved working on this paper with Elizabeth Anne Watkins, Olga Russakovsky, Ruth Fong, @andresmh and learned so much. Looking forward to presenting and discussing it at #CHI2023 in person! Will be my first CHI!!! 7/7

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 40mo ago

Replying to

To provide a holistic and nuanced understanding of trust in AI, we examined multiple aspects of it. General aspects include trustworthiness perception and trust attitude. Instance-specific aspects include AI output acceptance and AI adoption decision-making. 2/6

0

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 43mo ago

Replying to

🤝 We bridge the gap between XAI research done in the HCI and AI communities by directly connecting real AI end-users with the XAI methods literature. We do so by mocking up 4 XAI approaches that could be implemented into the app. 2/7

0

5

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 40mo ago

Replying to

(1) Intriguingly, we found a gap between general and instance-specific aspects. Participants assessed the AI as trustworthy and trusted it. Still, they selectively accepted the AI's outputs after verification and decided against AI adoption in certain high-stakes scenarios. 3/6

0

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 19mo ago

Replying to

We provide an in-depth discussion of our findings, bringing together the AI, HCI, and Psychology literature. We also contribute nuanced insights on people's interpretation of explanations from LLMs (e.g., are they faithful or not?), source clicking behavior, and more. 6/7

0

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 43mo ago

Replying to

2️⃣ XAI uses When participants saw mock-up explanations, they wanted to use them for purposes beyond understanding AI outputs. These highlight the broad range of user goals that should be considered in XAI development. 4/7

0

3

1

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 43mo ago

Replying to

3️⃣ XAI perceptions Finally, we found that participants' perceptions of XAI approaches varied based on their AI background. But overall, participants preferred part-based explanations that are similar to human reasoning and explanations. 5/7

0

2

0

Sunnie S. Y. Kim ☀️ @sunniesuhyoung@hci.social

· 28mo ago

Replying to

@thatandromeda @parastoo Sadly no, participants have to be over 18 years old. Thanks for asking!

0