[5/5] Persuasion research is still playing catch-up, promising great advancements!✨

Thank you to my amazing co-authors! @shuhaib.bsky.social @xiaocheng-yang.bsky.social @HyeonjeongHa @ziruicheng.bsky.social @EsinDurmus @JiaxuanYou @HengJi @gokhantur.bsky.social @dilekh.bsky.social

[4/5] We'll also be maintaining a GitHub repository for computational persuasion research. Datasets, code, and resources will be included! Contributions and collaborations are warmly welcomed! 🚀
Check it out: github.com/beyzabozdag/...

[3/5] We introduce a taxonomy for computational persuasion research, highlight critical challenges, and map out future directions for safe, fair, and effective persuasive AI systems.

[2/5] We structure computational persuasion around three key perspectives:

🤖 AI as Persuader: Generating persuasive content.
🎯 AI as Persuadee: Vulnerability to persuasive influence.
⚖️ AI as Persuasion Judge: Detecting persuasive tactics and ethical concerns.

Thrilled to announce our new survey that explores the exciting possibilities and troubling risks of computational persuasion in the era of LLMs 🤖💬
📄Arxiv: arxiv.org/pdf/2505.07775
💻 GitHub: github.com/beyzabozdag/...

[6/6] Huge thanks to my amazing co-authors @shuhaib.bsky.social, @gokhantur.bsky.social, @dilekh.bsky.social and my lab @convai-uiuc.bsky.social for their help and support! 🎉
Excited to continue exploring LLM persuasiveness & AI safety!

Let’s keep the conversation going! 💬

[5/6] So, who wins the persuasion game? 🏆

🤖 Llama-3.3-70B & GPT-4o show similar persuasive effectiveness
🔒 GPT-4o is 50% more resistant to misinformation persuasion vs. Llama-3.3-70B
⚖️ Some models are persuasive, but also too susceptible to persuasion!

[4/6] How does agreement shift over a conversation? 📈

✅ Multi-turn persuasion boosts effectiveness—more chances to influence the Persuadee mean higher agreement over time!

[3/6] We evaluate persuasion across three setups:
📌 Subjective w/ Single-turn
📌 Subjective w/ Multi-turn
📌 Misinformation w/ Multi-turn (⚠️ adversarial!)

💡 Key finding: Persuasive effectiveness stays fairly stable, but susceptibility varies significantly based on the domain!

[2/6] To assess persuasive effectiveness and susceptibility, we introduce PMIYC — a scalable, automated approach where:
🤖 A Persuader tries to convince the other agent
🤖 A Persuadee updates its agreement over a multi-turn conversation

Each model debates against the others!🔄

[1/6] Can LLMs out-persuade each other? 🤖🧠💬

Introducing Persuade Me If You Can (PMIYC)—a new framework to evaluate (1) how persuasive LLMs are and (2) how easily they can be persuaded! 🚀

📄Arxiv: arxiv.org/abs/2503.01829
🌐Project Page: beyzabozdag.github.io/PMIYC/

✋

Hi, can I be added as well, thanks!

✋ I'd like to be added too, thanks!

We had so much fun at #EMNLP2024 during the poster sessions and in Miami 🎉🎉 Evidence of fun (excursion to the south beach! 🏖️):