Mohit Chandra
@mohit__30
Followers
1K
Following
8K
Media
69
Statuses
664
PhDing @GeorgiaTech | Past: @MSFTResearch @Microsoft @iiit_hyderabad | Human-Centered NLP and Responsible AI for Mental Health | Opinions are personal
Chandigarh ✈️ Atlanta, GA
Joined July 2016
RT @munmun10: Loneliness can affect anyone, anywhere; it’s a public health issue. After year and half of work on the Technical Advisory Gro….
0
1
0
Hello Athens! 👋☀️. Excited to be attending #FAccT2025 and presenting our paper “From Lived Experience to Insight” on 24th June at 10:45 AM (New Stage C). Would love to catch up with old friends and make new ones and talk about AI and mental health 😄
1
0
48
We have much more covered in our paper!.Paper: Extremely grateful to my amazing co-authors 🙌:. @SuchismitaNaik, @denaefordrobin , @Ebele_Okoli_ , @munmun10 , Mahsa Ershadi, @GonzaloWorks , @javherriv , @Ananyabha , @warreth , Jina Suh. 10/10.
0
1
6
Operationalization of risk taxonomies is a challenging task. Current AI design pipelines primarily focus on evaluation of models through standard benchmarks which often lack the contextual information associated with human-AI interactions. Our Multi-path vignette-based.
1
0
0
Our work highlights the complex relationship between AI behaviors and resulting negative psychological impacts that are mediated through nuanced contexts. While personalization and alignment in AI could help in this case, there are open questions: . Which aspects of the user’s.
1
0
0
With shortage of mental health professionals around the world, AI agents can help. But our findings show it can lead to negative impacts. Therefore, we need to explore further general purpose AI systems should be used for support seeking purposes and, if so, how they can be.
1
0
1
Workshop participants recommended AI to avoid making assumptions about user’s intents and should set clear expectations by communicating if it is equipped to help achieve those goals. AI agents should use triggers to detect sensitive scenarios and redirect users to accessible.
1
0
0
During our workshop, expert participants provided recommendations in three broad areas 📝:. (1) How to Design Safer AI Conversational Agents. (2) How to Design Safer AI Conversational Agents. (3) How to Guide Users for Effective Engagement with AI Conversational Agents. 5/10.
1
1
2
Using our taxonomy, we identified 19 harmful AI behaviors, 21 negative psychological impacts, and 15 context categories. Through detailed vignettes we show how context mediates/exacerbates same AI behavior to different impacts and vice-versa 🔎. 4/10
1
1
1
We overcome the limitations associated with existing taxonomies/benchmarks through highlighting three components in our taxonomy ✅: . (1) AI behaviors.(2) Negative psychological impacts.(3) The contexts that either mediate or exacerbate these negative impacts. 3/10.
1
0
0
We conducted a two-phase study comprising of a survey study with 283 individuals with lived-mental health experience followed by three workshops with experts. We present a novel psychological risk taxonomy and introduce a framework for providing design recommendations. 2/10
1
0
0
Psychological risks of AI use have been under-represented in existing taxonomies. They often fail to capture the “lived experience” of users. So, how can we capture such risks and provide AI design recommendations? 🤔. We explore this in our paper, accepted at #FAccT 2025!.👇🧵
1
4
26
Thank you @GTResearchNews for covering our work! 😄 . I presented this work about two weeks ago at #NAACL2025. Checkout our paper: Blog: Code:
Think twice before you ask AI chatbots for psychiatric medication reaction advice, according to new research from @ICatGT's @mohit__30 and @munmun10.
0
4
21
I'll be presenting our work "Lived Experience Not Found" in the HC.1: Human-centered NLP oral session tomorrow at 10:15 am. Please join if you are around and would love to chat with you regarding Human-Centered NLP, Responsible AI and AI for Mental Well-being!. Paper:
First there was electricity, then the internet, now #LLMs. OK, we’re skipping a lot of history, but large-language models (LLMs) are a pretty big deal. Meet the next wave of inventors from @GeorgiaTech who are part of this new computing revolution. Highlighted are Munmun De
1
4
32
RT @inhwa_song_: (📣RT appreciated) I no longer have access to my prev account (@_inhwa_song) anymore 😭 Please re-connect with me through th….
0
3
0
RT @Sahithya_Ravi: Unexpected events grab human attention & push AI models beyond their training data. 🤔. How well do VLMs reason about the….
0
8
0
RT @rasbt: I think the problem is that self-reported LLM benchmarks don't account for inflation. I think as a rule of thumb, the longer a….
0
8
0
