USF Computational Social Science and AI - CSSAI Lab

We are excited to share that two of our Ph.D. students presented their work at the USF AI+X Symposium 2026.

1. Can a discussion with AI make us more intellectually humble?
================================================
You might have heard recent stories about the drawbacks of "Sycophantic AI", where chatbots give flattering and affirming responses. Instead, what if AI played the role of Socrates, who was somewhat unpopular for asking questions and probing people’s assumptions? We built a chatbot that does exactly that. Through a scaffolded conversation, it teaches people intellectually humble behaviors such as recognizing the limits of their knowledge and revising beliefs when strong evidence appears.

In a randomized experiment involving 400 participants, a short conversation with our Socratic-AI led to a measurable and durable increase in intellectual humility.

AI does not need to take away reflective agency. The same technology can make reflection easier. If chatbots are designed to flatter, they reinforce overconfidence. If they are designed to help us ask better questions, they may improve judgment.

Read more: Mahjabin and Baten (2026)
“General Intellectual Humility Is Malleable Through AI Mediated Reflective Dialogue” -- https://arxiv.org/abs/2603.23855

2. Why do negotiators fail to adapt when the situation changes?
================================================
Negotiation training is often built around static scenarios. But real negotiations are dynamic. Power shifts. Context changes. And people do not always adapt well.

We designed a controlled human experiment where 100 participants negotiated a job offer over six rounds. Midway through the negotiation, we introduced a sudden shift in leverage. What happened next was surprising. Negotiators became more competitive, less flexible, and more likely to damage the relationship, even when these changes did not improve their outcomes.

These failure patterns from the controlled simulation offer design implications for developing AI coaching tools that enhance humans' adaptability in negotiation.

Read more: Urmi and Baten (2026)
“Strategic Adaptation Under Contextual Change: Insights from a Dyadic Negotiation Testbed for AI Coaching Technologies” -- https://arxiv.org/abs/2602.04242

Can we nudge social media users to be less "rigid" in their beliefs? Our new paper in the IEEE Transactions on Affective Computing—the flagship journal of the field (IF 9.8)—tests a simple yet powerful intervention: randomization of algorithmic curation.

We found that while people have a natural "homophilic" tendency to engage with those they agree with, randomized recommendations can soften belief rigidity and promote more diverse network formation.

Read our paper, "Exploring the Role of Randomization on Belief Rigidity in Online Social Networks" here: https://lnkd.in/eSaXHkbe

Mohammad Ratul Mahjabin, one of our PhD students, is attending the ASEMFL 2025 Annual Meeting in Orlando to present a poster on his work titled "The Wisdom of Intellectually Humble Networks".

His research investigates how Intellectual Humility (IH) influences collective wisdom in networked settings. This event provides an excellent opportunity to showcase his work to a broader academic audience.

🔗 https://arxiv.org/abs/2502.02015

Congratulations to Mohammad Ratul Mahjabin, a PhD student of our lab, on being selected for a poster presentation at the upcoming ASEMFL 2025 Annual Meeting in Orlando. His simulation-based work investigates how Intellectual Humility (IH) influences collective wisdom in networked settings. This offers an excellent opportunity to present his work to a wider academic audience.

https://arxiv.org/abs/2502.02015

Accepted at EMNLP 2025! "MuseRAG: Idea Originality Scoring at Scale"

Creativity can be quantified in several ways. One key lens is social rarity: if too many people propose the same "creative" idea, it isn’t truly original. MuseRAG automates this originality scoring—resolving decades-standing technical hurdles in creativity assessment.

Preprint: https://arxiv.org/abs/2505.16232v1

Congratulations to Ratul Mahjabin, a Ph.D. student in the CSSAI Lab, on presenting his first grad-school paper at CogSci 2025.

The work examines how Intellectual Humility (IH) influences network performance on collective intelligence tasks. The findings show that IH leads to low error and polarization through balanced belief revision within the network.

We're excited to see more work on this front.

Paper Link: https://arxiv.org/abs/2502.02015

Our new paper just got published in Nature Humanities and Social Sciences Communications.

We show that scientists who heavily cite/collab with 'superstars' publish more—but innovate and disrupt less. The celebrity gravity in academia indeed comes at a price.

What can we do to incentivize bold ideas? Read to find out!

The innovation trade-off: how following superstars shapes academic novelty - Humanities and Social Sciences Communications Academic success is distributed unequally; a few top scientists receive the bulk of attention, citations, and resources. However, do these “superstars” foster leadership in scientific innovation? We employ a series of information-theoretic measures that quantify novelty, innovation, and impact f...

New from our lab: We built MuseRAG, a fully automated system that scores how original your ideas are—just like human judges, but faster and at scale.

It matches human annotators on 16k+ ideas, captures intent not just words, and brings psychometric rigor to creativity research.

Preprint: https://arxiv.org/pdf/2505.16232

Congratulations to Mohammad Ratul Mahjabin, a PhD student in the CSSAI Lab, for getting his first grad-school paper accepted! The work, "The Wisdom of Intellectually Humble Networks", has been accepted for an oral presentation at the top cognitive science conference, CogSci 2025.

The work employs theoretical modeling and data-calibrated simulations to probe how intellectual humility can make humans collectively wiser and less polarized. At a time of rising extremism and polarization, the need for humans to be more humble about the fallibility of their own knowledge has never been starker.

Paper link: https://arxiv.org/pdf/2502.02015

Mohammad Ratul Mahjabin, a PhD student in the CSSAI Lab, presented a poster at the USF Graduate Research Symposium 2025. The work, "The Wisdom of Intellectually Humble Networks", probes how intellectual humility—recognizing the limits of our own knowledge—can help societies become wiser and less polarized. Timely research for polarized times!