Philipp Lorenz-Spreen

Today, social media platforms hold the sole power to study the effects of feed-ranking algorithms. We developed a platform-independent method that reranks participants’ feeds in real time and used this method to conduct a preregistered 10-day field experiment with 1256 participants on X during the 2024 US presidential campaign. Our experiment used a large language model to rerank posts that expressed antidemocratic attitudes and partisan animosity (AAPA). Decreasing or increasing AAPA exposure shifted out-party partisan animosity by more than 2 points on a 100-point feeling thermometer, with no detectable differences across party lines, providing causal evidence that exposure to AAPA content alters affective polarization. This work establishes a method to study feed algorithms without requiring platform cooperation, enabling independent evaluation of ranking interventions in naturalistic settings.

New paper in Science:

In a platform-independent field experiment, we show that reranking content expressing antidemocratic attitudes and partisan animosity in social media feeds alters affective polarization.

🧵

The WHITE HOUSE 45 47 K PRESIDENTIAL ACTIONS RESTORING FREEDOM OF SPEECH AND ENDING FEDERAL CENSORSHIP The White House January 20, 2025

The WHITE HOUSE Offender Hall of Shame A record of the media's false and misleading stories flagged by The White House. Scroll for the Truth. The Washington Post EXPOSED OCBS NEWS EXPOSED EXPOSED i msnbc

How it started / How it’s going

🚨 New working paper 🚨

Can LLMs with reasoning + web search reliably fact-check political claims?

We evaluated 15 models from OpenAI, Google, Meta, and DeepSeek on 6,000+ PolitiFact claims (2007–2024).

Short answer: Not reliably—unless you give them curated evidence.

arxiv.org/abs/2511.18749

ロドプシングアニル酸シクラーゼと改変型 cGMP作動性 K⁺ チャネルを用いた神経抑制型オプトジェネティクス🌟
Optogenetic silencing using rhodopsin-GC and engineered cGMP-gated K+ channels🌟
www.science.org/doi/10.1126/...
@johvierock.bsky.social #OpenAccess

I’m glad to see on platform ranking algorithm work done without platform scientists in the mix.

I think it’s quite telling that, by the only means we have to verify Metas election collaborations results… we can’t confirm their claims of little to no effect.

The lines between top-down and bottom-up control are increasingly blurry. Do you mean enforcement on the content-level or on the platform-level? On the latter the enforcement is a job for politics, on the former the balance between autonomy and paternalism is a normative discussion to have.

Indeed, that is an issue. I imagine citizen-science type of Experiments, where platform designs are continuously tested and evaluated, also by participants. But that technology forces us to make decisions about information- and social environments is something we won’t get completely around?

A tiny minority of highly active users produce the majority of online political content, while most users consume content passively and remain largely silent.

Toxic comments are more likely to be expressed, even though most people disagree with them, creating false norms.
osf.io/preprints/so...

We haven’t really tried alternative Social Media. Just little bits of transparency about from where and how information is reaching us, is used by people immediately. If chosen well, such cues could empower users and reward quality. Some suggestions from some time ago: www.nature.com/articles/s41...

Internet erupts as MAGA influencers exposed for being based in other countries A large number of MAGA social media influencers were exposed over the weekend as being being based in other countries.Podcaster Matt Binder posted screenshots showing that Charlie’s Voice Rising, a po...

Social media could work differently. We teach digital literacy, but ppl often can’t apply those lessons w/in systems that obscure where info comes from & how it gets to them. What might social media look like if users had better signals of information provenance? www.rawstory.com/maga-foreign...

Because parties differed in overall posting activity, the share of misinformation within their communication did not necessarily reflect their contribution to total misinformation (Fig. 2, bottom). This was especially true for BSW: despite the highest misinformation rate, its 17 accounts produced only 1.17% of all posts and just 4.22% of total misinformation. Most misinformation originated from the AfD (36.5%), followed by the CDU (29.6%). The CSU accounted for 9.82% despite only 5.43% of all posts, while all other parties contributed less than 10%. Notably, the SPD generated the largest share of posts overall (23.07%) but only 7.34% of misinformation. Across individual platforms, three key patterns emerged. First, on nearly all platforms the AfD contributed the most misinformation, followed by the CDU. Second, this pattern was particularly pronounced on TikTok, where 54.01% of misinformation originated from the AfD. The party’s overall activity on TikTok was also higher than elsewhere: while its share of posts reached at most about 20% on Facebook, X, and Instagram, it accounted for nearly one - third (30.57%) on TikTok. Third, on Instagram it was not the AfD but the CDU that contributed the most to the overall level of misinformation (41.05%). This corresponded with a comparatively low posting activity by the AfD on Instagram (6.2%)

Interesting preprint from @lorenzspreen.bsky.social's team using LLMs to classify misleading social media posts of German politicians by platform and party belonging.

Link here: osf.io/preprints/so...

Algorithmen: Sind sie schuld am Hass in den Sozialen Medien? Algorithmen sind offenbar schuld an Hass und Polarisierung auf Social Media. Neue Forschung zeigt: Der Fokus auf Algorithmen verschleiert noch größere Probleme.

www.deutschlandfunk.de/polarisierun...

Autism and Vaccines QUESTIONS AND CONCERNS PAGE 2 OF 9 | ALL PAGES For Everyone NOV. 19, 2025 KEY POINTS The claim "vaccines do not cause autism" is not an evidence-based claim because studies have not ruled out the possibility that infant vaccines cause autism. Studies supporting a link have been ignored by health authorities. HHS has launched a comprehensive assessment of the causes of autism, including investigations on plausible biologic mechanisms and potential causal links.

Vaccines do not cause Autism* Pursuant to the Data Quality Act (DQA), which requires federal agencies to ensure the quality, objectivity, utility, and integrity of information they disseminate to the public, this webpage has been updated because the statement "Vaccines do not cause autism" is not an evidence-based claim. Scientific studies have not ruled out the possibility that infant vaccines contribute to the development of autism. However, this statement has historically been disseminated by the CDC and other federal health agencies within HHS to prevent vaccine hesitancy. HHS has launched a comprehensive assessment of the causes of autism, including investigations on plausible biologic mechanisms and potential causal links. This webpage will be updated with gold-standard science that results from the HHS comprehensive assessment of the causes of autism as required by the DQA. The following, as required by the DQA, details the state of the evidence and studies, and the lack thereof, regarding vaccines and autism spectrum disorder (autism) and outlines HHS future research directions to provide answers.

The official CDC website on autism and vaccines is now a screed of cherry-picked data and conspiratorial thinking claiming that actually vaccines *might* cause autism.

This is where we’re at folks…

🔥 NEW PUBLICATION: What is the state of generative AI and disinformation? Aqsa Farooq and I edited an #openaccess, special issue section of the International Journal of Communication, including a scene-setting Introduction article ⬇️

🧵[1/3]

ijoc.org/index.php/ij...

Heute läuft der deutsch-französische #EUSummit25. In der aktuellen geopolitischen Lage braucht es klare Akzente, doch digitale #Souveränität ist weit mehr als Cloud & KI. Im @grimmeinstitut.bsky.social Interview erkläre ich, warum wir digitale #Souveränität und Demokratie zusammendenken müssen:

Great perspective from #CDSM2025: causal inference is 'what-if' analysis. You don’t have to get everything perfect—or how dare you use the c-word. What matters is laying out your assumptions transparently and showing us what happens when they're violated.

Unser #DatensatzdesMonats: Einblicke in die Querdenken-Szene 2021: Einstellungen, Motive und politische Orientierung von Marvin Stein (Socium Universität Bremen).
#Facts: standardisierte, quantitative Online-Befragung auf Telegram im August 2021 mit 5.543 Interviews:

doi.org/10.4232/1.14540

Each dyad (a, b) moves through four discrete states over time, represented by coloured circles. The dyad remains in a given state for a certain duration, or "holding time", before transitioning to a new state according to state-specific transition probabilities, indicated by arrows showing all possible (non-zero) transitions. Paintings by Sofia M. Pereira & Judith von Nordheim.

New paper!

We propose a framework to empirically study animal social relationships by modelling social network (SN) data as time-series—that is, without the need to aggregate them over time.

www.biorxiv.org/content/10.1...

Behavioral data can be very detailed but are usually aggregated and normalized in ways that smother the dynamics. Ben wrote a continuous-time Markov model to improve on this, and also wrote simulations for exploring and validating pipelines. All the code is here: github.com/BenKawam/ASN...

Debate or talking past each other?
In a new preprint, we quantify the arguments that German politicians put forward for and against wealth taxation. Led by @saminenno.bsky.social, with more information in his thread below 👇

How do German MPs debate wealth taxation on social media?

Although wealth in Germany is distributed very unequally and taxation could address this, the topic rarely features in public debate. This was our starting point for examining politicians’ FB, Insta, and X posts from the past nine years 🧵1/5

We’re hiring a Postdoctoral Researcher to help lead my interdisciplinary team at FZI!
Work with us on the automated analysis and countering of #disinformation — at the intersection of AI, computational social science, and #democracy.

👉 Please apply and share! karriere.fzi.de/Vacancies/11...

Check out "Berlin ATProto Meetup" smokesignal.events/did:plc:oals...

⏰ Last chance to register for #CDSM2025!

Don't miss your chance to join us Nov 12–13 for two days of talks & debates at the intersection of causality, data science & AI.

💻 Online | 🎟️ Free
👉 causalscience.org

NEW: In a recent Oxford-led study by @mmosleh.bsky.social, researchers analysed millions of social media posts containing links to news stories, across seven different social platforms.

Read the full research paper here: www.pnas.org/doi/10.1073/...

Thank you @techpolicypress.bsky.social and @ramshajahangir.bsky.social for giving me the space to rant about the “end” of political advertising on social media in the EU!

How prevalent is misinformation really? Many detection methods rely on URL/Source-level, but they miss a big part of posts that don't include links.

In a new preprint from the group, led by @saminenno.bsky.social, we analyze 10x more posts from German politicians on four platforms.

thread below👇

Happy to share our new preprint “Content-based detection of misinformation expands its scope across politicians and platforms.” We analyzed misinformation at the text level in posts by German politicians on Facebook, Instagram, X, and TikTok.
osf.io/preprints/so... 🧵1/8

It is happening 👇

Selective Causal Focus: Research produced and funded by tech companies often either frames problems as user-driven, or solutions as the obligation of users (E.g. community notes). Distracting us from their design, business model, interface, and other causes steering us away from their profit model