Sign in

Ashwini Ashokkumar

@ashwinia.bsky.social
1.5K followers 524 following 39 posts

Assistant Prof of Psychology. Studies the social psychology of groups, identity, politics, & language. www.ashwinia.com

PostsRepliesMedia
Ashwini Ashokkumar @ashwinia.bsky.social · 11/10/2026
Amazing opportunity!!! ⬇️
020
Ashwini Ashokkumar @ashwinia.bsky.social · 11/09/2026
📣 Excited to announce the Computational Psychology preconference at @spspnews.bsky.social in Philly! We have a great lineup of speakers! Submissions open for data blitz talks and posters! Submit by Oct 8. More info: 🧵
041
Ashwini Ashokkumar @ashwinia.bsky.social · 27/08/2026
New paper powered by not one but two dissertations, led by @small-schulz.bsky.social and @crobertson500.bsky.social!
041
Reposted by Ashwini Ashokkumar
Claire E. Robertson, PhD @crobertson500.bsky.social · 24/08/2026
Is social media truly making us more politically polarized, or does it simply over-represent our most extreme opinions? In our new preprint, @small-schulz.bsky.social and I present *punishing evidence* that social media systematically overrepresents users’ most extreme political opinions. 1/n
25232
Ashwini Ashokkumar @ashwinia.bsky.social · 01/08/2026
Thanks, Sakshi! Hope all is well! 💜
010
Ashwini Ashokkumar @ashwinia.bsky.social · 01/08/2026
Thanks, Sanaz!
000
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
Thanks, Robb! It was a lot a fun to work on this together. 😌
010
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
Paper + supplement: doi.org/10.1038/s415... Preprint: osf.io/preprints/ps... Demo: treatmenteffect.app Data + code: codeocean.com/capsule/9843... We hope others will test where this works, where it fails, and how it can be used responsibly!
doi.org
Large language models can predict the results of social science experiments - Nature
Large language models can be used to estimate the results of social science experiments about as accurately as a group of human forecasters—even for experiments published after the...
240
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
We are grateful to Jamie Druckman for his thoughtful engagement with the paper and his @nature.com News & Views piece. doi.org/10.1038/d415...
140
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
Grateful to have worked on this with my amazing co-authors @lukebeehewitt.bsky.social, @isaiasghezae.bsky.social, and @robbwiller.bsky.social.
140
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
To make the method easy to try, @lukebeehewitt.bsky.social built this very cool demo: treatmenteffect.app
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
This capability also carries risks of misuse. We asked whether GPT-4 can identify which anti-vaccine Facebook posts would most reduce vaccine intentions. Predictions matched real effects from @jennyallen.bsky.social et al.'s 2024 study: the posts flagged as worst were actually highly harmful.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
Important caveat: Actual treatment effects in our archive were largely homogeneous across demographic groups -- only 7-18% of effects were significantly moderated by gender, race, or party. For an ideal test of bias, we'd need an archive with more heterogeneity in effects across subgroups.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
On bias: GPT-4 's predictions were equally accurate for women vs. men. We found a small, significant gap in accuracy favoring white over Black samples. We found a similar (small) gap in prediction error (but not correlation) favoring Republicans over Democrats.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
Researchers worried most about stereotyping, unequal accuracy across demographic groups, underestimating human variability, reproducibility, replacing human samples, and interpretability. We share these concerns.
140
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
We also surveyed 460 social scientists about whether they would use a validated tool like this. ~76% reported a >50% likelihood of using AI for at least one purpose, esp: power analysis, intervention selection, pilot testing, and identifying effects that need replication.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
What is this useful for? - LLM-only/LLM+human pilots help test effects before running human experiments. - LLM forecasts help identify which small-sample effects would replicate in larger held-out samples. - Expert+LLM judgments help select more effective interventions than expert-only selection.
141
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
An important limitation: LLM predictions were about 2x too large, on avg. Accuracy also was lower for small effects, treatments on stable attitudes, field outcomes, and non-text treatments. LLM predictions can help inform research but should not replace human evidence.
170
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
We next tested a harder archive: 15 megastudies, 606 experimental effects, 4.86M participants, including field experiments, behavioral outcomes, and non-text treatments. Accuracy was lower (r=.39), though similar to the pooled expert forecasts in the nine studies that collected them.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
How did this compare to human forecasting? GPT-4 matched pooled forecasts from 2,659 people. And the two offered partly independent information: averaging human + GPT-4 predictions improved accuracy beyond either alone.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
But could GPT-4 just be retrieving published results from its training data? The correlation remained high for experiments not public by GPT-4's training cutoff, and for studies still unpublished in June 2025.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
Main result: effect estimates derived from GPT-4-simulated participants were highly correlated with actual effects (r=.85; adjusted r=.92)! This correlation remained high for experiments from a wide range of social sciences.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
We assembled 70 preregistered experiments: 50 nationally representative TESS studies + 20 replications (469 effects; 119,330 participants). We prompted GPT-4 with study materials + demographic profiles, simulated responses, then compared predictions to actual effects.
130
Ashwini Ashokkumar @ashwinia.bsky.social · 31/07/2026
New paper in @nature.com with @lukebeehewitt.bsky.social, @isaiasghezae.bsky.social, & @robbwiller.bsky.social: Can LLMs predict results of social science experiments? Across 70 experiments, we find a strong correspondence (r=.85) between predicted and observed effects. doi.org/10.1038/s415... 🧵
15118
Reposted by Ashwini Ashokkumar
Isaias Ghezae @isaiasghezae.bsky.social · 19/04/2026
🚨New Working Paper!🚨 How do Americans mentally map political coalitions? We often talk as if U.S. politics is just two sorted camps. But when people think about which groups go with which, the picture is more complicated. 🧵👇
14113
Reposted by Ashwini Ashokkumar
Danica Dillion @danicajdillion.bsky.social · 29/07/2026
It was a huge honor to welcome everyone to the Mind & Machine Alignment Summit to discuss how to make the future of AI go well. Computer scientists, social scientists, and policy experts offered unique perspectives on steering AI toward human flourishing. 🧵
171
Ashwini Ashokkumar @ashwinia.bsky.social · 29/05/2026
Congratulations!!! 🤩🤩🤩
110
Reposted by Ashwini Ashokkumar
Sakshi Ghai @sakshighai.bsky.social · 27/05/2026
Honoured to share our new paper in @nature.com today. Across 12 countries in eastern and southern Africa and Southeast Asia, 1 in 6 internet-using children had experienced technology-facilitated child sexual exploitation and abuse. A thread 🧵
34615
Reposted by Ashwini Ashokkumar
Marcel Roman @mfroman.bsky.social · 27/05/2026
🚨NEW PUBLICATION🚨 in Science Advances. We meta-reanalyze 100 conjoint experiments to assess which immigrants citizens prefer around the world. Our evidence reinforces existing findings, identifies novel insights, and provides a basis for future research 1/n www.science.org/doi/10.1126/...
science.org
Which immigrants do citizens prefer? A meta-reanalysis of 100 conjoint experiments
Meta-reanalysis of 100 conjoint experiments reveals patterns of similarity and variation in public immigration preferences worldwide.
15928
Reposted by Ashwini Ashokkumar
Neil Lewis, Jr. @neillewisjr.bsky.social · 22/05/2026
3/4. If you do those things, it will not necessarily make the world less chaotic...but, it will provide you with the community needed to get you through whatever life may throw your way...having that strong sense of community will be one of the greatest sources of satisfaction in your life."
1365
Reposted by Ashwini Ashokkumar
Neil Lewis, Jr. @neillewisjr.bsky.social · 22/05/2026
2/4. "In a world that glorifies the artificial, I encourage you to cherish real intelligence. Nurture it as if your health and well-being depend on it—because it does. Invest in the people around you, and allow them to invest in you...
15215
Reposted by Ashwini Ashokkumar
Robb Willer @robbwiller.bsky.social · 22/04/2026
...and this video of @nmalhotra.bsky.social and me discussing the study! www.youtube.com/watch?v=z2E3...
youtube.com
Political economist + sociologist's tourney tests climate change messages | Uncommon Collaborators
YouTube video by Stanford Woods Institute for the Environment
093
Reposted by Ashwini Ashokkumar
Robb Willer @robbwiller.bsky.social · 22/04/2026
For a short, public-facing overview of implications for talking about climate change across political divides, @woods.stanford.edu has a helpful explainer here: woods.stanford.edu/news/how-talk-about-climate-change-across-political-divides
woods.stanford.edu
How to talk about climate change across political divides
As climate impacts intensify while political progress remains stalled, new Stanford research examines which messages can shift public beliefs across partisan divides and strengthen climate communicati...
183
Reposted by Ashwini Ashokkumar
Robb Willer @robbwiller.bsky.social · 22/04/2026
Climate change impacts are here, but public support for action remains polarized. Which climate messages move people? In a ~13,500-person megastudy, we tested 10 of the most-cited messages. Six increased pro-environmental attitudes in the U.S.—but only by 1–4 percentage points. 🧵
611846
Reposted by Ashwini Ashokkumar
Jin X. Goh | 吴晋勋 @jinxungoh.bsky.social · 14/04/2026
TWO social psychologists got it!!!! Congrats to Mina Cikara and @olsonista.bsky.social !!!!
0444
Ashwini Ashokkumar @ashwinia.bsky.social · 07/04/2026
Thanks, Steven. I appreciate it. 🙏
010
Ashwini Ashokkumar @ashwinia.bsky.social · 07/04/2026
www.theindiaforum.in/caste/algori...
theindiaforum.in
Algorithms Don't Need Your Surname to Guess Your Caste. That's a Problem
A Dalit student graduating in the mid-2020s will encounter screening systems built on decades of unequal hiring, skewed promotion data, and biased public records. Unless institutions change how algori...
141
Ashwini Ashokkumar @ashwinia.bsky.social · 06/04/2026
We study identity, group dynamics, language, & politics. Especially encouraging applicants with experience/background in managing and analyzing data. Lab + research: www.ashwinia.com Happy to answer quick questions by DM!
ashwinia.com
We-Search Lab
We-Search Lab
010
Ashwini Ashokkumar @ashwinia.bsky.social · 06/04/2026
🚨🚨 My lab is hiring a Pre-doc Researcher/Lab Manager! 🚨🚨 Full-time pre-doc research opening in my lab at Harvard Psych (start: Summer/Fall 2026). Review begins April 10 Apply: jobs.smartrecruiters.com/HarvardUnive... Please repost + tag folks who might be a good fit! 👇
jobs.smartrecruiters.com
Research Assistant II
Company Description: By working at Harvard University, you join a vibrant community that advances Harvard's world-changing mission in meaningful ways, inspires innovation and collaboration, and builds...
13133
Reposted by Ashwini Ashokkumar
Sarah J. Jackson @sjjphd.bsky.social · 08/03/2026
Sonny Burton is set to be executed by the state of Alabama on Thursday. There is a direct line between a government that murders its own citizens and the devaluing of life abroad. The sin has always started at home.
nbcnews.com
Alabama set to execute man who did not kill anyone
Unless Alabama Gov. Kay Ivey intervenes, Charles “Sonny” Burton, 75, will be executed Thursday. He is among thousands in America convicted under a controversial felony murder law.
131285613
Reposted by Ashwini Ashokkumar
William J. Brady @williambrady.bsky.social · 26/02/2026
New work from @drsanaz.bsky.social : using smart phone sensing data, one of the biggest predictors of people higher on authoritarian measure was Facebook and social media use #spsp
1172
Ashwini Ashokkumar @ashwinia.bsky.social · 26/02/2026
Hi Louisa, sorry about the DM thing. 😄 Would you mind emailing please?
110
Ashwini Ashokkumar @ashwinia.bsky.social · 24/02/2026
Especially encouraging applicants with computational, big data, and/or experimental expertise! Lab + research: www.ashwinia.com Happy to answer quick questions by DM!
ashwinia.com
We-Search Lab
We-Search Lab
110
Ashwini Ashokkumar @ashwinia.bsky.social · 24/02/2026
🚨🚨 My lab is hiring a Postdoc! 🚨🚨 Postdoc opening in my lab at Harvard Psych (start: Summer/Fall 2026). We study identity, group dynamics, language, & politics. Rolling review begins 3/14 Apply: academicpositions.harvard.edu/postings/15806 Please repost + tag folks who might be a good fit! 👇
academicpositions.harvard.edu
Postdoctoral Fellow
Dr. Ashwini Ashokkumar’s lab at the Department of Psychology at Harvard University invites applications for a Postdoctoral Fellow position to start in Summer 2026. The lab conducts research on social ...
12534
Reposted by Ashwini Ashokkumar
Richard Baker @bakerart.bsky.social · 21/02/2026
Shawls, pastel (one of my old May Week theme drawings) #Cambridge #mayball #pastel
Pastel sketch of ladies assembling to attend a May Ball
732951263
Reposted by Ashwini Ashokkumar
Feyaad Allie @feyaadallie.bsky.social · 17/02/2026
Call for Papers: Workshop on Contention Under Repression We invite scholars in the Greater Boston area to submit proposals for a one-day workshop on 4/17 on collective action, civil society, contentious politics, etc. across global contexts Details: tinyurl.com/wrkshp417 Submission Deadline: 3/15
Conference Details
171
Reposted by Ashwini Ashokkumar
Eric Shuman @ericmshuman.bsky.social · 19/01/2026
I am looking to recruit a postdoc to join my lab next fall, & work on projects focused on online mobilization with social media datasets - see more details here: bit.ly/4r49x9b . If anyone knows of someone with relevant skills who is looking for a postdoc- please pass this on!
bit.ly
Research Associate in Psychology
Dr. Eric Shuman at the University of Virginia invites applications for a post-doctoral research associate with expertise in the social psychology of social protest, social change, and/or in big data, ...
01110
Reposted by Ashwini Ashokkumar
Molly Crabapple @mollycrabapple.bsky.social · 18/11/2025
Hey authors! Check to see if Anthropic stole your book to train their slop generator on. You’re entitled to $1500 per stolen Work. Look up your work, and if you’re in the database, file a claim secure.anthropiccopyrightsettlement.com/lookup/
secure.anthropiccopyrightsettlement.com
Submit a Claim
7730602523
Reposted by Ashwini Ashokkumar
Mohamed A. Hussein @mhusseinlab.bsky.social · 01/09/2025
🚨New Paper🚨 Elected officials are increasingly extreme. E.g., a recent analysis of 84,000 state-level candidates found that extreme candidates are now winning at the highest rates in 30 years. Why are people increasingly drawn to extreme candidates?
6338
Reposted by Ashwini Ashokkumar
Joel Le Forestier @joelleforestier.bsky.social · 21/08/2025
📣📣We're going for round two!!!📣📣 I'll be reviewing graduate applicants again this year. Come join my growing lab! Info about the lab: joelleforestier.com Info about the social program: www.psychology.pitt.edu/graduate/soc...
joelleforestier.com
Identity & Intergroup Relations Lab
Identity & Intergroup Relations Lab in the Department of Psychology at the University of Pittsburgh, directed by Dr. Joel Le Forestier, studying intergroup experiences, prejudice, and interventions.
22212