Research | Shivansh Gupta

Work in progress

The Price of Progress: Gender Bias after #MeToo

Vasundhara Sharma, Deepa Mani, Aditya Karanam, and Shivansh Gupta

2026

Work in progress

Abs

Submitted to Nature. This project studies how social movements influence bias in student evaluations of faculty, with a focus on behavioral and perceptional changes.
Disclosing the Machine: The Effect of AI Transparency Mandates on Crowdfunding Performance

Vasundhara Sharma, Anitesh Barua, Ashish Agarwal, Aketi Gayatri, and Shivansh Gupta

2026

Work in progress

Abs

Draft in preparation. This paper studies how transparency about artificial intelligence reshapes market behavior in creative funding contexts. Using a platform-level policy change as a source of exogenous variation, we show that mandatory AI disclosure leads to significant declines in funding outcomes. We argue that disclosure activates a multi-channel devaluation process, wherein perceptions of reduced human effort, diminished authenticity, and heightened skepticism jointly suppress support. By linking causal estimates with discourse-based evidence, the study highlights how transparency can trigger adverse belief formation in environments where value is closely tied to human authorship.
The Spread of the Political Video DeepFakes on Social Media

Vasundhara Sharma, Anitesh Barua, Ashish Agarwal, and Shivansh Gupta

2026

Work in progress

Abs

Draft in preparation.

Research assistance

Curbing Hidden Advertisements in Children’s Videos: The Limits of Platform Policies

Vasundhara Sharma, Sumeet Kumar, and Ashish Agarwal

2025

Research assistance

Abs

Children’s YouTube videos often embed promotions that young viewers struggle to identify. Analyzing 200+ channels using LLM-based detection, we find YouTube’s policy did not reduce promotions and lowered disclosure. Brand-affiliated creators neither reduced promotions nor maintained disclosure, while non-affiliated creators reduced promotions but disclosed less, revealing limits of platform self-regulation.
Platform Design to Curb Misinformation

Vasundhara Sharma, Anitesh Barua, and Ashish Agarwal

2025

Research assistance

Abs

We propose and test a crowd-based content moderation approach to combat the spread of misinformation. In this method, users can flag posts they believe contain misinformation, with such flags serving as visible cues to others. These cues then influence users’ moderation and sharing decisions. We assess users’ willingness to participate in moderation and identify the underlying drivers. Thereafter, we examine whether such flags can address two primary causes for the spread of misinformation -lack of knowledge and scrutiny, specifically when posts align with one’s ideology. We developed a social media application and conducted a randomized controlled experiment. Participants were shown both true and false posts about politics, COVID-19, and the Russia-Ukraine conflict, along with social cues like flags and share counts. Our results demonstrate that users with a higher level of knowledge are more likely to flag misinformation and that their flags are not influenced by their personal beliefs. The presence of these flags encourages other users, especially those with less information, to be more cautious about sharing content. The presence of flags also reduces the spread of misinformation, even when users’ beliefs align with false posts. We also establish that users’ flagging and sharing behavior are driven by their updated opinions regarding the accuracy of the post and the need for impression management. Our study reveals a key limitation in the current approach to platform moderation, where user feedback is not made visible to others. It demonstrates that a crowd-based approach can be effective in curbing misinformation.

Patent

Method and System for Classifying and Detecting True Well Control Events During Drilling Operations

Jaspreet Singh, Somnath Das, Shivansh Gupta, Idris Kagdi, and Rohit Kumar

2023

Patent

Abs

Indian Patent, Published and under review, Patent No. 202311057956