|
David Rand@dgrand.bsky.social |
2167 followers 1491 following 49 posts
Prof at MIT working on misinformation/fake news, social media, polarization, intuition vs deliberation, cooperation, politics, and religion. (he/him)
davidrand-cooperation.com/
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Want to integrate GPT4 (or other LLMs) into Qualtrics, like we did in our recent paper on debunking conspiracy theories?
@tomcostello.bsky.social has it all set up, and you can download the qsf here: publish.obsidian.md/qualtrics-do...
0 replies 10 reposts 18 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
📢POSTDOC OPPORTUNITY📢
Come join me Gord Pennycook and @tomcostello.bsky.social
as we launch a research program using dialogues with LLMs as a tool to study human psychology and decision-making! Located at MIT or Cornell, and open applicants from all kinds of intellectual backgrounds/training!
0 replies 12 reposts 12 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Want to integrate GPT4 (or other LLMs) into Qualtrics, like we did in our recent paper on debunking conspiracy theories?
@tomcostello.bsky.social has it all set up, and you can download the qsf here: publish.obsidian.md/qualtrics-do...
0 replies 10 reposts 18 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨New in Science🚨
What FB news drove COVID vax hesitancy in US? False misinfo?
Not so much: We find unflagged ‘vax-skeptical’ news had *46X larger* impact than flagged misinfo
Why? Flagged misinfo had bigger impact when seen, but ~100x fewer views
www.science.org/doi/10.1126/...
0 replies 4 reposts 4 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Excited for this paper to be out!
0 replies 2 reposts 6 likes
|
Cameron Martel
@cameronmartel.bsky.social
[ View ] |
🚨New in PNAS Nexus🚨
Twitter field exps:
-Users block counter-partisans 12x more than co-partisans
-Dems block Reps more than vice-versa
Survey exps:
-Why block? To not see blocked user's posts
-Dem/Rep diff prob bc Dems block misinfo/toxic sharers more
academic.oup.com/pnasnexus/ar...
1 replies 10 reposts 23 likes
|
|
Cameron Martel
@cameronmartel.bsky.social
[ View ] |
🚨New review🚨 Do warning labels on misinformation actually work?
In Current Opinion in Psychology, @dgrand.bsky.social & I show that the answer is YES: warnings can reduce belief & sharing by ~25-50%!
We also review moderators (eg more visible=bigger effect)
www.sciencedirect.com/science/arti...
1 replies 37 reposts 68 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Agreed!
0 replies 0 reposts 0 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Ya this jives w work suggesting that eg misinfo warnings labeled as coming from AI are less effective than labeled as from fact-checkers dl.acm.org/doi/fullHtml...
So our treatment might work even better if people thought it was human- but maybe not (if eg theyd be more defensive). Future work!
0 replies 0 reposts 1 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨WP🚨
Conspiracy beliefs famously resist correction, right?
WRONG: We show that brief convos w GPT4 reduce conspiracy beliefs by ~20pp (d~1)!
🡆Tailored AI response rebuts specific evidence offered by believers
🡆Effect undiminished after 2mo
🡆Works even on entrenched beliefs
osf.io/preprints/ps...
12 replies 52 reposts 113 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
For MPSA folks, I'm going to impromptu present this paper in the next session: Fri, April 5, 11:30am to 1:00pm CDT Floor: 7th, Dearborn 2
0 replies 0 reposts 1 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Showtime!
0 replies 0 reposts 1 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
It's almost showtime!
0 replies 0 reposts 1 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
After a 4.5 hr flight delay, finally arriving in Chicago for MPSA! If you want some fun, come watch me and my 8yo son co-present in a session tomorrow (Fri) 945am Chicago time, Palmer House floor 3 salon 2
0 replies 0 reposts 6 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
It's a great question, and one we haven't tested yet (partly for ethical reasons, we don't want to talk people out of true beliefs). But I think it's reasonable to expect that it will work
0 replies 0 reposts 2 likes
|
Ross Dahlke
@rossdahlke.bsky.social
[ View ] |
Conversations with chat-based generative AI can durably reduce conspiratorial beliefs and worldview, finds
@tomcostello.bsky.social @gordpennycook.bsky.social @dgrand.bsky.social, suggesting that AI is a scalable potential solution to some deep social issue doi.org/10.31234/osf...
0 replies 6 reposts 6 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Thanks! Yes they knew they were talking to a computer. The polarized accusations topic was a real grabbag, things like Rs are purposefully keeping people uneducated, or anti abortion to increase number of adoptable kids, and similar for Ds. variation in effect size across topics wasn't significant
0 replies 0 reposts 2 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Got it, makes sense!
0 replies 0 reposts 1 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
What makes it feel manipulative for your perspective? To me, the issue is the potential for the AI to use inaccurate info (which it still might here). But if the info is accurate, is it still manipulative? (also it might nke be clear, but we explicitly told them it was AI they were talking to)
1 replies 0 reposts 0 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Prob with that is that it doesn't hold the actual evidence constant - so you'd need to have a condition where they talk to GPT but are *told* it's a human. Something like that is on our to-do list
2 replies 0 reposts 5 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Right!
0 replies 0 reposts 1 likes
|
Mike Caulfield
@mikecaulfield.bsky.social
[ View ] |
I am famously suspicious of some of these larger test models, but the fact people respond to reason/examples more than the other strategies doesn't surprise me. It's always been ridiculous to me that the working thesis of the field is that people don't care about evidence.
6 replies 3 reposts 24 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Pps were asked to identify & describe a particular conspiracy theory they believed in & provid details about evidence or experiences supporting their belief, then indicated their belief in an AI-summary of their statement using a 0-100 scale anchored at False and True
1 replies 0 reposts 8 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
heres a sample of how it works
2 replies 0 reposts 11 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
H/T lead author @tomcostello.bsky.social and co-conspirator @gordpennycook.bsky.social
0 replies 0 reposts 3 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨WP🚨
Conspiracy beliefs famously resist correction, right?
WRONG: We show that brief convos w GPT4 reduce conspiracy beliefs by ~20pp (d~1)!
🡆Tailored AI response rebuts specific evidence offered by believers
🡆Effect undiminished after 2mo
🡆Works even on entrenched beliefs
osf.io/preprints/ps...
12 replies 52 reposts 113 likes
|
Nora Newcombe
@noranewcombe.bsky.social
[ View ] |
Candice Odgers reviews Jonathan Haidt’s new book on social media and youth mental health. A very polite but very complete takedown. Required reading. www.nature.com/articles/d41...
6 replies 52 reposts 79 likes
|
|
Cameron Martel
@cameronmartel.bsky.social
[ View ] |
🚨Out in Psych Sci🚨 Prompting accuracy can increase news sharing quality - but is this true for those on the political right? In an ADVERSARIAL COLLABORATION we find: ➡️ Acc prompts increase sharing quality of Republicans ➡️ Some evidence of greater efficacy for those on left v right
1 replies 12 reposts 25 likes
|
Brian Guay
@brianguay.bsky.social
[ View ] |
*Less than half* of misinformation studies expose people to both false and true info, and ONLY 7% measure discernment (the difference between believing/sharing true and false info)!!
misinforeview.hks.harvard.edu/article/what...
Short 🧵 on why you should be measuring discernment👇
3 replies 25 reposts 37 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
If you're going to MPSA in Chicago and want to chat, send me a message or email!
0 replies 0 reposts 0 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨New PNASNexus🚨
We join work on misinformation & harmful language:
-More harmful language in tweets w low-quality news links β=0.1 & in false headlines β=0.19
-Users who share more misinfo use more harmful language in non-news tweets β=0.13 academic.oup.com/pnasnexus/ar...
w @mmosleh.bsky.social
2 replies 11 reposts 23 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Thanks for sharing!!
0 replies 0 reposts 1 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Many thanks!
0 replies 0 reposts 0 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Our dataset of 14k headlines available online:
osf.io/q5h49/
0 replies 0 reposts 0 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
CONCLUSION -Misinformation & harmful language *are* related in important ways- but not so strongly related that harmful language is a useful diagnostic for info quality -Shows opportunities to integrate largely disconnected strands of research & understand psychological connections
1 replies 1 reposts 3 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
RESULTS -Tweets with links to lower-quality news domains are more likely to contain harmful language -False headlines are more likely to contain harmful language than true headlines -Users who share more low quality links use more harmful language - even in non-news posts
1 replies 1 reposts 5 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
We study 8.6 million posts from 6832 Twitter users -classifiers identify harmful language -URL news domain quality scores to measure info quality Also analyze 14k true and false headlines (as evaluated by professional fact-checkers)
1 replies 0 reposts 1 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
But they might also be independent: - false claims need not involve harmful language - can insult and denigrate targets without making inaccurate claims So we wanted to investigate if they were actually related or not empirically
1 replies 0 reposts 1 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Misinfo & harmful language are both problematic, but treated largely independently. They could be connected: -hateful/toxic posts may use inaccurate or misleading claims about their targets to insult or belittle -posts that seek to mislead may use harmful lang as persuasive tool
1 replies 0 reposts 3 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨New PNASNexus🚨
We join work on misinformation & harmful language:
-More harmful language in tweets w low-quality news links β=0.1 & in false headlines β=0.19
-Users who share more misinfo use more harmful language in non-news tweets β=0.13 academic.oup.com/pnasnexus/ar...
w @mmosleh.bsky.social
2 replies 11 reposts 23 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨New WP🚨
Field experiments with 33 million FB users & 75k Twitter users: Ads prompting users to think about accuracy reduce misinformation sharing!
Accuracy prompts offer platforms a content-neutral approach that is scalable and preservers user autonomy osf.io/preprints/ps...
2 replies 27 reposts 48 likes
|
Alexander J. Stewart
@al-cibiades.bsky.social
[ View ] |
New paper in PNAS "The distorting effects of producer strategies: Why engagement does not reveal consumer preferences for misinformation" with @jplotkin.bsky.social, @dgrand.bsky.social and @arechar.bsky.social.
www.pnas.org/doi/epdf/10....
1 replies 11 reposts 16 likes
|
Lisa Fazio
@lkfazio.bsky.social
[ View ] |
Catch me next Wed March 6th at the Harvard/Northeastern Misinformation Speaker Series (live & on zoom) for the public reveal of the results from our large collaborative megastudy testing nine misinformation interventions @dgrand.bsky.social @lewan.bsky.social
shorensteincenter.org/new-event/mi...
1 replies 7 reposts 21 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨WP🚨
We test 9 online samples and find clear tradeoffs between attentiveness and representativeness - which sample is best depends on research q and priorities. For social/political qs I rec Bovitz/Lucid, for complex designs I rec Cloud/Prolific osf.io/preprints/ps...
1 replies 58 reposts 80 likes
|
Erez Yoeli
@erezyoeli.bsky.social
[ View ] |
If you run online surveys, you should read this paper.
0 replies 2 reposts 5 likes
|
Uma Karmarkar
@umakarma.bsky.social
[ View ] |
Amitai Shenhav and his lab are great people who do great science, and they're looking for a lab manager...
0 replies 2 reposts 3 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨WP🚨
We test 9 online samples and find clear tradeoffs between attentiveness and representativeness - which sample is best depends on research q and priorities. For social/political qs I rec Bovitz/Lucid, for complex designs I rec Cloud/Prolific osf.io/preprints/ps...
1 replies 58 reposts 80 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
Excited to present this paper at SPSP JDM pre-conference today, 4pm PST room 11b. Feel free to swing by (even if not registered for pre conf)!
0 replies 3 reposts 7 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
🚨New WP🚨
Field experiments with 33 million FB users & 75k Twitter users: Ads prompting users to think about accuracy reduce misinformation sharing!
Accuracy prompts offer platforms a content-neutral approach that is scalable and preservers user autonomy osf.io/preprints/ps...
2 replies 27 reposts 48 likes
|
|
David Rand
@dgrand.bsky.social
[ View ] |
I'll be at SPSP on Friday - if anyone is interested in meeting up, send me an email or DM!
0 replies 0 reposts 1 likes