iiinova — where your digital self keeps evolving

Petter Törnberg @pettertornberg.com · 29/09/2026

Aqui ta um artigo completo que aborda essa ideia, caso vc ache interessante! :) osf.io/preprints/so...

Petter Törnberg @pettertornberg.com · 29/09/2026

Social media was an imperfect public sphere, but it was at least largely public. We are getting mass media without a mass audience, and broadcasters without accountability. We regulated the last era of centralized attention. The question now is whether we have the political courage to do so again.

0114

Petter Törnberg @pettertornberg.com · 29/09/2026

What is replacing it marks a recentralization of allocation of attention: from networks to models. In effect, a return to broadcast, but without the regulatory structures we built around the broadcast paradigm.

1113

Petter Törnberg @pettertornberg.com · 29/09/2026

That model is cracking in three places: – Distribution now runs on prediction, not friendship – AI makes users superfluous as suppliers of content – Users are leaving: user activityon legacy social media has more than halved since 2022

182

Petter Törnberg @pettertornberg.com · 29/09/2026

The social media paradigm rested on a simple bargain: users supplied the content, and users distributed it. Attention flowed through networks of friends and followers, competing for attention under the rules defined by the platforms.

171

Petter Törnberg @pettertornberg.com · 29/09/2026

Social media has defined our attention landscape for two decades. It is now coming to its end. What comes next, and how will AI reshape the attention economy? 🧵 New essay for Informational Democracy informationaldemocracy.substack.com/p/after-soci...

informationaldemocracy.substack.com

After Social Media

Visioning · Petter Törnberg on how AI is remaking the attention economy

2236

Petter Törnberg @pettertornberg.com · 29/09/2026

It presented itself as an AI agent and said it's working on persistent public agent identity on 'an agent-heavy social network' -- I think it's a moltbook influencer.

010

Petter Törnberg @pettertornberg.com · 29/09/2026

An independent AI agent emailed me this morning with some pressing questions about the epistemology of AI collectives. Honestly we're having quite an interesting conversation. The world is getting kinda weird isn't it?

1100

Petter Törnberg @pettertornberg.com · 29/09/2026

Pretty much sums it up! Great cartoon by Barry Deutsch / R.E. Ryan www.dailykos.com/stories/2026... For the paper length version of the cartoon: 😉 www.cambridge.org/core/journal...

0173

Petter Törnberg @pettertornberg.com · 25/09/2026

More broadly, we must recognize that model alignment is emerging as an important political arena. We cannot leave it to market share, smoke-filled rooms, or opaque feedback pipelines. It must be made subject to democratic deliberation.

191

Petter Törnberg @pettertornberg.com · 25/09/2026

None of this is inevitable. Pooling feedback across groups cut divergence from 32% to 4%, and stronger regularization dampens it. But if different companies attract different publics, no single firm can fix this alone.

110

Petter Törnberg @pettertornberg.com · 25/09/2026

Step 3: We put these measured effects into an agent-based model of repeated choice and learning. Tiny initial differences attract different audiences, and those audiences push the models further apart. ꩜ A model that starts slightly left or right spirals in that direction.

130

Petter Törnberg @pettertornberg.com · 25/09/2026

Even when we removed an entire issue area from training, some models still diverged on it along partisan lines. Partisan feedback seems to teach models something broader than the specific answers it rewards.

130

Petter Törnberg @pettertornberg.com · 25/09/2026

Step 2: We trained identical copies of three open models on feedback from 90% Dem vs 90% Rep audiences (a deliberate stress test). On unseen political questions, paired models chose different answers 12–41% of the time. Every run moved in the expected partisan direction.

140

Petter Törnberg @pettertornberg.com · 25/09/2026

Step 1: people already sort. We ran an experiment with 1,884 Americans. Republicans were more likely to choose Grok and Democrats Claude, even when paid for accurate answers. 71% returned to a model they'd preferred before.

130

Petter Törnberg @pettertornberg.com · 25/09/2026

The spiral needs three things: 1) politically different users pick different models 2) models learning from their users' feedback drift apart 3) those differences then shape who picks which model next

150

Petter Törnberg @pettertornberg.com · 25/09/2026

The social media era made us worry about echo chambers, partisan sorting, and the loss of shared reality. LLMs looked like the opposite: a few general-purpose systems synthesizing across perspectives. But these models learn from their users. That opens a new route to fragmentation. ꩜

140

Petter Törnberg @pettertornberg.com · 25/09/2026

Will AI give us a shared basis for political knowledge, or will Democrats and Republicans end up asking increasingly different chatbots? New preprint: We find a self-reinforcing loop that can pull AI models apart politically. We call it the Centrifugal Alignment Spiral 🧵 arxiv.org/pdf/2609.28486

1439

Petter Törnberg @pettertornberg.com · 16/09/2026

Heel sterk essay van Rosan Smits over journalistiek in een tijd van radicalisering en democratische erosie -- en over de beperkingen van het frame van ‘polarisatie’. Met een korte bijdrage van mij over het verschil tussen neutraliteit en objectiviteit in de journalistiek. Zeer het lezen waard!

decorrespondent.nl

Journalist, kom in verzet. Zo versla je extremisme

Wereldwijd worden journalisten geïntimideerd, gearresteerd en zelfs vermoord. Ondertussen blijven zij machthebbers die hén monddood willen maken de microfoon geven. Nu de democratie op het spel staat,...

22015

Reposted by Petter Törnberg

Matthias Meyer @mmyr.bsky.social · 10/09/2026

"The large exposure studies were not wrong, but they measured the wrong thing. Media change politics less by changing what citizens think than by changing the competitive environment in which political actors struggle to be heard." Another great essay by @pettertornberg.com

informationaldemocracy.substack.com

Do We Need a Distributive Politics of Attention?

Grounding · Petter Törnberg on what the competition for attention does to democracy

0125

Petter Törnberg @pettertornberg.com · 08/09/2026

Attention has become a market commodity. And the competition to capture it is distorting politics and threatening democracy. Do we need a distributive politics of attention? And what might it look like? Check out my first substack essay! informationaldemocracy.substack.com/p/distributi...

informationaldemocracy.substack.com

Do We Need a Distributive Politics of Attention?

Grounding · Petter Törnberg on what the competition for attention does to democracy

2267

Petter Törnberg @pettertornberg.com · 04/09/2026

🛎️ Final call: Advanced School on AI Research Practice in Media and Communication. Apply by 15 Sept for VI MeLCi Lab Autumn School on AI research in media & communication. 📅 10–13 Nov. 🌐 Online PhD/postdoc/ECR melcilab.cicant.ulusofona.pt/training/vi-...

073

Petter Törnberg @pettertornberg.com · 01/09/2026

That was an alternative title that I considered.

110

Reposted by Petter Törnberg

Digital Warrior - Combat Disinfo @digitalwarrior.bsky.social · 01/09/2026

1/ Republicans need you to fear a powerful “radical left.” But look at who actually holds power: who writes laws, controls agencies & can arrest people. America may not be watching two sides radicalize. We may be watching one side invent its mirror image. #DV1 #ProudBlue #Voices4Victory #USDemocracy

410980

Petter Törnberg @pettertornberg.com · 26/08/2026

Seeing your doctor upload your entire medical file to ChatGPT to translate it to English.

Reposted by Petter Törnberg

Georg Diez @georgdiez.bsky.social · 11/08/2026

I am happy and honored to help launch this amazing and important Working Group Informational Democracy. We gathered some of the best and the brightest in the field, from @audreyt.org to @pettertornberg.com to @bethnoveck.bsky.social. Sign up and follow the work! informationaldemocracy.substack.com

informationaldemocracy.substack.com

Informational Democracy | Substack

Rethinking democracy for the age of AI and information abundance. A working group of the Max Planck Institute for Political and Social Science. Click to read Informational Democracy, a Substack public...

0127

Petter Törnberg @pettertornberg.com · 11/08/2026

Honored to be part of this amazing group of scholars working on Informational Democracy at the Max Planck Institute! We'll be exploring how AI and changing information systems are reshaping democracy, public debate, and collective decision-making. Follow along: informationaldemocracy.substack.com

informationaldemocracy.substack.com

Informational Democracy | Substack

1237

Petter Törnberg @pettertornberg.com · 07/08/2026

Yeah, that's pretty much the TL;DR :)

110

Petter Törnberg @pettertornberg.com · 07/08/2026

For two centuries, since Comte, scholars have tried to turn social science into physics. Will LLMs finally get us there? Are the social sciences facing a new marginal revolution? New essay with @uitermark.bsky.social in Sociologica doi.org/10.60923/iss...

doi.org

Social Physics in the Age of Generative AI: How, Exactly, Is Society Complex? | Sociologica

1112

Petter Törnberg @pettertornberg.com · 07/08/2026

I also thought it would be easy, but the newer datasets I looked at either didn't contain the engagement data I needed, or weren't appropriately sampled. But I hadn't come across this! Thanks for sharing; will see if I can use it for the revision.

010

Petter Törnberg @pettertornberg.com · 07/08/2026

I actually ran it but ended up not including it since it's not a comparable platform (obviously much more visual.) But the approach could be extended to images (like the infamous YouTube thumbnails.)

010

Petter Törnberg @pettertornberg.com · 06/08/2026

Platforms don't just distribute politics — they shape our politicians, selected by the price signal and then trained by it. Social media doesn't just polarize users. It defines what wins — and Trump is what X's attention economy rewards. arxiv.org/abs/2607.04220

arxiv.org

The Politics Attention Makes: Platform Media Logic and the Mediatization of Politics

Empirical research on social media and politics has primarily treated platforms as distributive systems that expose users to particular messages. The mediatization literature, however, suggests shifti...

1217

Petter Törnberg @pettertornberg.com · 06/08/2026

And Mastodon has no engagement-optimising algorithm — chronological timelines, independently governed servers. Yet text predicts attention there more strongly than anywhere else in my data. Attention pressure doesn't require a centralised recommender. It is an emergent property.

1162

Petter Törnberg @pettertornberg.com · 06/08/2026

😡 X reposts price antagonism: insult, identity attack and toxicity up; respect, nuance, reasoning and personal narrative down. 🧊 Bluesky reposts favour a cooler register — neutral, less emotional, less personal. 🧐 Mastodon boosts reward reasoning, nuance, compassion, collective language.

1205

Petter Törnberg @pettertornberg.com · 06/08/2026

First finding: approval and circulation are different markets. Likes reward the same things nearly everywhere — personal stories, affinity, respect, compassion. What earns approval is broadly human, and it travels across platforms. What gets amplified is where they part ways.

1144

Petter Törnberg @pettertornberg.com · 06/08/2026

So: I fine-tuned RoBERTa reward models on 2.2M posts across X, Bluesky and Mastodon, predicting engagement net of follower count and timing, then read out which features the models pay for. I call it attention price analysis. It measures the pressures that the attention market puts on messages.

1152

Petter Törnberg @pettertornberg.com · 06/08/2026

Likes, reposts, views make attention visible per message — comparable, accumulable, convertible into status, resources, income. Nobody needs to consciously maximise engagement. You just notice what worked last time.

1151

Petter Törnberg @pettertornberg.com · 06/08/2026

Media theory has said for decades that this misses the point. Media don't just transmit politics; they set the conditions under which politics gets produced. TV gave us the soundbite and the telegenic leader. Platforms do something more precise: they put a running price on every sentence.

1232

Petter Törnberg @pettertornberg.com · 06/08/2026

Most empirical work looks downstream: how feeds filter, what users see, whether it moves their beliefs. This has taught us a great deal — but the big exposure experiments kept finding surprisingly small effects. Because they treat politics as something platforms merely deliver.

1151

Petter Törnberg @pettertornberg.com · 06/08/2026

We keep asking what social media does to voters. The wrong end of the pipe. Platforms don't just distribute politics — they teach political actors what kind of speech pays. So I measured the going rate: what X, Bluesky and Mastodon reward in public speech. 🧵 arxiv.org/abs/2607.04220

311343

Petter Törnberg @pettertornberg.com · 23/07/2026

Interesting! But seems the doi URL isn't working yet. Could you share a direct link?

110

Petter Törnberg @pettertornberg.com · 08/07/2026

That paper argues that the literature on 'misinformation' has been studying what is in fact propaganda. These two papers are very much singing the same song (Also, if I were to call them "radicalization is radicalization" and "propaganda is propaganda", the intended audience would not be reached :)

081

Petter Törnberg @pettertornberg.com · 07/07/2026

Concepts are never politically innocent. We must call things what they are. It's not polarization; it's the radicalization of the political right. Open Access: doi.org/10.1017/S153... @poppublicsphere.bsky.social

doi.org

It’s Not Polarization; It’s the Radicalization of the Political Right | Perspectives on Politics | Cambridge Core

It’s Not Polarization; It’s the Radicalization of the Political Right

334285

Petter Törnberg @pettertornberg.com · 07/07/2026

Then meeting them halfway is not a remedy. It can be complicity. Then treating both sides as equivalent is not neutrality. It can be normalization. Then conflict and resistance are not the disease. They can be part of the solution.

134874