Lightnews — Scholar-powered news

Reposted by Ben Tappin

Felix M. Simon @felixsimon.bsky.social · 5d

🗞️ 🤖 Weekend reading anyone? For the launch of @transformernews.ai as a standalone publication, they invited me to contribute a piece on what persuasive AI might mean for democracy and elections.

Here’s the result…

buff.ly/OJsNmpK

AI is persuasive, but that’s not the real problem for democracy

Opinion: Felix M Simon argues that AI is unlikely to significantly shape election results in the near future, but warns that it could damage democracy through a steady erosion of institutional trust.

www.transformernews.ai

1 8 5

Reposted by Ben Tappin

Per Engzell @pengzell.bsky.social · Sep 1

WE ARE HIRING! 2 Lecturers in Quantitative Social Science. Want a friendly interdisciplinary department in one of the world's most vibrant cities? This just might be for you.

Apply by: 10 Oct

www.ucl.ac.uk/work-at-ucl/...

3 160 150

Reposted by Ben Tappin

Mike Masnick @mmasnick.bsky.social · Aug 5

Wrote about how the UK's online age verification requirements are already proving to be a disaster (as UK regulators were clearly warned it would be) and how unhelpful the UK's response to this mess has been, including their tech minister saying anyone who complains supports predators.

Techdirt @techdirt.com · Aug 4

Didn’t Take Long To Reveal The UK’s Online Safety Act Is Exactly The Privacy-Crushing Failure Everyone Warned About

Well, well, well. The “age assurance” part of the UK's Online Safety Act has finally gone into effect, with its age checking requirements kicking in a week and a half ago. And what…

Didn’t Take Long To Reveal The UK’s Online Safety Act Is Exactly The Privacy-Crushing Failure Everyone Warned About

Well, well, well. The “age assurance” part of the UK's Online Safety Act has finally gone into effect, with its age checking requirements kicking in a week and a half ago. And what do you know? It's turned out to be exactly the privacy-invading, freedom-crushing, technically unworkable disaster that everyone with half a brain predicted it would be. Let's start with the most obvious sign that this law is working exactly as poorly as critics warned: …

www.techdirt.com

6 390 810

Reposted by Ben Tappin

Prolific @joinprolific.bsky.social · Jul 24

The largest investigation of AI persuasion with 76,977 participants across 3 large-scale experiments. Excellent work by @ox.ac.uk PhD candidate @kobihackenburg.bsky.social.

19 LLMs. 707 political issues. 466,769 fact-checkable claims evaluated.

arxiv.org/abs/2507.13919

#AcademicSky #MLSky #PhDSky

A still image of the PhD candidate Kobi Hackenburg's AI persuasion research paper titled "The Levers of Political Persuasion with Conversational AI."

1 3 9

Reposted by Ben Tappin

Helen Margetts @helenmargetts.bsky.social · Jul 22

I am so proud of the brilliant @oii.ox.ac.uk DPhil @kobihackenburg.bsky.social and this wonderful “bees-knees” paper on conversational AI and political persuasion @AISecurityInst - it is a “must-read”, comments welcome!

Kobi Hackenburg @kobihackenburg.bsky.social · Jul 21

Today (w/ @ox.ac.uk @stanford @MIT @LSE) we’re sharing the results of the largest AI persuasion experiments to date: 76k participants, 19  LLMs, 707 political issues.

We examine “levers” of AI persuasion: model scale, post-training, prompting, personalization, & more!

🧵:

3 16

Ben Tappin @benmtappin.bsky.social · Jul 22

A great post, this point especially:

"I don’t mean to argue all research needs to be slow and fully documented. When we are just starting in a new area, it’s chaos. But at some point, by the time results are reported, the workflow needs to be professionalized. Research is not a hobby. It’s a job."

Richard McElreath 🐈‍⬛ @rmcelreath.bsky.social · Jul 9

How can we reform science? I have some ideas. But I am not sure you’ll like them, because they don’t promise much. elevanth.org/blog/2025/07...

Which Kind of Science Reform

What hope is there for science reform, if we can't agree on what to reform? Right now, principles are more important than practices.

elevanth.org

Reposted by Ben Tappin

David Rand @dgrand.bsky.social · Jul 21

VERY excited about this massive AI persuasion experiment - big effects on UK policy positions, driven mostly by post-training (rather than hyperpersonalization or model scale); and the more info the model provides, the more persuasive it is

Kobi Hackenburg @kobihackenburg.bsky.social · Jul 21

Today (w/ @ox.ac.uk @stanford @MIT @LSE) we’re sharing the results of the largest AI persuasion experiments to date: 76k participants, 19  LLMs, 707 political issues.

We examine “levers” of AI persuasion: model scale, post-training, prompting, personalization, & more!

🧵:

8 25

Reposted by Ben Tappin

LSE Department of Psychological & Behavioural Science @lsepbs.bsky.social · Jul 21

Exciting new research by @benmtappin.bsky.social and colleagues ⬇️

Kobi Hackenburg @kobihackenburg.bsky.social · Jul 21

Today (w/ @ox.ac.uk @stanford @MIT @LSE) we’re sharing the results of the largest AI persuasion experiments to date: 76k participants, 19  LLMs, 707 political issues.

We examine “levers” of AI persuasion: model scale, post-training, prompting, personalization, & more!

🧵:

2 5

Ben Tappin @benmtappin.bsky.social · Jul 21

👇New experiments in which we aimed to map the levers and scope of political persuasion with conversational AI models.

It was a tremendous privilege to lead on this work alongside the brilliant @kobihackenburg.bsky.social. The paper is packed with results and we'd love your comments!

Kobi Hackenburg @kobihackenburg.bsky.social · Jul 21

Today (w/ @ox.ac.uk @stanford @MIT @LSE) we’re sharing the results of the largest AI persuasion experiments to date: 76k participants, 19  LLMs, 707 political issues.

We examine “levers” of AI persuasion: model scale, post-training, prompting, personalization, & more!

🧵:

9

Ben Tappin @benmtappin.bsky.social · Jul 20

Have now read this paper in detail. It’s a tour de force. If you’re interested in the potential impact of AI on election outcomes you should put it by your bedside. Key takeaway: let’s remain alive to, but healthily skeptical of, the possibility of large impacts: knightcolumbia.org/content/dont...

1 7 17

Reposted by Ben Tappin

Gordon Pennycook @gordpennycook.bsky.social · May 29

New paper in PSPB! journals.sagepub.com/doi/10.1177/...

Well, actually, not "new". We first put this paper online way back Dec 2022... in any case, we think it's really cool!

We find that conspiracy believers tend to be overconfident & really don't seem to realize that most disagree with them

14 120 350

Ben Tappin @benmtappin.bsky.social · May 29

You should’ve followed Rose’s example more thoroughly and chucked the paper off the back of a boat…

1 2

Reposted by Ben Tappin

Thom Scott-Phillips @thomscottphillips.bsky.social · May 23

New Substack post

It's very personal: my story of a 20-year academic career, and the many challenges of theoretical and cross-disciplinary work

As I put it in the subtitle: There is a lot of success and a lot of pain here, and no happy ending

thomscottphillips.substack.com/p/happy-in-t...

Happy In Theory

This is the short story of my long, 20 year search for a stable academic home. There is a lot of success and a lot of pain here, and no happy ending.

thomscottphillips.substack.com

18 42 110

Ben Tappin @benmtappin.bsky.social · May 21

Given this discussion you might be interested in this paper where we simulated the return to message pre-testing assuming different values of the correlation (& other params). We also unearth some papers on the survey-field correlation in treatment fx (very thin) www.benmtappin.com/papers/Tappi...

2

Ben Tappin @benmtappin.bsky.social · May 21

Ah yes I see more where you’re coming from now - many thanks for persisting! I agree and think it’s actually distinct from the decay mechanism. I’m most used to thinking about persuasive message treatments like political ads, so wasn’t considering those that remove barriers or reduce cognitive load

2 1

Ben Tappin @benmtappin.bsky.social · May 20

Hence my interest in understanding your mechanism as it hadn’t occurred to me before!

Ben Tappin @benmtappin.bsky.social · May 20

Additional context: I’ve thought about this before and agree the correlation is unlikely to be strong. The mechanisms most plausible to me are a) treatment effect heterogeneity and b) some interventions decay faster than others - not detectable on intentions, which are usually measured immediately.

1

Ben Tappin @benmtappin.bsky.social · May 20

Thanks! I’m still a little unclear (likely my fault). Say we had 10 ads encouraging eating veggie, with some rank order of impact on intention to eat veggie. You claim the rank order is likely different on actually eating veggie because the ads change the cost-benefit setup of intention vs. action?

1

Reposted by Ben Tappin

Ryan Briggs @ryancbriggs.net · May 19

I think the current state of social science research is pretty bad and I wrote something for @asteriskmag.bsky.social about it. asteriskmag.com/issues/10/ca...

Text from an article:

Given the current state of evidence production in the social sciences, I believe that many - perhaps most - attempts to use social scientific evidence to inform policy will not lead to better outcomes.
This is not because of politics or the challenges of scaling small programs. The problem is more immediate. Much of social science research is of poor quality, and sorting the trustworthy work from bad work is difficult, costly, and time-consuming.

2 25 49

Reposted by Ben Tappin

Public Opinion Analytics Lab (POAL) @poalab.bsky.social · May 19

Some snapshots from our event on Friday, where we discussed the future of polling, the challenges ahead, and how AI is shaping survey research. Massive THANK YOU to all panelists for sharing their wisdom with us! Here’s a thread with some key take-aways:

1 8 20

Ben Tappin @benmtappin.bsky.social · May 19

Hi Fabian, sympathetic to your paper + this specific point! Interested to hear more re: your reasoning. Say we have 10 interventions. Their effects are correlated across intention A and action A. In your view is there a key mechanism why the value of this correlation depends on the cost of action A?

1

Reposted by Ben Tappin

Julia M. Rohrer @dingdingpeng.the100.ci · May 11

Thanks to everybody who chimed in!

I arrived at the conclusion that (1) there's a lot of interesting stuff about interactions and (2) the figure I was looking for does not exist.

So, I made it myself! Here's a simple illustration of how to control for confounding in interactions:>

69 280 1.1K

Reposted by Ben Tappin

Kristoffer Magnusson @rpsychologist.com · May 5

🚨New R package! {easymediation}🚨

The *Simplest* and *Most Correct* Way to Do Causal Mediation Analysis

Are you tired of explaining mediation analysis to your colleagues? Just send them this package.

github.com/rpsychologis...

Screenshot showing a troll mediation analysis in R

6 23 95

Reposted by Ben Tappin

Julia M. Rohrer @dingdingpeng.the100.ci · Apr 28

PSA: If you want to compare whether an association varies between groups (i.e., you're interested in an interaction) but you also note that the groups differ in some variable that could "explain away" any interaction -- then you need to interact that variable with the predictor to control for it.

[80] Interaction Effects Need Interaction Controls - Data Colada

In a recent referee report I argued something I have argued in several reports before: if the effect of interest in a regression is an interaction, the control variables addressing possible confounds ...

datacolada.org

6 11 69