Semra Sevi

Semra Sevi

@semrasevi.bsky.social

Assistant Professor of Political Science at the University of Toronto www.semrasevi.com

1.5K followers3K following45 posts

Science communication has never been more important. In this animation, @yamilrvelez.bsky.social, Donald Green and I break down our research exploring whether AI chatbots can increase political engagement among young, politically unaligned voters. Link to animation: www.youtube.com/watch?v=iCuX...

youtube.comCan AI Chatbots Increase Political Engagement of Young Unaligned Voters?YouTube video by Semra Sevi
Semra Sevi @semrasevi.bsky.social

🚨Excited to share our new paper published in PNAS (joint with @yamilrvelez.bsky.social and Don Green)! AI can enhance political knowledge and provide balanced information about politics with proper guardrails and vetted sources (e.g., party platforms). www.pnas.org/doi/full/10....

Semra Sevi reposted

🚨New WP: Can an AI voter guide (grounded in information from a nonpartisan, fact-checked source) help voters’ decision making? 🚨 We built and evaluated an LLM-based chatbot that provided voting info in CA & TX (N=2,474) right before the 2024 election. 🧵👇

Semra Sevi reposted

This paper was a blast to work on. The challenge: present party positions across many issues, in real time, using language voters actually use. 🧵 on why we went with a more involved retrieval-based approach and where I think these tools are headed.

Semra Sevi @semrasevi.bsky.social

🚨Excited to share our new paper published in PNAS (joint with @yamilrvelez.bsky.social and Don Green)! AI can enhance political knowledge and provide balanced information about politics with proper guardrails and vetted sources (e.g., party platforms). www.pnas.org/doi/full/10....

View full thread

Last week, Science and Nature published two articles highlighting the potential for AI chatbots to persuade and manipulate voters. We take a different approach: using AI to make political information more accessible.

Whether AI is ultimately used for good or ill largely depends on how govts, campaigns, and civil society orgs navigate this technological transformation. We hope our work spurs more innovation in building tools that expand access to democracy.

View full thread

Despite these knowledge gains, we observed little evidence of voters shifting toward the more proximate party, suggesting that issue proximity may not figure as prominently in the partisan attachments of young voters.

Across three experiments in the US, we found that VAA Bot enhanced knowledge of party stances on voters’ core issues by ~13 percentage points, with smaller spillovers on general issue knowledge (~4pp).

Semra Sevi reposted

@utoronto.ca, do better! The Uni is excluding recent hires from cost-of-living adjustments that an arbitrator decided ALL faculty/librarians deserve. This harms junior colleagues struggling w/ inflation: we’ve organized a petition demanding @utoronto.ca change course. Spread the word! www.utfa.org

utfa.orgUniversity of Toronto Faculty AssociationThe University of Toronto Faculty Association exists to protect and advance the well being of the faculty, librarians and research associates of the University of Toronto, the University of St. Michae...
View full thread

So why second at all? ✅ Shared party ✅ Common values ✅ Constituency interests In other words: homophily, not horse-trading. Sometimes, MPs just support what they believe in, not because they expect payback.

Why is reciprocity so weak in 🇨🇦? Strong party discipline limits side deals even in the more flexible world of PMBs. Our study, using a rare real-world lottery, shows: Legislative support often reflects shared values, not traded favours. Not all politics is transactional.

View full thread

Weak evidence for strategic seconding. MPs with better lottery spots are slightly more likely to second others, and there's almost no evidence that favours are returned in future parliaments.

In Canada, MPs are randomly assigned a spot in a lottery that determines who can introduce a private member’s bill (PMB). This lets us test: · Who supports whom · Whether support gets repaid · If it’s loyalty, strategy or something else

After a PMB is introduced, MPs can formally second it to show support. If reciprocity exists, we’d expect MPs with good lottery spots to second others hoping to get support back when it’s their turn.

🚨NEW PAPER: Do incumbents really have an edge in elections? Research says yes. But what happens when party labels are added to the ballot? A natural experiment from Canadian elections tells an interesting story👇🧵

View full thread

FINDING #2: Party matters more than person Incumbency advantage: Liberals pre-1972: +16 pts Conservatives pre-1972: +8 pts (not significant) Post-1972: Both parties? Advantage vanishes.

🚨NEW PAPER: Do incumbents really have an edge in elections? Research says yes. But what happens when party labels are added to the ballot? A natural experiment from Canadian elections tells an interesting story👇🧵

Before 1972, Canadian ballots showed only: • Candidate names • Occupations No party labels. No shortcuts for voters. Then came a 1970 law: Starting in 1972, ballots began listing party affiliations alongside candidate names.

Using data from 1867 to 2021, and a RDD, I estimate the causal impact of incumbency on electoral success. FINDING #1: The incumbency advantage shrank dramatically. ✅ Before 1972: Incumbents had a 15-point edge. ❌ After 1972: Just 2 points, and no longer significant.

Can AI reduce prejudice? 🇺🇸 In our new preprint, GPT-4o moved the needle on trans rights in a national U.S. study using morally tailored messages. The shift was real and short-lived. For more, read 👇 Comments and feedback are welcome.

John Holbein@johnholbein1.bsky.social

Prejudice is common in modern societies. Reducing prejudice is difficult to do, particularly in a cost-effective & scalable manner. Can AI help reduce prejudice? My new working paper w @semrasevi.bsky.social, @mbosley.bsky.social & Crabtree examines this question! papers.ssrn.com/sol3/papers....

Semra Sevi reposted

You are up to date.