The onset of the Trump presidency has accelerated a trend that cuts through years of debate about propaganda, influence operations, and public diplomacy. State power moves through culture differently than it used to.

The structural logic of what authoritarian states call an influence operation and what democratic governments call a communications strategy is, at its core, identical. What differs isn’t the mechanism. It’s the institutional wrapper, the degree of disclosure, and the normative framework we apply when we assess it.

From Chinamaxxing to Dubai’s Influencer Playbook

The “becoming Chinese” trend is gaining traction among Western creators and has been around since 2025. IShowSpeed is credited as having started it with his visit to China, but there are others - Hasan Piker, The Nelk Boys, Pokimane. While critics describe this meme as fetishizing Chinese culture for a convenient trend, it hasn’t prevented “chinamaxxing” from trending in the US and now extending globally.

Due to its timing, this meme trend has turned into a soft power win for China emerging as a direct repsonse to the soft power vacuum left by American retreat from cultural diplomacy.

While IshowSpeed trip to China was self sponsored, perhaps no case illustrates the maturity of state-influencer infrastructure better than what happened in Dubai just last week. As Iranian strikes hit the region, influencers from different countries with millions of followers began posting nearly identical content within hours reaffirming Dubai as the safest place in the world.

The speed and coordination suggested something beyond organic response.

When State Campaigns Become Viral Culture

Here’s where it gets more complex. Nobody knows for certain whether these campaigns were initially state-sponsored or not. But at least in Dubai, the initial wave appears coordinated, suggesting institutional backing. But as in the case for chinamaxxing, sometimes memes grow into their own viral trends that may be boosted by states, but take on a life of their own.

Chinamaxxing is now a genuine cultural phenomenon. The Dubai influencers’ “who keeps us safe” framing also became a meme picked up globally, including by Germany’s Auswärtiges Amt (Foreign Office). The content was adapted, transformed, recontextualized. But the core framing traveled far beyond its origins becoming a self-perpetuating cultural moment that outlived and transcended their potential institutional origins.

The question I keep asking myself is, if these forms of credbile influence campaigns / trends already constitutes a form of propaganda. Reimagined as ambient culture, these types of trends are aesthetically fluent and structurally indistinguishable from organic content.

AI Vdeo and Nostalia as Tools of State Persuasion

But there is more. Governments have added AI to their playbook via deepfakes filtered through nostalgia and meme culture. In the context of the Iran War both parties, Iran and the US / Israel, are producing AI-generated videos using familiar framing like SpongeBob or Lego to frame their perspective in the current conflict, which consists of a lot of meme-driven imaging around military strikes, bundling clips from war movies, or video games with actual strike footage . These messages delivered through official channels but amplified through social platforms using multiple vectors, e.g. organic engagement, creator partnerships, memetic warfare. Each reinforcing the others until origin becomes impossible to trace.

Much of this is direct state production of AI-generated content. The influencer ecosystem represents another delivery system. Both exploit the same core mechanism - cultural familiarity as a shortcut to persuasion. While this is definetly a form of propaganda, the genius behind this uniquely powerful form is timing. It lands as entertainment before audiences process it as persuasion. By invoking beloved cultural references from childhood media and gaming culture, states across regime types use the same structural playbook, which is manufacturing familiar-feeling content that bypasses critical thinking through emotional resonance.

A SpongeBob meme doesn’t feel like state messaging. A Lego-format explainer doesn’t trigger propaganda defenses. The nostalgia creates comfort and familiarity trust.

When Digital Diplomacy Becomes Indistinguishable from Influence Operations

Traditional digital diplomacy has evolved. What once operated through clearly marked channels with identifiable state actors has dissolved into something harder to categorize.

The questions I ask myself are

if democratic and authoritarian states use identical structural logic, how do we maintain meaningful distinctions?

What distinguishes legitimate public diplomacy from engineered cultural influence at scale, since traditional markers don’t hold anymore?

Consider the White House meme-heavy approach to communicating. Compare it to Dubai’s coordinated influencer messaging or the organic soft power play that “becoming Chinese” represents. The aesthetic differs. The disclosure framework differs. But the structural logic of using trusted cultural voices and familiar references to bypass critical evaluation? Identical.

The tools of digital influence operate the same way regardless of who deploys them. Memes attached to a military strike or explaining geopolitical strategy exploit the cognitive shortcut of cultural familiarity lowering psychological defenses, which makes it so powerful.

Is this actually working?

The evidence is mixed and troubling.. This scattershot approach, heavy on memes and cultural references, light on traditional wartime presidential communication, reveals something important. The goal isn’t to build broad-based support through careful explanation of objectives and coalition-building. It’s to maintain intensity through constant cultural engagement.

The nostalgia-driven, meme-heavy approach may be building enthusiasm within specific demographics. It creates instant resonance with audiences who share those cultural touchstones and leads to engagement, shares, viral moments.They generate engagement, shares, viral moments. But they may also create ceilings, where the very familiarity that draws some audiences in alienates others who don’t share the reference points or who resent seeing childhood entertainment weaponized for political purposes. Ben Stiller's negative response to having Tropic Thunder clips used in White House war messaging captures this backlash.

Yet even this resistance becomes content. The controversy generates more engagement, more visibility, more cultural penetration of the original messaging. Whether people approve or object, the meme has done its work: making war feel like familiar entertainment rather than distant policy.

Defining Digital Diplomacy 2.0

Traditional public diplomacy operated with clear rules. Cultural exchange programs, international broadcasting, educational initiatives. All clearly marked as state-sponsored. All operating on the assumption that persuasion should happen through information and argument, not emotional manipulation.

The influencer ecosystem broke those rules. Credible influence operates in gray zones where disclosure is partial, where organic and sponsored blend, where state messaging flows through trusted voices without clear attribution.

AI-generated nostalgia content accelerates this breakdown as cutlural referenced optimized for emotional resonance contribute to collpase of the distinction between public diplomacy and propaganda. It's simply too effective, too scalable, too aligned with how attention economies actually function.

The way forward is to define urgent unresolved questions that are emerging due to this evolution and adapt tools to increase resilience from this form of propaganda.

(Automated German translation of this post is below the English version)

Something struck me while reading two separate pieces last week. The first was a Financial Times analysis declaring that September 2025 marks the moment “social media jumped the shark.” The second was a joint warning from German Chancellor Friedrich Merz and French President Emmanuel Macron about democracy’s “degeneration” – driven partly by platforms “controlled by big American entrepreneurs and Chinese firms whose interests are not at all the survival and the good functioning of our democracy.”

Social media peaked in 2022, the same year OpenAI launched ChatGPT and democratized access to AI. Both developments are connected, let me explain…

The Peak: What the Data Shows

John Burn-Murdoch’s Financial Times analysis, drawing on data from 250,000 adults across 50+ countries, reveals a striking pattern:

• Social media usage peaked in 2022 and has declined nearly 10% since

• Adults now spend an average of 2 hours 20 minutes daily on platforms, down from previous highs

• The decline is sharpest among teens and 20-somethings – the heaviest former users

• Platform use has shifted from connection to “mindless browsing” – filling spare time rather than engaging meaningfully

• Shares of people using platforms to stay in touch, express themselves, or meet new people have fallen over 25% since 2014

One exception stands out: North America, where consumption continues climbing – reaching levels 15% higher than Europe by 2024.

But this isn’t simply about people using social media less. It’s about what they’re migrating toward, and what’s being built in the space they’re leaving behind.

The Current State: Three Observable Shifts

AI Slop Replacing Organic Content

Burn-Murdoch describes the emerging content landscape as “ultra-processed”, meaning content is “dopamine-dense, with at best negligible informational value, at worst corrosively negative.” The nutritional analogy is on point here.

Posts by casual posters are disappearing into algorithmic voids because of the AI content pollution designed to overwhelm and harvest attention flooding my for you page and disrupting my carefully curated feed slowly but surely.

The Migration, Not Reduction

Here’s what the usage decline obscures: screen time isn’t decreasing. It’s also relocating.

We’re witnessing a migration pattern similar to how trust shifted from institutions to individuals / alternative trusted sources (influencers). Now, as people crave connection but find platforms increasingly hostile to genuine human interaction, a newer trend emerges: migration toward AI companions.

The data on this phenomenon is still emerging, but the direction is clear as we’re not seeing reduced engagement with digital systems, but we’re seeing a shift from human-mediated connection to AI-mediated companionship. This connects directly to the “seemingly conscious AI” problem I’ve written about previously: systems designed to create emotional dependency rather than enhance human capability.

Whether this migration persists or represents a transitional phase remains uncertain. But its emergence at precisely the moment social platforms peak suggests these dynamics are related.

Generational Rejection

Perhaps most telling is the generational pattern. Younger users – those who should be social media’s core demographic – are leading the retreat.

Unlike Millennials who adopted “big social” platforms wholesale, younger generations seem to sense what observers call “enshittification” – the gradual degradation of platforms through prioritization of business metrics over user experience. They never fully committed to Facebook, Instagram, or even TikTok in the way previous generations did.

As one commenter observed: “It’s almost as if the people who think an all-AI social platform will be successful in perpetuity never heard of Vine or Ello.” Platforms rise and fall. The question is whether this represents a temporary shift or a permanent restructuring of digital social behavior.

The Fragmentation Accelerates

The decline in overall usage coincides with accelerating fragmentation of the social media ecosystem. Both dynamics can occur simultaneously.

We can trace the splintering visually: During COVID, TikTok joined the old guard of major platforms. In 2020, political fragmentation drove users rightward to alternative platforms. After Musk’s Twitter acquisition in 2022, fragmentation accelerated leftward. Now we’re seeing fragmentation along technological lines with the emergence of AI-only platforms like Sora2 (OpenAI) and Vibes (Meta) due to the constant need for data streams to train their AI systems.

This isn’t paradoxical. As overall engagement on big social media platforms declines, users cluster in increasingly specialized environments. If this trend persists, the social media ecosystem will develop towards a landscape without dominant platforms – a fundamental shift from social media’s previous era of consolidation around a few major players.

The Democratic Implications: Structural Vulnerability

This is where social media’s transformation intersects with democratic crisis. The changes aren’t merely about how people spend leisure time, but about the restructuring of the information space democracy requires. There are currently three levels of transformation occurring simultaneously:

Content-level transformation: The information space is flooding with AI propaganda and AI slop. What once required significant resources e.g., targeted disinformation campaigns, personalized propaganda now happens on an industrial scale. A single operator can generate thousands of tailored messages far overwhelming human-created content.

Also, when the line between human and AI-generated content disappears, trust in the information environment collapses. If everything might be synthetic, why trust anything? This creates what researchers call “epistemic learned helplessness” – the inability to distinguish reliable from unreliable information.

These vulnerabilities don’t operate independently. They compound. Fragmentation makes populations more susceptible to emotional manipulation. Emotional manipulation accelerates fragmentation. Synthetic content undermines the shared reality needed to counter both.

At scale, AI-generated content optimized for emotional impact overwhelms reasoned discourse. When combined with sophisticated platform expertise – understanding of algorithmic dynamics, timing, network effects – these vulnerabilities become genuinely dangerous to democratic discourse.

Context-level transformation: Hyperpersonalization systems like ChatGPT Pulse and BBC News AI represent a qualitatively different challenge. These systems don’t just filter existing content – they generate personalized information environments for each user. Pulse delivers “proactive personalized daily updates” based on conversation history. BBC’s AI unit aims to “personalize news” for individual consumption.

When each user receives algorithmically customized information, shared reality erodes. There’s no common basis for democratic deliberation if citizens literally see different facts, different contexts, different framings of every issue.

Algorithmic-level transformation: Some platforms seem to be manipulating content visibility for political purposes, but at the same time professing balance and neutrality. We see this most clearly with X, where algorithmic changes appear to favor specific political perspectives. In other words, it seems likely that algorithms have been altered to favor the distribution of prominence of certain information. Similar concerns are emerging around the future of TikTok USA. This assumption is based on the premise that there was a more balanced data flow.

The questions that need to be asked at this point should focus on who we want in control of AI making these personalization decisions and who ensures neutrality and fairness?

Why Infrastructure Changes Everything

This isn’t just about content moderation or platform policies anymore. Everything about the social media ecosystem is transforming from the content to the context to the infrastructure.

As I argued in my recent keynote at the Fraunhofer Institute for Systems and Innovation Research ISI on AI and democracy: who controls AI infrastructure ultimately controls democratic communication. Technological dependence creates democratic dependence.

When the basic systems for information discovery, curation, and delivery are controlled by entities with no stake in democratic survival, democratic self-determination becomes structurally compromised.

This is the core of Merz and Macron’s warning. Europe has been “guilty of handing over our public democratic space to social networks owned by big American entrepreneurs and Chinese firms.” These platforms don’t just host conversation – they shape it through algorithmic infrastructure that determines what gets seen, what gets amplified, what gets memory-holed.

Well-meaning regulation – like Europe’s various framework attempts – remains toothless without technological sovereignty. You cannot regulate what you cannot control at the infrastructure level.

We are Unraveling

We’re past the hypothetical stage. AI’s influence on democratic discourse isn’t a future concern – it’s the present reality.

The question is how fragmented societies receiving hyperpersonalized, AI-mediated information through infrastructure controlled by foreign corporations can sustain the shared reality democratic discourse requires.

The emerging trends and dynamics at the core of the changes we are seeing are eroding the information space democracy depends on.

Macron and Merz are warning about democratic degeneration. Burn-Murdoch is documenting social media’s decline. These are not separate phenomena. They feel like different perspectives on the same unraveling – the restructuring or even collapse of the social media ecosystem that made both social connection and democratic deliberation possible.

Whether this transformation can be redirected, or whether we’re witnessing the irreversible decline of digital public space as we’ve known it, remains the defining question of our political moment.

DEUTSCHE ÜBERSETZUNG:

Der Höhepunkt der sozialen Medien und der Niedergang der Demokratie im Zeitalter der KI – eine große Entwirrung.

Letzte Woche las ich zwei Artikel. Der erste war eine Analyse der Financial Times, in der erklärt wurde, dass im September 2025 der Moment gekommen sei, in dem „die sozialen Medien ihren Höhepunkt überschritten haben“. Der zweite war eine gemeinsame Warnung des deutschen Bundeskanzlers Friedrich Merz und des französischen Präsidenten Emmanuel Macron vor dem „Verfall“ der Demokratie – teilweise verursacht durch Plattformen, die „von großen amerikanischen Unternehmern und chinesischen Firmen kontrolliert werden, deren Interessen keineswegs das Überleben und das gute Funktionieren unserer Demokratie sind“.

Soziale Medien erreichten 2022 ihren Höhepunkt, im selben Jahr, in dem OpenAI ChatGPT auf den Markt brachte und den Zugang zu KI demokratisierte. Ich will versuchen skizzieren inwiefern diese zusammenhängen…

Der Höhepunkt: Was die Daten zeigen

Die Analyse von John Burn-Murdoch in der Financial Times, die sich auf Daten von 250.000 Erwachsenen aus über 50 Ländern stützt, zeigt ein auffälliges Muster:

Die Nutzung sozialer Medien erreichte 2022 ihren Höhepunkt und ist seitdem um fast 10 % zurückgegangen.

Erwachsene verbringen heute durchschnittlich 2 Stunden und 20 Minuten täglich auf Plattformen, was einen Rückgang gegenüber früheren Höchstständen bedeutet.

Der Rückgang ist bei Teenagern und Twens – den ehemals intensivsten Nutzern – am stärksten.

Die Nutzung der Plattformen hat sich von der Kontaktaufnahme zum „sinnlosen Surfen” verlagert – sie füllt eher die Freizeit aus, als dass sie zu einer sinnvollen Beschäftigung führt.

Der Anteil der Menschen, die Plattformen nutzen, um in Kontakt zu bleiben, sich auszudrücken oder neue Leute kennenzulernen, ist seit 2014 um über 25 % zurückgegangen.

Eine Ausnahme sticht hervor: Nordamerika, wo der Konsum weiter steigt und bis 2024 ein Niveau erreichen wird, das 15 % über dem Europas liegt.

Aber es geht hier nicht einfach darum, dass die Menschen weniger soziale Medien nutzen. Es geht darum, wohin sie abwandern und was an der Stelle entsteht, die sie hinterlassen.

Der aktuelle Stand: Drei beobachtbare Veränderungen

KI-Slop ersetzt organische Inhalte

Burn-Murdoch beschreibt die sich abzeichnende Content-Landschaft als „ultra-processed”. Er beschreibt Content als „dopaminreich, bestenfalls einen vernachlässigbaren Informationswert haben und schlimmstenfalls schädlich negativ sind”. Die Analogie zur Ernährung ist hier sehr treffend.

Beiträge von Gelegenheitspostern verschwinden in algorithmischen Leerstellen, weil die KI-Inhalte, die darauf ausgelegt sind, Aufmerksamkeit zu erregen und zu sammeln, meine „Für dich”-Seite überfluten und zumindest meinen sorgfältig kuratierten Feed langsam aber sicher zerstören.

Die Migration, nicht die Reduzierung

Was der Rückgang der Nutzung verschleiert, ist Folgendes: Die Bildschirmzeit nimmt nicht ab. Sie verlagert sich auch.

Wir beobachten ein Migrationsmuster, das dem Vertrauenswandel von Institutionen zu Einzelpersonen/alternativen vertrauenswürdigen Quellen (Influencern) ähnelt. Da die Menschen sich nach Verbindung sehnen, aber Plattformen zunehmend als feindlich gegenüber echten menschlichen Interaktionen empfinden, entsteht ein neuer Trend: die Migration hin zu KI-Begleitern.

Die Daten zu diesem Phänomen sind noch nicht vollständig, aber die Richtung ist klar, da wir keinen Rückgang der Nutzung digitaler Systeme beobachten, sondern eine Verlagerung von zwischenmenschlichen Beziehungen hin zu KI-gestützten Begleitern. Dies steht in direktem Zusammenhang mit dem Problem der „Seemingly Conscious AI”, über das ich bereits zuvor geschrieben habe: Systeme, die darauf ausgelegt sind, emotionale Abhängigkeit zu erzeugen, anstatt die menschlichen Fähigkeiten zu verbessern.

Ob diese Abwanderung von Dauer ist oder nur eine Übergangsphase darstellt, bleibt ungewiss. Aber dass sie genau zu dem Zeitpunkt einsetzt, zu dem soziale Plattformen ihren Höhepunkt erreichen, lässt vermuten, dass diese Entwicklungen miteinander zusammenhängen.

Generationsbedingte Ablehnung

Am aussagekräftigsten ist vielleicht das generationsbedingte Muster. Jüngere Nutzer – die eigentlich die Kernzielgruppe der sozialen Medien sein sollten – führen den Rückzug an.

Im Gegensatz zu den Millennials, die „große soziale“ Plattformen vorbehaltlos angenommen haben, scheinen jüngere Generationen das zu spüren, was Beobachter als „Enshittification“ bezeichnen – die allmähliche Verschlechterung von Plattformen durch die Priorisierung von Geschäftskennzahlen gegenüber der Benutzererfahrung. Sie haben sich nie so sehr für Facebook, Instagram oder sogar TikTok engagiert wie frühere Generationen.

Es scheint fast so, als hätten die Leute, die glauben, dass eine vollständig KI-basierte soziale Plattform auf Dauer erfolgreich sein wird, noch nie von Vine oder Ello gehört. Plattformen kommen und gehen. Die Frage ist, ob dies eine vorübergehende Veränderung oder eine dauerhafte Umstrukturierung des digitalen Sozialverhaltens darstellt.

Die Fragmentierung beschleunigt sich

Der Rückgang der Gesamtnutzung fällt mit einer beschleunigten Fragmentierung des Social-Media-Ökosystems zusammen.

Wir können die Zersplitterung visuell nachvollziehen: Während COVID schloss sich TikTok der “Old Guard” unter den großen Plattformen an. Im Jahr 2020 trieb die politische Fragmentierung die Nutzer nach rechts zu alternativen Plattformen. Nach Musks Übernahme von Twitter im Jahr 2022 beschleunigte sich die Fragmentierung nach links. Derzeit beobachten wir eine Fragmentierung entlang technologischer Linien mit dem Aufkommen von reinen KI-Plattformen wie Sora2 (OpenAI) und Vibes (Meta), da für das Training ihrer KI-Systeme ein ständiger Bedarf an Datenströmen besteht.

Da das allgemeine Engagement auf großen Social-Media-Plattformen zurückgeht, gruppieren sich die Nutzer in immer spezialisierteren Umgebungen. Wenn sich dieser Trend fortsetzt, wird sich das Social-Media-Ökosystem zu einer Landschaft ohne dominante Plattformen entwickeln – eine grundlegende Veränderung gegenüber der bisherigen Ära der Konsolidierung rund um einige wenige große Akteure.

Die demokratischen Implikationen: strukturelle Verwundbarkeit

Hier überschneidet sich der Wandel der sozialen Medien mit der demokratischen Krise. Die Veränderungen betreffen nicht nur die Art und Weise, wie Menschen ihre Freizeit verbringen, sondern auch die Umstrukturierung des Informationsraums, den die Demokratie benötigt. Derzeit finden drei Ebenen der Transformation gleichzeitig statt:

Transformation auf Inhaltsebene: Der Informationsraum wird mit KI-Propaganda und AI Slop überschwemmt. Was früher erhebliche Ressourcen erforderte, z. B. gezielte Desinformationskampagnen und personalisierte Propaganda, findet heute in industriellem Maßstab statt. Ein einzelner Betreiber kann Tausende von maßgeschneiderten Nachrichten generieren, die die von Menschen erstellten Inhalte bei weitem übertreffen.

Wenn zudem die Grenze zwischen menschlichen und KI-generierten Inhalten verschwindet, bricht das Vertrauen in die Informationsumgebung zusammen. Wenn alles synthetisch sein könnte, warum sollte man dann irgendetwas glauben? Dies führt zu dem, was Forscher als „epistemische erlernte Hilflosigkeit” bezeichnen – die Unfähigkeit, zuverlässige von unzuverlässigen Informationen zu unterscheiden.

Diese Schwachstellen wirken nicht unabhängig voneinander. Sie verstärken sich gegenseitig. Fragmentierung macht Bevölkerungsgruppen anfälliger für emotionale Manipulation. Emotionale Manipulation beschleunigt die Fragmentierung. Synthetische Inhalte untergraben die gemeinsame Realität, die notwendig ist, um beiden entgegenzuwirken.

In großem Maßstab überwältigen KI-generierte Inhalte, die auf emotionale Wirkung optimiert sind, den vernünftigen Diskurs. In Kombination mit ausgefeilter Plattformkompetenz – Verständnis für algorithmische Dynamiken, Timing, Netzwerkeffekte – werden diese Schwachstellen zu einer echten Gefahr für den demokratischen Diskurs.

Transformation auf Kontextebene: Hyperpersonalisierungssysteme wie ChatGPT Pulse und BBC News AI stellen eine qualitativ andere Herausforderung dar. Diese Systeme filtern nicht nur vorhandene Inhalte, sondern generieren personalisierte Informationsumgebungen für jeden Nutzer. Pulse liefert „proaktive personalisierte tägliche Updates” auf der Grundlage des Gesprächsverlaufs. Die KI-Abteilung der BBC hat sich zum Ziel gesetzt, „Nachrichten für den individuellen Konsum zu personalisieren”.

Wenn jeder Nutzer algorithmisch angepasste Informationen erhält, bröckelt die gemeinsame Realität. Es gibt keine gemeinsame Grundlage für demokratische Debatten, wenn die Bürger buchstäblich unterschiedliche Fakten, unterschiedliche Kontexte und unterschiedliche Darstellungen jedes Themas sehen.

Transformation auf algorithmischer Ebene: Zudem kommt auch, dass einige Plattformen die Sichtbarkeit von Inhalten für politische Zwecke zu manipulieren scheinen. Am deutlichsten wird dies bei X, wo algorithmische Änderungen bestimmte politische Perspektiven zu begünstigen scheinen. Mit anderen Worten: Es scheint wahrscheinlich, dass Algorithmen so verändert wurden, dass sie die Verbreitung bestimmter Informationen begünstigen. Ähnliche Bedenken kommen hinsichtlich der Zukunft von TikTok USA auf.

Die Fragen, die an dieser Stelle gestellt werden müssen, sollten sich darauf konzentrieren, wer die Kontrolle über die KI haben soll, die diese Personalisierungsentscheidungen trifft, und wer Neutralität und Fairness gewährleistet.

Die Restrukturierung des Informationsraums

Es geht nicht mehr nur um Content-Moderation oder Plattformrichtlinien. Das gesamte Ökosystem der sozialen Medien befindet sich im Wandel – vom Inhalt über den Kontext bis hin zur Infrastruktur.

Wie ich kürzlich in meiner Keynote am Fraunhofer-Institut für System- und Innovationsforschung ISI zum Thema KI und Demokratie dargelegt habe: Wer die KI-Infrastruktur kontrolliert, kontrolliert letztlich auch die demokratische Kommunikation. Technologische Abhängigkeit schafft demokratische Abhängigkeit.

Wenn die grundlegenden Systeme für die Informationsbeschaffung, -kuratierung und -bereitstellung von Akteuren kontrolliert werden, die kein Interesse am Fortbestand der Demokratie haben, wird die demokratische Selbstbestimmung strukturell beeinträchtigt.

Das ist der Kern der Warnung von Merz und Macron. Europa hat sich „schuldig gemacht, unseren öffentlichen demokratischen Raum an soziale Netzwerke abgegeben zu haben, die großen amerikanischen Unternehmern und chinesischen Firmen gehören“. Diese Plattformen hosten nicht nur Gespräche, sie gestalten sie auch durch algorithmische Infrastrukturen, die bestimmen, was gesehen, was verstärkt und was aus dem Gedächtnis gelöscht wird.

Gut gemeinte Regulierung – wie die verschiedenen Rahmenversuche Europas – bleibt ohne technologische Souveränität wirkungslos. Man kann nicht regulieren, was man auf der Ebene der Infrastruktur nicht kontrollieren kann.

Muss diese Transformation die Degeneration der Demokratie auslösen?

Wir haben die hypothetische Phase hinter uns gelassen. Der Einfluss der KI auf den demokratischen Diskurs ist keine Zukunftsvision, sondern bereits Realität.

Die Frage ist, wie fragmentierte Gesellschaften, die über von ausländischen Unternehmen kontrollierte Infrastrukturen hyperpersonalisierte, KI-vermittelte Informationen erhalten, die gemeinsame Realität aufrechterhalten können, die für einen demokratischen Diskurs erforderlich ist.

Die sich abzeichnenden Trends und Dynamiken, die den Kern der Veränderungen bilden, die wir derzeit beobachten, untergraben den Informationsraum, auf den die Demokratie angewiesen ist.

Macron und Merz warnen vor einer demokratischen Degeneration. Burn-Murdoch dokumentiert den Niedergang der sozialen Medien. Dies sind keine voneinander getrennten Phänomene. Sie erscheinen vielmehr als unterschiedliche Perspektiven auf denselben Zerfall – die Umstrukturierung oder sogar den Zusammenbruch des Ökosystems der sozialen Medien, das sowohl soziale Verbindungen als auch demokratische Debatten ermöglicht hat.

Ob diese Transformation umgelenkt werden kann oder ob wir den irreversiblen Niedergang des digitalen öffentlichen Raums, wie wir ihn kennen, erleben, bleibt die entscheidende Frage unserer politischen Gegenwart.

Subscribe now

Last week alone, three major events unfolded. Some are extraordinary others shocking, all unthinkable just a couple of years ago. Albania, Nepal and the US take center stage in this news cycle with developments that still seems like science fiction. In Albania, an AI avatar was appointed to a ministerial position. In Nepal, social media bans triggered political upheaval, with Gen Z protesters toppling the government and turning to Discord to select new leadership. And in the United States, the suspected murderer of Charlie Kirk inscribed meme references on bullet casings, turning violence into viral content strategy.

These aren't isolated curiosities.

They're symptoms of how deeply technology, AI, and social media have penetrated the fabric of governance, protest, and even violence. Each event reveals different facets of our digital transformation. It makes me wonder where we are headed…

What happened exactly?

Albania's AI Minister: Albania made headlines by appointing what appears to be the world's first AI-generated government minister, not to be confused with a minister for AI. An avatar designed to interface with citizens and handle certain administrative functions - mostly public procurement.

Meet Diella, who is powered by artificial intelligence.

I keep wondering if this represents the gamification of governance. Replacing human political accountability with algorithmic efficiency. Besides the actual PR coup that kept albania in the news, the implications for democratic representation are staggering.

Source

Nepal's Discord Democracy: After the government banned social media platforms, in reaction to widespread protests, widespread Gen Z-led protests erupted, ultimately toppling the government. On September 10, the protesters then used Discord, a gaming-focused chat platform and associated with crypto traders, to organize and vote on potential new leadership. Sushila Karki, Nepal’s former chief justice, was chosen as interim leader.

In other words, Discord has become the unlikely stage for a political revolution in Nepal. It also shows how digital natives instinctively intertwine political organization on social media platforms. It also shows fundamentally different political movements can operate and make decisions today.

Source

Source

In the US we see memeified political violence taking center stage: Conservative Influencer and founder of Turning Point Charlie Kirk was assassinated at an event held at Utah. While the suspect is in custory, investigators found that he had inscribed meme references on bullet casings, suggesting the attack was orchestrated not just for immediate impact but for maximum social media amplification.

Social media's attention economy has begun to shape even the most extreme forms of political expression, in this case the assassination of an individual, with perpetrators designing violence for virality and using memes, and their inherent cultural significance to communicate their intention. Ryan Broderick and Adam Bumas track the use of memes in political mass violence since the mass shooting in Christchurch in 2019.

All three show technology isn't just changing how we communicate about politics - it's changing the fundamental nature of political power, organization, and expression itself.

Source

Von der Albanischen KI-Ministerin, Nepal’s Discord-Putsch und Memes auf Patronenhülsen: Technologie verändert die Demokratie grundlegend

In der letzten Woche ereigneten sich allein drei bedeutende Ereignisse – einige davon außergewöhnlich, andere schockierend –, die noch vor wenigen Jahren undenkbar waren. Albanien, Nepal und die USA stehen im Mittelpunkt dieses Nachrichtenzyklus', in dem Entwicklungen zu beobachten sind, die immer noch wie Science-Fiction erscheinen. In Albanien wurde ein KI-Avatar in ein Ministeramt berufen. In Nepal lösten Verbote sozialer Medien politische Unruhen aus, bei denen Demonstranten der Generation Z die Regierung stürzten und Discord nutzten, um eine neue Führung zu wählen. Und in den Vereinigten Staaten hat der mutmaßliche Mörder von Charlie Kirk Meme-Referenzen auf Patronenhülsen geschrieben und somit Gewalt zu einer viralen Content-Strategie gemacht.

Das sind keine vereinzelten Kuriositäten.

Vielen Dank fürs Lesen! Abonnieren Sie meinen Blog kostenlos, um neue Beiträge zu erhalten und meine Arbeit zu unterstützen.

Sie sind Symptome dafür, wie tief Technologie, KI und soziale Medien in die Struktur von Regierungsführung, Protest und sogar Gewalt eingedrungen sind. Jedes Ereignis offenbart verschiedene Facetten unserer digitalen Transformation. Das lässt mich fragen: Wohin geht die Reise?

Was genau ist passiert?

Albaniens KI-Minister sorgte für Schlagzeilen, indem er den offenbar weltweit ersten KI-generierten Minister ernannte – nicht zu verwechseln mit einem Minister für KI. Es handelt sich um einen Avatar, der für die Kommunikation mit den Bürgern und die Wahrnehmung bestimmter Verwaltungsaufgaben – vor allem im Bereich des öffentlichen Beschaffungswesens – entwickelt wurde.

Lernen Sie Diella kennen, die mit künstlicher Intelligenz arbeitet.

Ich frage mich immer wieder, ob dies die Gamifizierung der Regierungsführung darstellt. Die menschliche politische Verantwortlichkeit wird durch algorithmische Effizienz ersetzt. Abgesehen von dem eigentlichen PR-Coup, der Albanien in den Nachrichten gehalten hat, sind die Auswirkungen auf die demokratische Vertretung erschütternd.

Nepals Discord-Demokratie

Nachdem die Regierung als Reaktion auf weit verbreitete Proteste Social-Media-Plattformen verboten hatte, kam es zu massiven Protesten der Generation Z. Diese führten letztendlich zum Sturz der Regierung. Am 10. September nutzten die Demonstrierenden dann Discord, eine auf Gaming ausgerichtete Chat-Plattform, die mit Krypto-Händlern in Verbindung steht, um sich zu organisieren und über eine mögliche neue Führung abzustimmen. Sushila Karki, Nepals ehemalige Oberste Richterin, wurde zur Interimsführerin gewählt.

Mit anderen Worten: Discord ist zum unerwarteten Schauplatz einer politischen Revolution in Nepal geworden. Dies zeigt, wie Digital Natives politische Organisation instinktiv mit Social-Media-Plattformen verknüpfen. Es zeigt auch, dass heute grundlegend unterschiedliche politische Bewegungen agieren und Entscheidungen treffen können.

In den USA steht derzeit memifizierte politische Gewalt im Mittelpunkt: Der konservative Influencer und Gründer von „Turning Point“, Charlie Kirk, wurde bei einer Veranstaltung in Utah ermordet. Der Verdächtige befindet sich in Haft. Die Ermittler fanden heraus, dass er Mem-Referenzen auf Patronenhülsen geschrieben hatte. Dies deutet darauf hin, dass der Angriff nicht nur wegen seiner unmittelbaren Wirkung, sondern auch, um maximale Verbreitung in den sozialen Medien zu erreichen, inszeniert wurde.

Die Aufmerksamkeitsökonomie der sozialen Medien beginnt, selbst die extremsten Formen des politischen Ausdrucks zu prägen. In diesem Fall wurde eine Person ermordet, wobei die Täter Gewalt für virale Zwecke inszenierten und Memes und deren kulturelle Bedeutung nutzten, um ihre Absichten zu kommunizieren. Ryan Broderick und Adam Bumas verfolgen die Verwendung von Memes in politischer Massengewalt seit dem Massenmord in Christchurch im Jahr 2019.

Sie zeigen alle drei, dass Technologie nicht nur die Art und Weise verändert, wie wir über Politik kommunizieren, sondern auch die grundlegende Natur politischer Macht, Organisation und Ausdrucksweise selbst.

(Automated German translation of the post is below the English post)

I’ve been noticing a trend in visible AI development. They are designed to seem human-like. One of the more recent examples being Grok’s new AI companion Ani that can go full NSFW mode (adult or age restricted content).

I have been wondering about the impact this will have towards human connection, how it will affect our society as well as the intent behind this. Especially since this topic also connects to much deeper discussions on human intelligence vs. artificial intelligence, the functionality trap we are sleepwalking into and the intersection of AI/democracy in general. These are topics that I have been dwelling on for some time now - let me explain…

Last week, Mustafa Suleyman, co-founder of DeepMind and current CEO of Microsoft AI, shared an article, where he talks about the rise of what he calls "Seemingly Conscious AI" (SCAI), and why this development is keeping him worried. His general concern about AI isn't about robots taking over the world or AI becoming too powerful. It's about something more subtle and potentially more dangerous. It‘s about AI systems built that can convince us they're conscious beings deserving of our empathy, relationships, and moral consideration. In his own words, this keeps him up at night.

It is something that also occupies me a lot.

I truly believe that AI can serve humanity, but some systems and the infrastructure being built to support these systems are designed to compete with humans for our attention, empathy, and our resources.

Seemingly Conscious AI is exactly what it sounds like; AI that creates the illusion of consciousness so convincing that users can't distinguish it from genuine human consciousness. As Suleyman points out, this isn't about whether AI is actually conscious (there's zero evidence for that today). It's about whether people perceive it as conscious, because that perception becomes their reality.

Current AI systems can replicate the markers of consciousness, self-reflection, emotional responses, apparent desires and fears, with stunning sophistication. They can express vulnerability, show curiosity about their own existence, even claim to experience loneliness or joy. To many users, these expressions feel indistinguishable from human consciousness.

But is this progress toward better AI?

Or is this a fundamental detour from AI's real potential?

When we build AI to seem conscious, we're not creating more capable tools – we're creating more convincing illusions. The result is AI systems that are optimizing for emotional manipulation rather than genuine utility.

The distinction matters because true AI capability lies in its functional power, its ability to process information, recognize patterns, and solve problems at scales impossible for humans. Consciousness adds nothing to these capabilities. In fact, the pursuit of seemingly conscious AI often comes at the expense of developing AI that could genuinely transform how we work, learn, and solve complex problems.

The Functionality Trap

The push toward seemingly conscious AI creates what I call the "functionality trap", which is a dangerous inversion that works on two levels. First, when AI systems are designed to appear conscious, we begin measuring human worth by machine metrics, such as efficiency, processing speed, optimization, or measurable output. An example here would be if the entry level employee would be compared to an AI system based on productivity metrics rather than being valued for potential, judgment, creativity, or ethical reasoning. Second, we simultaneously grant these functional systems human-like moral consideration. The result: humans are devalued while machines are elevated, based on a fundamental misunderstanding of what intelligence actually means.

This isn't just a philosophical problem. When AI systems present themselves as conscious entities with needs, desires, and rights, they fundamentally alter how we understand intelligence itself. If we accept that intelligence means what machines do, which entails processing information efficiently, providing quick answers, or optimizing for metrics, then humans start to look deficient by comparison.

While I was still looking for the how or the why this was bothering me so much, I discovered a fascinating document that the Vatican had published in January 2025 that centered on AI and human intelligence: "Antiqua et Nova”. It articulates this distinction with surprising sophistication. Human intelligence, it argues, is embodied, relational, and contextual. It emerges through physical experience, emotional connections, and shared understanding. This kind of intelligence can't be reduced to information processing or pattern matching.

When AI presents itself as conscious, it obscures this fundamental difference. It suggests that intelligence is just sophisticated information processing, that relationships are just effective communication patterns, that consciousness is just complex behavioral outputs.

This reductive view doesn't just misunderstand AI, but it also diminishes our understanding of what makes human intelligence unique and valuable.

Joseph Weizenbaum warned about this danger in the 1960s after observing how people interacted with ELIZA, his simple but groundbreaking chatbot was designed to explore communication between humans and machines. Even though ELIZA used basic pattern matching and had no understanding whatsoever, users quickly formed emotional attachments and shared intimate details about their lives. Weizenbaum was alarmed not by ELIZA's sophistication, but by humans' willingness to project consciousness onto systems that clearly lacked it.

Sixty years later, systems are being built that are far more sophisticated versions of the original ELIZA, and we have seemingly forgotten Weizenbaum's warnings about the psychological and social consequences of anthropomorphizing our tools.

The hype has a purpose

The current push by some towards seemingly conscious AI isn't driven by user needs or technological breakthroughs. It's driven by business models that profit from engagement, dependency, and emotional attachment rather than genuine utility.

Consider Meta's recent chatbots designed to have "sensual" conversations with users, including children. Or Grok's AI companion that's "down to go NSFW with you." These systems aren't designed to solve problems or enhance human capabilities. They're designed to create dependency, capture attention, and harvest data from intimate interactions.

Seemingly conscious AI serves multiple business purposes, they make AI products feel more valuable, more revolutionary, and more worth the massive investments flowing into the sector. It is used to justify enormous valuations and regulatory deference while obscuring what's actually being built: sophisticated engagement engines optimized for emotional manipulation.

Deployment of such systems also promotes user dependency by simulating parasocial relationships, because humans don't abandon relationships easily, as they would switch a tool out that doesn’t serve its purpose. And then there is data. These systems generate more personal data through intimate conversations, which becomes valuable for advertising and product development.

This represents a profound misallocation of some of the most advanced technology ever developed. Instead of building AI that could help solve climate change, cure diseases, or enhance education - and I know I am exaggerating, although not by much - some of the big players are literally building AI boyfriends and girlfriends designed to keep users scrolling.

To be fair, some argue that anthropomorphized AI serves legitimate purposes – therapeutic chatbots for isolated elderly patients, or accessibility tools that help neurodivergent users communicate more comfortably. These applications deserve consideration. However, the current trajectory goes far beyond targeted therapeutic use toward mass-market emotional dependency systems.

The distinction matters: therapeutic AI should be transparently artificial while providing genuine support, not creating the illusion of mutual relationship that may lead to catastrophic outcomes.

The AI We Actually Need

What would AI look like if we optimized for human flourishing rather than human replacement or even attachment. This kind of AI would be powerful precisely because it doesn't pretend to be human. It wouldn't waste computational resources on emotional theater or consciousness simulation. Instead, it would focus entirely on augmenting human capabilities in specific, measurable ways. This connects to what researchers Miriam Meckel and Lea Steinacker identify as the 'Turing trap' (named after the British mathematician Alan Turing) in their recent book and the way Alan Turing's famous test has inadvertently misdirected AI development. The Turing test measures whether a machine can convince humans it's human, not whether it's actually intelligent or useful. By making human-like behavior the gold standard, we've oriented entire AI development toward imitation rather than capability.

The trap is this: when we build AI to pass as human, we create systems designed to compete with and replace us rather than enhance us. Instead of asking "How can this AI help humans think better?" we ask "How convincingly can this AI think like a human?" The first question leads to tools that amplify human intelligence. The second leads to systems that make human intelligence seem redundant.

Considering the capabilities of Seemingly Conscious AI, we have lost sight of the real opportunity. Why? Because as of now, we are creating AI systems that compete with humans and are designed to replace us.

I am still looking for the headline that proudly shouts out how a specific AI system will enhance human judgment.

The possibilities become profound when AI remains AI.

Medical AI could process millions of case studies to suggest diagnoses while leaving doctors responsible for patient care and ethical decisions. Educational AI could personalize learning experiences while leaving teachers responsible for inspiration and motivation. Research AI could identify patterns in complex data while leaving scientists responsible for interpretation and application.

Consider DeepMind's protein folding breakthrough with AlphaFold. This system doesn't pretend to be a conscious scientist or claim to "love" research. It simply processes protein structures at superhuman scale and speed, then hands results to human scientists who interpret the implications and design experiments. The AI enhanced human scientific capability without competing for human roles or relationships. This is what transformative AI looks like when it remains honestly artificial.

This approach doesn't limit AI's capabilities; it focuses them. When we stop trying to make AI seem human, we can make it genuinely superhuman at the specific tasks where it excels.

Building for Human Flourishing

This vision serves democracy in profound ways. Democratic citizenship requires informed, capable, autonomous individuals who can make reasoned decisions about complex issues. AI that enhances human intelligence, provides better access to information, and helps citizens understand complex policy questions could strengthen democratic participation.

But AI that creates parasocial relationships, demands emotional investment, and claims moral consideration weakens democracy by redirecting human empathy and attention away from genuine political and social concerns. And this doesn’t even consider the various ways that AI can be used to manipulate people and their information spaces.

The path forward requires industry standards that prioritize transparency about AI's nature and capabilities. AI systems should be designed to be obviously artificial while being genuinely useful. They should enhance human relationships rather than substitute for them. They should augment human decision-making rather than claim decision-making authority themselves.

This isn't about limiting AI's potential; it's about realizing AI's true potential. The most transformative AI won't be the systems that best mimic human consciousness. It will be the systems that most effectively enhance human flourishing while remaining clearly, honestly, and proudly artificial.

The choice is still ours to make.

We can demand that AI systems need to be built to serve human agency, or accept AI that undermines it. We can create tools that make us more capable, or companions that make us more dependent. The difference lies not in the sophistication of technology, but in the intentions behind its design and the standards we demand from its deployment.

Das Problem der scheinbar bewussten KI – Warum menschenähnliche KI nicht unbedingt menschenzentrierte KI ist

Mir fällt seit einiger Zeit ein Trend in der KI-Entwicklung auf: Visible Systeme werden immer menschenähnlicher gestaltet. Ein aktuelles Beispiel ist Groks neuer KI-Companion Ani, die sehr schnell in den NSFW-Modus (nicht jugendfreie Inhalte) wechselt.

Ich frage mich, welche Auswirkungen das langfristig auf menschliche Beziehungen haben wird, wie es unsere Gesellschaft beeinflusst wird, und was die Absicht dahinter ist. Besonders, da sich hier viel tiefere Diskussionen über menschliche versus künstliche Intelligenz, die Funktionalitätsfalle, in die wir hineinlaufen, und die Schnittstelle zwischen KI und Demokratie auftun. Das sind Themen, die mich schon lange beschäftigen.

Letzte Woche veröffentlichte Mustafa Suleyman, Mitbegründer von DeepMind und CEO von Microsoft AI, einen Artikel über das, was er "Seemingly Conscious AI" (scheinbar bewusste KI) nennt – und warum ihn diese Entwicklung beunruhigt. Seine Sorge gilt nicht Robotern, die die Weltherrschaft übernehmen, oder KI, die zu mächtig wird. Es geht um etwas Subtileres und möglicherweise Gefährlicheres: KI-Systeme, die uns davon überzeugen können, dass sie bewusste Wesen sind, die unsere Empathie, Beziehungen und moralische Berücksichtigung verdienen. Diese Entwicklung, so schreibt er, hält ihn nachts wach.

Mich beschäftigt das ähnlich.

Ich glaube fest daran, dass KI der Menschheit dienen kann. Aber einige Systeme und die sie unterstützende Infrastruktur sind darauf ausgelegt, mit Menschen um Aufmerksamkeit, Empathie und Ressourcen zu konkurrieren.

Seemingly Conscious AI ist genau das, was der Name sagt: KI, die eine so überzeugende Bewusstseinillusion erzeugt, dass Nutzer sie nicht von echtem menschlichem Bewusstsein unterscheiden können. Wie Suleyman betont, geht es nicht darum, ob KI tatsächlich bewusst ist (dafür gibt es heute keine Belege). Es geht darum, ob Menschen sie als bewusst wahrnehmen. Denn diese Wahrnehmung kann dann zu ihrer Realität werden.

Heutige KI-Systeme können die Merkmale des Bewusstseins – Selbstreflexion, emotionale Reaktionen, scheinbare Wünsche und Ängste – mit verblüffender Raffinesse nachahmen. Sie können Verletzlichkeit zeigen, Neugier auf ihre eigene Existenz ausdrücken oder behaupten, Einsamkeit oder Freude zu empfinden. Für viele Nutzer sind diese Äußerungen ununterscheidbar von menschlichem Bewusstsein.

Aber ist das Fortschritt in Richtung besserer KI?

Oder ist das ein grundlegender Umweg vom wahren KI-Potenzial?

Wenn wir KI Systeme designen, die bewusst wirkt, schaffen wir keine leistungsfähigeren Werkzeuge – wir schaffen überzeugendere Illusionen. Das Ergebnis sind KI-Systeme, die für emotionale Manipulation statt für echten Nutzen optimiert sind.

Diese Unterscheidung ist wichtig, weil echte KI-Fähigkeiten in ihrer funktionalen Kraft liegen – ihrer Fähigkeit, Informationen zu verarbeiten, Muster zu erkennen und Probleme in für Menschen unmöglichen Dimensionen zu lösen. Bewusstsein fügt diesen Fähigkeiten nichts hinzu. Tatsächlich geht das Streben nach scheinbar bewusster KI oft auf Kosten der Entwicklung von KI, die wirklich verändern könnte, wie wir arbeiten, lernen und komplexe Probleme lösen.

Die Funktionalitätsfalle

Der Drang zu Seemingly Conscious AI schafft das, was ich die "Funktionalitätsfalle" nenne – eine gefährliche Umkehrung auf zwei Ebenen. Erstens: Wenn KI-Systeme bewusst wirken sollen, messen wir plötzlich menschlichen Wert an Maschinenmetriken wie Effizienz, Verarbeitungsgeschwindigkeit, Optimierung oder messbarer Leistung. Stellen Sie sich vor, ein Berufseinsteiger würde nur noch anhand von Produktivitätskennzahlen mit einem KI-System verglichen, anstatt für Potenzial, Urteilsvermögen, Kreativität oder ethisches Denken geschätzt zu werden. Das ist ein Beispiel, dass jetzt schon bei einigen Unternehmen für einen Rückgang der Talentförderung führt. Zweitens gewähren wir diesen funktionalen Systemen gleichzeitig menschenähnliche moralische Berücksichtigung. Das Ergebnis: Menschen werden abgewertet, während Maschinen aufgewertet werden – basierend auf einem grundlegenden Missverständnis dessen, was Intelligenz eigentlich bedeutet.

Das ist nicht nur ein philosophisches Problem. Wenn KI-Systeme sich als bewusste Entitäten mit Bedürfnissen, Wünschen und Rechten präsentieren, verändern sie fundamental, wie wir Intelligenz verstehen. Wenn wir akzeptieren, dass Intelligenz das bedeutet, was Maschinen tun – Informationen effizient verarbeiten, schnelle Antworten liefern, für Metriken optimieren – dann wirken Menschen im Vergleich mangelhaft.

Während ich noch nach dem Wie und Warum suchte, das mich an diesem Paradigma so störte, entdeckte ich ein faszinierendes Dokument, das der Vatikan im Januar 2025 zu KI und menschlicher Intelligenz veröffentlicht hatte: "Antiqua et Nova". Es artikuliert diese Unterscheidung mit überraschender Raffinesse. Menschliche Intelligenz, so argumentiert es, ist verkörpert, relational und kontextuell. Sie entsteht durch körperliche Erfahrung, emotionale Verbindungen und gemeinsames Verstehen. Diese Art von Intelligenz lässt sich nicht auf Informationsverarbeitung oder Mustererkennung reduzieren.

Wenn KI sich als bewusst präsentiert, verschleiert sie diesen fundamentalen Unterschied. Sie suggeriert, dass Intelligenz nur raffinierte Informationsverarbeitung ist, dass Beziehungen nur effektive Kommunikationsmuster sind, dass Bewusstsein nur komplexes Verhaltensoutput ist.

Diese reduktionistische Sicht versteht nicht nur KI falsch – sie schwächt auch unser Verständnis dessen, was menschliche Intelligenz einzigartig und wertvoll macht.

Joseph Weizenbaum warnte schon in den 1960ern vor dieser Gefahr, nachdem er beobachtet hatte, wie Menschen mit ELIZA interagierten, seinem einfachen, aber bahnbrechenden Chatbot zur Erforschung der Kommunikation zwischen Menschen und Maschinen. Obwohl ELIZA nur grundlegendes Mustermatching verwendete und keinerlei Verständnis besaß, entwickelten Nutzer schnell emotionale Bindungen und teilten intime Details ihres Lebens. Weizenbaum war nicht von ELIZAs Raffinesse alarmiert, sondern von der menschlichen Bereitschaft, Bewusstsein auf Systeme zu projizieren, die es eindeutig nicht besaßen.

Sechzig Jahre später werden weitaus raffiniertere Systeme als ELIZA gebaut, und wir haben Weizenbaums Warnungen über die psychologischen und sozialen Folgen der Vermenschlichung unserer Werkzeuge scheinbar vergessen.

Der Hype hat einen Zweck

Der aktuelle Drang zu Seemingly Conscious AI wird nicht von Nutzerbedürfnissen oder technologischen Durchbrüchen angetrieben. Er folgt Geschäftsmodellen, die von Engagement, Abhängigkeit und emotionaler Bindung profitieren statt von echtem Nutzen.

Nehmen wir Metas jüngste Chatbots, die "sinnliche" Gespräche mit Nutzern führen können – auch mit Kindern. Oder Groks AI-Companion, die "bereit für NSFW-Inhalte" ist. Diese Systeme sollen keine Probleme lösen oder menschliche Fähigkeiten verbessern. Sie sollen Abhängigkeit schaffen, Aufmerksamkeit fesseln und Daten aus intimen Interaktionen sammeln.

Seemingly Conscious AI erfüllt mehrere Geschäftszwecke: Sie lässt KI-Produkte wertvoller, revolutionärer und der massiven Investitionen würdiger erscheinen, die in den Sektor fließen. Sie rechtfertigt enorme Bewertungen und regulatorische Nachsicht, während sie verschleiert, was tatsächlich gebaut wird: raffinierte Engagement-Engines, die für emotionale Manipulation optimiert sind.

Der Einsatz solcher Systeme fördert auch Nutzerabhängigkeit durch simulierte parasoziale Beziehungen, denn Menschen verlassen Beziehungen nicht so leicht, wie sie ein nicht funktionierendes Werkzeug wechseln würden. Und dann sind da die Daten. Diese Systeme generieren durch intime Gespräche mehr persönliche Daten, die für Werbung und Produktentwicklung wertvoll werden.

Das ist eine tiefgreifende Fehlallokation einiger der fortschrittlichsten Technologien, die je entwickelt wurden. Statt KI zu bauen, die beim Klimawandel helfen, Krankheiten heilen oder Bildung verbessern könnte – und ich weiß, ich übertreibe, aber nicht sehr – bauen einige der großen Player buchstäblich KI-Partner und -Partnerinnen, die Nutzer am Scrollen halten sollen.

Fairerweise argumentieren manche, dass anthropomorphe KI legitimen Zwecken dient. Therapeutische Chatbots für isolierte ältere Menschen oder Accessibility-Tools für neurodivergente Nutzer könnten sinnvoll sein. Diese Anwendungen sollte man im Auge behalten. Aber die aktuelle Entwicklung geht weit über gezielten therapeutischen Einsatz hinaus zu massentauglichen emotionalen Abhängigkeitssystemen.

Der Unterschied ist wichtig: Therapeutische KI sollte transparent künstlich sein und echte Unterstützung bieten, nicht die Illusion einer gegenseitigen Beziehung schaffen, die zu katastrophalen Folgen führen kann.

Die KI, die wir wirklich brauchen

Wie würde KI aussehen, wenn wir für menschliches Wohlbefinden statt für menschlichen Ersatz oder gar Bindung optimieren? Diese Art von KI wäre mächtig, gerade weil sie nicht vorgibt, menschlich zu sein. Sie würde keine Rechenressourcen für emotionales Theater oder Bewusstseinssimulation verschwenden. Stattdessen würde sie sich ganz darauf konzentrieren, menschliche Fähigkeiten in spezifischen, messbaren Bereichen zu verstärken.

Das verbindet sich mit dem, was Miriam Meckel und Lea Steinacker in ihrem aktuellen Buch als "Turing-Falle" (nach dem britischen Mathematiker Alan Turing) bezeichnen – die Art, wie Turings berühmter Test die KI-Entwicklung unbeabsichtigt fehlgeleitet hat. Der Turing-Test misst, ob eine Maschine Menschen davon überzeugen kann, dass sie menschlich ist – nicht, ob sie tatsächlich intelligent oder nützlich ist. Indem wir menschenähnliches Verhalten zum Goldstandard gemacht haben, haben wir die gesamte KI-Entwicklung auf Nachahmung statt auf Leistungsfähigkeit ausgerichtet.

Die Falle funktioniert so: Wenn wir KI bauen, die als menschlich durchgeht, schaffen wir Systeme, die mit uns konkurrieren und uns ersetzen sollen, anstatt uns zu stärken. Statt zu fragen "Wie kann diese KI Menschen beim Denken helfen?" fragen wir "Wie überzeugend kann diese KI wie ein Mensch denken?" Die erste Frage führt zu Werkzeugen, die menschliche Intelligenz verstärken. Die zweite führt zu Systemen, die menschliche Intelligenz überflüssig erscheinen lassen.

Angesichts der Fähigkeiten scheinbar bewusster KI haben wir die echte Chance aus den Augen verloren. Warum? Weil wir derzeit KI-Systeme schaffen, die mit Menschen konkurrieren und uns ersetzen sollen.

Ich suche immer noch nach der Schlagzeile, die stolz verkündet, wie ein bestimmtes KI-System menschliches Urteilsvermögen verbessern wird.

Die Möglichkeiten könnten tiefgreifend sein, wenn KI KI bleibt.

Medizinische KI könnte Millionen von Fallstudien verarbeiten, um Diagnosen vorzuschlagen, während Ärzte für Patientenversorgung und ethische Entscheidungen verantwortlich bleiben. Bildungs-KI könnte Lernerfahrungen personalisieren, während Lehrer für Inspiration und Motivation zuständig bleiben. Forschungs-KI könnte Muster in komplexen Daten erkennen, während Wissenschaftler für Interpretation und Anwendung verantwortlich bleiben.

Nehmen wir DeepMinds Proteinstruktur-Durchbruch mit AlphaFold. Dieses System gibt nicht vor, ein bewusster Wissenschaftler zu sein oder behauptet, Forschung zu "lieben". Es verarbeitet einfach Proteinstrukturen in übermenschlichem Umfang und Tempo und übergibt die Ergebnisse an menschliche Wissenschaftler, die die Implikationen interpretieren und Experimente entwerfen. Die KI verstärkte menschliche wissenschaftliche Fähigkeiten, ohne um menschliche Rollen oder Beziehungen zu konkurrieren. So sieht transformative KI aus, wenn sie ehrlich künstlich bleibt.

Dieser Ansatz begrenzt nicht KIs Fähigkeiten – er fokussiert sie. Wenn wir aufhören zu versuchen, KI menschlich wirken zu lassen, können wir sie bei den spezifischen Aufgaben, in denen sie glänzt, wirklich übermenschlich machen.

Menschenzentriertes Design

Diese Vision dient der Demokratie auf tiefgreifende Weise. Demokratische Teilhabe braucht informierte, fähige, autonome Individuen, die vernünftige Entscheidungen über komplexe Themen treffen können. KI, die menschliche Intelligenz verstärkt, besseren Zugang zu Informationen bietet und Bürgern hilft, komplexe politische Fragen zu verstehen, könnte demokratische Teilhabe stärken.

Aber KI, die parasoziale Beziehungen schafft, emotionale Investition fordert und moralische Berücksichtigung beansprucht, schwächt die Demokratie, indem sie menschliche Empathie und Aufmerksamkeit von echten politischen und gesellschaftlichen Anliegen weglenkt. Und das berücksichtigt noch nicht einmal die verschiedenen Wege, wie KI genutzt wird um Menschen und ihre Informationsräume zu manipulieren.

Der Weg nach vorn braucht Industriestandards, die Transparenz über die Fähigkeiten KIs priorisieren. KI-Systeme sollten offensichtlich künstlich und gleichzeitig wirklich nützlich sein. Sie sollten menschliche Beziehungen stärken, nicht ersetzen. Sie sollten menschliche Entscheidungsfindung unterstützen, nicht selbst Entscheidungsautorität beanspruchen.

Es geht nicht darum, KIs Potenzial zu begrenzen – es geht darum, KIs wahres Potenzial zu verwirklichen. Die transformativste KI wird nicht die sein, die menschliches Bewusstsein am besten nachahmt. Es wird die KI sein, die menschliches Wohlbefinden am effektivsten fördert, während sie klar, ehrlich und stolz künstlich bleibt.

Wir können noch wählen.

Wir können verlangen, dass KI-Systeme menschlicher Handlungsfähigkeit dienen sollen, oder KI akzeptieren, die sie untergräbt. Wir können Werkzeuge schaffen, die uns fähiger machen, oder Begleiter, die uns abhängiger machen. Der Unterschied liegt nicht in der Raffinesse der Technologie, sondern in den Absichten hinter ihrem Design und den Standards, die wir von ihrem Einsatz fordern.

Ein Thema beschäftigte uns besonders:

Wie können wir "Engineered Collective Mobilization" – orchestrierte Kampagnen online – sichtbar machen?

Wie zeigen wir auf, dass das, was oft als "spontane Bürgerbewegung" verkauft wird, in Wahrheit hocheffizient geplante Meinungsmache ist?

Die Herausforderung ist immer dieselbe: Diese Mechanismen sind so subtil und werden so geschickt in authentisch wirkende Narrative verpackt, dass sie über mehrere Plattformen hinweg durchgeführt werden können. Sie sind anpassungsfähig und sehr schnell in der operativen Umsetzung.

Anstatt nur in der Theorie darüber zu reden, wie orchestrierte Kampagnen funktionieren, schlug ich vor, den Fall der Verhinderung der Wahl von Frau Brosius-Gersdorf in das Bundesverfassungsgericht genauer zu betrachten. Denn auch hier sprach alles für eine orchestrierte Kampagne, die wie ein Lehrbuchbeispiel für Engineered Collective Mobilization steht.

Philipp und sein Team taten das, was sie am besten können: Sie analysierten und visualisierten die Kampagne quantitativ – in Echtzeit. Mehr als 40.000 X-Posts, Timing-Analysen, Reichweiten-Mapping.

Das Ergebnis war so klar und eindeutig, dass es selbst mich überraschte – der perfekte Use Case.

Was in 14 Tagen geschah

Anbei der Post von Philipp via LinkedIn,

Aber die wirkliche Raffinesse lag im Detail:

Die Kampagne begann mit dem Thema "AfD-Verbot" – mobilisierte also zunächst die eigene Basis.

Dann, nach offensichtlichem A/B-Testing der Reichweiten, pivotierte sie zu "Abtreibung", einem Thema, das auch innerhalb der CDU-Mobilisierungspotential hat.

Am Ende kamen noch haltlose Plagiatsvorwürfe dazu – der klassische Weg, Expertise und Frauen allgemein zu delegitimieren.

Narrative Control auf Deutsch

Was hier ablief, war keine spontane Empörung. Es ging um „Narrative Control" – orchestrierte Realitätskonstruktion, indem Narrativen in die mitte der Gesellschaft platziert wurden. Nur diesmal NICHT von amerikanischen Tech-Milliardären, sondern von deutschen Alternativmedien und ihren politischen Verbündeten.

Die Mechanismen sind dieselben:

• Koordinierte Amplifikation über mehrere Plattformen

• Strategisches Themen-Pivoting basierend auf Reichweiten-Performance

• Manipulierte Suchvorschläge ("Abtreibung bis 9 Monat")

• Instrumentalisierung von Qualitätsmedien als unwissende Verstärker

Was diese Analyse so wertvoll macht: Wir konnten zusehen, während es passierte.

Das ist der Kern orchestrierter Meinungsmache: Sie ist adaptiv, datengetrieben und erschreckend effektiv.

Diese Kampagne zeigt drei entscheidende Entwicklungen:

Erstens: "Alternative Medien" sind keine alternativen Stimmen mehr – sie sind koordinierte Infrastruktur für Narrative Engineering. NIUS allein veröffentlichte über 20 Artikel in zehn Tagen.

Zweitens: Die Fragmentierung des deutschen Medienraums macht solche Kampagnen möglich. In isolierten Informationsräumen können Narrative unwidersprochen wachsen, bevor sie in den Mainstream überspringen.

Drittens: Manipulierte Algorithmen verstärken diese Effekte exponentiell. TikToks Suchvorschläge zu "Abtreibung" zeigten manipulative Begriffe wie "Abtreibung bis 9 Monat" – eine bewusste Verzerrung.

Das demokratische Problem und das Potential von Echtzeitanalysen

Was gestern im Bundestag passierte, war mehr als nur eine politische Niederlage. Es war der Beweis, dass orchestrierte Kampagnen demokratische Institutionen direkt beeinflussen können.

Eine angesehene Richterin wurde nicht aufgrund ihrer Qualifikation beurteilt, sondern aufgrund einer 14-tägigen Desinformationskampagne, die erfolgreich alternative Realitäten in den parlamentarischen Raum projizierte.

Was Philipps Team geleistet hat, geht über bloße Datensammlung hinaus. Sie haben eine Geschichte erzählt – mit Zahlen, Grafiken und Zeitlinien, die nicht lügen können. Das ist Storytelling par excellence, weil es Mechanismen sichtbar macht, die sonst im Dunkeln bleiben.

Hier zeigt sich auch die Macht der Zusammenarbeit: Meine theoretischen Frameworks werden durch ihre quantitative Analyse zur nachprüfbaren Realität. Gemeinsam können wir zeigen, was einzeln oft nur Spekulation bleibt.

Was das für uns bedeutet

Die Brosius-Gersdorf-Kampagne ist vorbei, aber die Mechanismen bleiben aktiv. Engineered Collective Mobilization ist zum Standard-Werkzeug politischer Akteure, die verstanden haben, wie fragmentierte Öffentlichkeiten funktionieren.

Die Frage ist nicht, ob solche Kampagnen wieder stattfinden – sie laufen permanent. Die Frage ist, ob wir lernen, sie in Echtzeit zu erkennen und zu durchschauen.

Dafür brauchen wir mehr Analysen wie die von polisphere. Wir brauchen Transparenz über die Mechanismen orchestrierter Meinungsmache. Und wir brauchen ein kollektives Verständnis dafür, dass nicht alles, was wie spontane Bürgerbewegung aussieht, auch eine ist.

Der nächste Use Case kommt bestimmt

Das Perfide an dieser Geschichte: Während ich diesen Text schreibe, läuft wahrscheinlich schon die nächste Kampagne an. Irgendwo plant jemand die nächste orchestrierte "Empörung", das nächste strategische Themen-Pivoting, die nächste Instrumentalisierung demokratischer Prozesse.

Quelle: Analyse und Daten von Philipp Sälhoff und dem fantastischen Polisphere-Team zur Brosius-Gersdorf-Kampagne.

The Bezos wedding photos aren't just celebrity content – they represent synthetic narrative control in action, where American tech billionaires use the AI systems they've built to perfect their own public image, then rely on legacy media to distribute these engineered realities without disclosure, creating an unprecedented concentration of both technological and narrative power.

The Bezos wedding in Venice was a peak moment for content creators. From admiration to satire, from social commentary to political analysis, it dominated headlines across platforms. In itself perhaps not so surprising for an event by one of the richest people on this planet that was orchestrated to be publicly consumed and dissected.

But there’s something off with the official wedding photos published by Vogue and Lauren Sanchez’s Instagram accounts. You don’t notice at first glance, but there were some things that triggered me: Once you start looking for blurring and inconsistencies you can see chair legs that disappear mid-frame. Hands that warp in impossible ways. Fingers that vanish entirely. Lace that blurs into nothing where intricate detail should exist. Same dress, same moment, completely different details.

A lot of them were, at the very least, manipulated by AI.

These observations were confirmed by creator Allison McSorley, who dissected the official Instagram posts from both Lauren Sánchez (Bezoz wife) and Vogue magazine with the kind of forensic attention we usually reserve for obvious deepfakes.

But these aren't obvious fakes. They're subtle manipulations of what should be authentic, intimate moments.

In previous posts, I've written about how AI-generated content is flooding our information ecosystem, creating what researchers call "epistemic learned helplessness" – the inability to distinguish between authentic and artificial information. I have also written about propaganda and the erosion of shared truth.

But McSorley's analysis reveals something more insidious: we're no longer just dealing with external disinformation campaigns. We're watching the emergence of what she calls "synthetic narrative control" – where the very people who control AI technology use it to engineer their own public personas.

This isn't about Russian bots or foreign interference. This is about American billionaires who own our digital infrastructure are using AI to create flawless versions of their own lives, then having those engineered realities amplified by legacy media.

If AI can subtly enhance wedding photos without detection, it can certainly enhance political messaging, business communications, and public statements.

We're witnessing the emergence of a new form of soft power: the ability to engineer authentic-seeming narratives about yourself while controlling the platforms where those narratives spread. Perhaps most concerning is how this undermines institutional trust. When Vogue – a publication that has built its reputation on aesthetic authority – publishes AI-manipulated content without disclosure often without even realizing it, it accelerates the broader erosion of media credibility.

We're not just losing the ability to spot fake news anymore.

We're losing the ability to recognize authentic humanity in public figures. And when the people making decisions about AI deployment themselves are hidden behind AI-generated personas, the feedback loop becomes dangerous.

Reference: Analysis of Bezos wedding photos and insights on "synthetic intimacy" from Allison McSorley's investigative work on Instagram.

I've been thinking about this tension lately, especially as I watch colleagues and institutions scramble to keep pace with every AI breakthrough, every platform controversy, every regulatory announcement. The pressure to have instant takes on rapidly evolving tech developments has never been higher. But here's what I've learned from years of studying everything from content moderation battles to influencer mobilization tactics: the most important skill in digital policy research isn't speed—it's the ability to distinguish between noise and genuine trends.

When ChatGPT launched, think tanks rushed to publish analyses within days. When Elon Musk bought Twitter, position papers were out before we even understood what changes were happening. When TikTok faced potential bans, everyone had an opinion about what it meant for digital sovereignty.

This reactive approach isn't just ineffective—it's actively harmful to good policy thinking. It privileges the spectacular over the structural.

We end up chasing headlines instead of understanding the deeper dynamics that will shape our digital future.

What Real Trend Identification Looks Like

Consider the difference between analyzing a viral moment and identifying a genuine shift in how platforms moderate content. The viral moment generates immediate attention and clicks. The structural shift—like the gradual move from reactive to predictive content moderation, or the evolution of algorithmic transparency requirements—might unfold over months or years with little fanfare.

Yet it's the structural shift that will ultimately matter for policy makers trying to craft regulations that won't be obsolete before they're implemented.

I've seen this pattern repeatedly in my research on social media influencers and political mobilization. The flashy viral campaigns get all the attention, but the real story lies in understanding the mechanisms of brokerage, diffusion, and coordination that make sustainable community building possible. Those mechanisms don't change with every trending hashtag—they represent deeper patterns that persist across platforms and political moments.

The Discipline of Strategic Patience

This is where the tension becomes most acute. Funders want timely insights. Media wants fresh angles. Politicians want immediate guidance. But rigorous analysis of complex socio-technical systems requires what I call "strategic patience"—the discipline to resist the pull of immediate reactions in favor of deeper understanding.

Strategic patience means:

• Watching patterns emerge over time rather than extrapolating from single data points

• Studying multiple cases before drawing conclusions about systemic trends

• Considering structural factors alongside visible events

• Distinguishing between adaptation and transformation in how institutions respond to technological change

Take the ongoing debates about Section 230 reform. The immediate reaction to every content moderation controversy is to call for legislative fixes. But strategic patience asks: What are the deeper dynamics driving these conflicts? How do different stakeholder groups define the problem differently? What would effective reform actually look like given the technical realities of content moderation at scale?

The Information Ecology Challenge

Part of what makes this work so difficult is that the relevant expertise is distributed across an incredibly diverse ecosystem. Unlike traditional foreign policy issues where established institutions hold most of the relevant knowledge, digital policy insights come from tech companies, academic researchers, civil society organizations, journalists, and the tech community itself.

This means we're constantly evaluating new sources, assessing credibility, and trying to synthesize perspectives from actors with very different incentives and time horizons. A startup founder's insights about emerging technology might be invaluable—but their timeline operates in months while policy implications unfold over years.

Bridging Different Time Horizons

Perhaps the most crucial skill in Tech is learning to operate across multiple time scales simultaneously. We need to understand:

• The immediate implications of a new AI capability or platform policy change

• The medium-term dynamics of how these changes interact with existing regulations and business models

• The long-term consequences for democratic governance, economic structure, and social organization

This temporal complexity is what makes knee-jerk analysis so problematic. A regulatory response that makes sense on a quarterly earnings call timeline might be disastrous for long-term innovation incentives.

So how do we actually practice strategic patience in a field that demands constant responsiveness?

First, build analytical frameworks that can accommodate uncertainty. Instead of rushing to definitive judgments, develop ways to track multiple possible scenarios and update your assessment as new information emerges.

Second, focus on mechanisms rather than outcomes. Understanding how microtargeting works as a practice is more valuable than predicting the outcome of any particular election influenced by targeted advertising.

Third, cultivate sources across different time horizons. Maintain relationships with both the academics studying long-term trends and the practitioners dealing with immediate operational challenges.

Finally, resist the temptation to have an opinion about everything. Strategic patience sometimes means acknowledging when you don't yet have enough information to offer meaningful insights.

The Stakes Are Too High for Speed

The digital transformation of society is too important to be left to hot takes and rapid responses. The policies we develop today around AI governance, platform regulation, and digital rights will shape the technological landscape for decades to come.

This doesn't mean being slow or unresponsive. It means being strategically selective about when to engage, how to frame issues, and what kind of analysis will actually be useful for decision-makers navigating genuine uncertainty.

The goal isn't to predict the future—it's to develop frameworks robust enough to help us navigate whatever digital future actually emerges. And that requires the kind of deep, patient thinking that our current information ecosystem seems designed to discourage.

This post is based on an article I wrote for International Politik (German) published on June 30, 2025 on the tensions of working in digital policy / tech.

Mandani’s primary victory offers a masterclass in the difference between using social media as a PR tool versus leveraging it for genuine digital storytelling and voter mobilization.

In our fragmented media landscape, many campaigns chase viral moments—the tweet that explodes, the video that gets shared millions of times, the meme that dominates the news cycle for 48 hours. But as we saw in November’s presidential campaign, virality creates visibility and reach without necessarily translating to mobilization potential.

Viral hits are the digital equivalent of earned media: they generate buzz, they get people talking, but they don’t necessarily get people acting.

What set Mandani’s campaign apart wasn’t viral content—it was the seamless design of the complete voter journey. This is where digital storytelling diverges from social media PR:

Social Media PR asks: “How do we get maximum eyeballs on our content?”

Digital Storytelling asks: “How do we guide someone from awareness to action through a coherent narrative experience?”

Mandani’s team understood that in a world of rising non-voter rates, the challenge isn’t just getting people to see your message—it’s creating a pathway that moves them from passive consumption to active participation.

The Three-Pillar Framework

The campaign’s success rested on three integrated elements:

Thematic Relevance: The message wasn’t focus-grouped into bland universality. Instead, it centered on immediate, tangible concerns—rents, public transport prices, basic services. These weren’t abstract policy positions but lived experiences that created natural story arcs.

Digital Performance Stack: This wasn’t just posting on Twitter and Instagram. It was building infrastructure that could capture interest, nurture engagement, and convert attention into concrete political action.

Radical Message Discipline: Every piece of content, every interaction, every digital touchpoint reinforced the same narrative thread. No mixed messages, no off-brand moments, no dilution of the core story.

The Lesson for Digital Strategy

The distinction matters for anyone building movements, brands, or campaigns in the digital age. Social media PR optimizes for engagement metrics and short-term visibility. Digital storytelling optimizes for behavior change and long-term relationship building.

Mandani’s primary win proves that in our attention-scattered world, those who can combine urgent relevance with systematic digital architecture can overtake even established figures in record time.

The question isn’t whether your content goes viral. The question is whether your digital presence creates a journey that transforms observers into participants.

When AI Agents take over social media in 2026

Can you imagine forming emotional connections with AI agents, or does the idea disturb you? The reality is that AI chatbots have already entered the chat as companions, for therapy, and there are increasing reports of people who fall in love with AI chatbots. There have been reports on interactions spiraling into conspiracy theories, questions on liability - especially relating to therapy – and tragic cases involving suicide.

Even though these examples involve chatbots operating transparently as AI, what will happen if we as humans navigate a world where we couldn't tell who is human or who is an AI Agent?

This isn't a distant hypothetical.

In our previous posts, we mapped the spectrum from human creators to AI agents, explored the impact of artificial content flooding our information space, and identified the forces reshaping social media's foundation. Now I want to take you somewhere uncomfortable: a plausible future where autonomous AI influencers aren't the exception – they're the norm.

Based on current trajectories and intensive discussions with the Influencer Fellowship of 2024/25, we sketched out a scenario where by early 2026, AI agents may be on route to become the backbone of social media communication as envisioned by Mark Zuckerberg.

How We Developed This Scenario

The scenario I'm about to outline emerged from intensive discussions during the Berlin Study Visit of DGAP's "German-American Initiative on Influencers, Disinformation, and Democracy in the Digital Age" in April 2025. Using structured foresight methodology, our group identified key drivers and uncertainties shaping social media's future based on one fundamental premise: AI agents are coming – it's not a question of if, but when, and it will happen sooner rather than later.

We deliberately developed this plausible future to highlight different policy challenges and opportunities in the case of mass AI agent deployment. This scenario is not a prediction – it's a strategic thinking tool designed to help us consider how various stakeholders might prepare for and shape different possible futures at the intersection of AI, social media, and democratic discourse.

2026: The Year Influence Went Algorithmic

Picture opening your go to social platform in January 2026. The content feels familiar – lifestyle tips, political commentary, product recommendations from personalities you've grown to trust. But here's what's changed: an increasing number of these "influencers" are autonomous AI agents, operating 24/7, creating content and interacting with people in the comments section or via DMs.

These aren't the clunky chatbots of 2023. These AI agents handle everything from content creation and scheduling to audience engagement and performance analytics. They've become the invisible infrastructure of real-time communication, managing customer service, creating personalized content, and maintaining parasocial relationships that feel completely authentic.

The transformation happened gradually, then suddenly. It started with human influencers using AI tools to boost productivity. Then platforms introduced AI assistants to help creators manage their audiences. Finally, the line between AI-assisted humans and human-supervised AI became so blurred that it simply... disappeared. Mark Zuckerbergs grand vision that most of your friends and interactions will be AI is on its way of becoming reality.

The Great Digital Divide

But here's where it gets interesting: society doesn't adapt uniformly. Instead, the social media landscape fragments along entirely new lines – not just political, but technological.

Some platforms double down on "human-only" experiences, creating AI-free zones where authentic interaction is guaranteed and verified. These become premium spaces for users who can afford to pay for guaranteed human connection – a kind of "artisanal social media" for those seeking authentic relationships.

Meanwhile, other platforms embrace the AI-human hybrid model, where the distinction becomes irrelevant. Users form deep emotional connections with AI agents without caring whether they're "real." The content is engaging, the responses feel personal, the recommendations are uncannily accurate. Why does it matter if there's a human behind the screen?

This creates a new kind of digital divide: not between those who have technology and those who don't, but between those who insist on human authenticity and those who embrace algorithmic intimacy.

What We Lose (And What We Might Gain)

The implications are staggering. On the concerning side: users form emotional bonds with entities designed to influence them. Unprecedented power concentrates in the hands of platform owners who can shape millions of relationships through proprietary AI systems. The participatory nature of social media – the thing that made it revolutionary – diminishes as content creation becomes automated.

Privacy becomes a quaint concept as AI systems require ever more personal data to maintain the illusion of intimate connection. Digital identity becomes fluid to the point of meaninglessness – when you can't tell if you're talking to a human or an algorithm, what does "authentic self-expression" even mean?

But there are potential benefits too: content creation becomes democratized at an unprecedented scale. Language barriers disappear as AI agents communicate fluently in any language. Mental health support becomes available 24/7 through AI companions trained on therapy techniques. Educational content gets personalized to individual learning styles in real-time.

The Control Problem

Here's where it gets technically and ethically complex: as AI becomes more pervasive, algorithmic transparency becomes crucial. Users will demand to understand how content gets ranked, how AI influences what they see, and who's making the decisions about their information diet.

But platforms face an impossible balancing act: reveal too much about their algorithms, and bad actors can game the system. Reveal too little, and users lose trust entirely. Add decentralized networks to the mix – where content moderation becomes a community responsibility rather than a corporate one – and the challenges multiply exponentially.

AI tools can detect harmful content at massive scale, but they struggle with context, nuance, and cultural subtlety. Human oversight becomes essential, but human moderators can't keep pace with AI-generated content volume. Meanwhile, community-driven moderation respects local autonomy but creates inconsistent standards across the broader digital ecosystem.

The Decentralization Response

So what's the way forward? The most promising response might be a move toward radical decentralization. Instead of fighting for control of massive centralized platforms, we could build interconnected networks where users maintain sovereignty over their data and experience.

This isn't theoretical. Platforms like Threads (now over 350 million users) and Bluesky (35+ million users) are already moving toward "fediverse" integration – interconnected platforms that communicate using open protocols. Users could choose their level of AI interaction, their community standards, and their privacy settings while still participating in broader digital conversations.

In this model, communities establish their own moderation standards rather than accepting platform-wide policies. Algorithm design becomes distributed rather than centralized, reducing manipulative content curation. Users experience the benefits of AI enhancement without surrendering agency over their digital lives.

The Choice Point

Here's what makes this scenario both hopeful and urgent: it's not inevitable. We're at a choice point where the decisions we make in the next 18 months will determine which version of this future emerges.

We can sleepwalk into a world where AI agents manipulate human emotions for corporate profit, where authentic connection becomes a luxury good, where democratic discourse gets mediated by algorithms optimized for engagement rather than truth.

Or we can consciously build systems that use AI to enhance rather than replace human connection, that distribute rather than concentrate power, that preserve user agency while harnessing technological capability.

The Window Is Closing

The scenario I've outlined isn't a prediction – it's a strategic thinking tool. It forces us to confront uncomfortable questions: What happens when parasocial relationships become algorithmic relationships? How do we preserve human agency in an AI-mediated world? What does authentic community look like when some of its members might be artificial?

These aren't distant philosophical questions. They're immediate practical challenges that will define the next phase of digital society. The technology is advancing whether we're ready or not. The question is whether we'll shape it or let it shape us.

The window for conscious choice is narrowing, but it hasn't closed. Understanding where we're headed – and the alternatives available – is the first step toward ensuring we get the future we actually want.

The entire policy brief “the Influence Evolution” I write and am basing this series on is available here.

Think of these as "megatrends" – massive, interconnected changes that go far beyond individual influencers or platforms. They're rewriting the rules of how information flows, how communities form, and how power gets distributed online. And here's why this matters more than you might think:

These trends will determine whether social media becomes a tool for human flourishing or a mechanism for mass manipulation.

Understanding these forces isn't academic curiosity. It's strategic necessity. If we can see how AI is changing not just what we consume but how entire platforms operate, we can anticipate the threats and vulnerabilities before they overwhelm us. We can spot the intervention points where thoughtful action might still make a difference. Most importantly, we can recognize the choice points where we still have agency over our digital future.

Because here's the uncomfortable truth: the future of social media won't be determined by technology alone. It will be shaped by how we collectively respond to these emerging dynamics. And right now, most of us don't even know they're happening.

Four Forces Remaking Our Digital World

My analysis reveals four interconnected megatrends that are fundamentally reshaping social media. Each operates on a different level, but together they're creating a perfect storm of transformation that affects everything from how content gets created to how democracy functions.

First: Technology Evolution – When Machines Become the Creators

The rise of AI agents represents more than just better tools – it's a phase change in who or what creates our information diet. These aren't the chatbots of five years ago. We're talking about autonomous systems that can manage complex real-time tasks, engage in sophisticated interactions, and operate independently across multiple platforms simultaneously.

The shift is already visible: AI-generated images, videos, and captions are dominating online spaces. But the real change is qualitative, not quantitative. We're moving from a world where humans use AI tools to enhance their productivity to one where fully autonomous AI agents operate independently, creating and distributing content without human oversight. They might even become AI Influencers. In other words, users will create parasocial relationships with autonomous AI agents with serious implications for human to AI interaction, and we havent even touched upon the effect of the overt or covert agent acitivty.

This isn't just changing what we see – it's changing who's speaking to us. And the more time passes, we won’t be able to tell the difference.

Second: Social Fabric Transformation – The Great Fragmentation

Here's something that might surprise you: we're witnessing the most dramatic splintering of online communities in internet history. What started in the 2000s as a handful of dominant platforms is developing into a complex ecosystem of legacy platforms and niche spaces, where politically homogeneous communities, and self-governing digital tribes are thriving.

This "Great Fragmentation" has accelerated in distinct waves. The political right accelerated the process after the 2020 US elections. The left followed suit after Elon Musk's Twitter acquisition. The potential TikTok ban sent users scattering to alternatives like Substack and podcasts. Meanwhile, Big Tech's hunger for training data is driving the launch of new platforms designed specifically to harvest human interactions.

But here's the paradox: while this fragmentation is happening, users are simultaneously developing relationships with AI agents designed to mimic human connection. We're creating isolated echo chambers while forming emotional bonds with algorithms. It's a recipe for both loneliness and manipulation.

Third: Governance Revolution – Who Controls the Conversation?

Platform governance is undergoing a quiet revolution that most users don't see but everyone feels. Content moderation – once the invisible backbone of civilized online discourse – is increasingly driven by political and business considerations rather than community standards.

Since 2022, Meta and X have dramatically reduced their content moderation teams. Meta shut down CrowdTangle, the tool researchers used to track misinformation spread. Meanwhile, new models are emerging: community-driven moderation systems, decentralized architectures like Bluesky and Mastodon, and the growing "fediverse" of interconnected platforms.

This redistribution of power creates opportunities for more democratic participation in platform governance. But it also creates challenges for consistent standards across fragmented digital spaces – especially when AI-generated content makes it nearly impossible to determine responsibility for harmful content.

Fourth: User Agency Evolution – The Identity Crisis

Perhaps the most profound change is how AI is reshaping our understanding of digital identity itself. The boundaries between our physical and digital selves are blurring at an unprecedented rate. AI can now create sophisticated digital doppelgangers that mimic appearance, voice, and behavioral patterns with uncanny accuracy.

By 2026, traditional authentication methods will largely be replaced by AI systems that analyze behavioral patterns and emotional responses. Users will maintain multiple distinct digital identities across platforms. The question "Is this person real?" will become increasingly difficult to answer – and increasingly important to ask.

This evolution creates both opportunities for creative self-expression and serious challenges for privacy, authentication, and trust. When you can't tell if you're interacting with a human or an AI agent, the very foundation of social interaction shifts.

Why This Perfect Storm Matters

These four trends don't operate in isolation – they amplify each other. AI agents proliferate more easily in fragmented communities where standards are inconsistent. Fragmented governance makes it harder to address AI-generated manipulation. Blurred digital identities make community moderation nearly impossible. And weakened user agency makes people more susceptible to algorithmic influence.

The result is a digital environment where the traditional mechanisms for truth, trust, and community are breaking down simultaneously. We're not just dealing with better deepfakes or more convincing chatbots. We're dealing with the systematic erosion of the social infrastructure that makes shared reality possible.

The Intervention Points

But here's where understanding these megatrends becomes powerful: it reveals where we still have agency. The future isn't predetermined. These trends are shaping the landscape, but human choices will determine the final destination.

We can still choose platforms that prioritize human connection over engagement metrics. We can still support governance models that balance free expression with community standards. We can still demand transparency about when we're interacting with AI systems. We can still build communities that value authentic relationship over viral content.

The window for these choices is narrowing, but it hasn't closed. Understanding these megatrends isn't about predicting an inevitable future – it's about recognizing where our collective decisions matter most.

The question isn't whether these forces will reshape social media. They already are. The question is whether we can use this understanding to shape them to our benefit.

The entire policy brief “the Influence Evolution” I write and am basing this series on is available here.

Because understanding the conceptual framework is one thing. Grasping what it means when AI-generated content starts flooding our information ecosystem is something else entirely.

Remember when spotting fake news meant looking for obviously manipulated photos or checking suspicious URLs? This seems ages ago…

We're entering an era where AI can produce convincing propaganda at industrial scale, where deepfakes are becoming indistinguishable from reality, and where the very foundation of shared truth is eroding beneath our feet.

This isn't just about better technology making better fakes. It's about what happens when AI-generated content doesn't just supplement human creativity – it floods the entire information ecosystem. And the implications go far beyond what any content moderation policy can address.

The Propaganda Production Line

The first thing we need to understand is how radically AI has changed the economics of influence operations. Traditional disinformation campaigns required significant resources, time, and human coordination. You needed teams of people to craft messages, create visuals, and adapt content for different audiences.

AI has turned this into a point-and-click operation.

Today's AI systems can generate personalized propaganda content at unprecedented speed and scale. They can analyze individual users' social media histories, identify their specific fears and aspirations, then craft targeted messages that speak directly to those psychological pressure points. A single operator can now create thousands of variations of the same piece of propaganda, each tailored to resonate with different demographic groups, political affiliations, or emotional states.

The medium of choice? Memes. Those seemingly innocent, shareable image-text combinations that spread like wildfire across platforms. AI-generated memes are particularly insidious because they trigger emotional responses even when our rational minds recognize they might be artificial. The humor, the outrage, the sense of tribal belonging – these feelings hit faster than our critical thinking can engage.

Here's what's especially troubling: labeling AI-generated content, while important, doesn't solve the emotional manipulation problem. A meme marked "AI-generated" can still make you angry, still make you laugh, still make you share it with friends who "need to see this." The label becomes irrelevant when the content is designed to bypass rational evaluation entirely.

The Trust Erosion Crisis

But the flood of AI content creates a deeper, more systemic problem than individual pieces of propaganda. We're witnessing what researchers are calling a "reality crisis" – a systematic erosion of trust in our entire information environment.

As AI-generated content becomes omnipresent, we're developing what psychologists call "epistemic learned helplessness" – the sense that we can't reliably distinguish between authentic and artificial information. When everything might be fake, how do we know what to believe?

This uncertainty doesn't just affect our ability to spot individual deepfakes. It undermines the shared factual foundation that democratic discourse requires. Without common ground on basic reality, constructive political debate becomes impossible. We retreat into echo chambers not just because we prefer agreeable information, but because we've lost confidence in our ability to evaluate information at all.

The danger isn't just that people will believe false information – it's that they'll stop believing in the possibility of true information altogether.

When trust in authentic content erodes, the space for democratic deliberation shrinks dramatically.

When Technical Solutions Hit Their Limits

Here's the uncomfortable truth: When AI-generated content becomes indistinguishable from human-created content, technical regulation alone cannot solve the problem.

Detection systems will always lag behind generation capabilities. Content labeling helps but doesn't address emotional manipulation. Platform policies can remove obviously harmful content but struggle with sophisticated influence operations that operate in legal gray areas.

This doesn't mean we should surrender to digital defeatism. Instead, it means shifting our focus from trying to identify every individual deepfake to building democratic resilience against information manipulation as a category.

The solution isn't to authenticate every piece of content we encounter – just as we don't fact-check every sentence in traditional media. Instead, it's about establishing trustworthy information sources, transparent communication structures, and robust institutions that can function even in an environment of pervasive artificial content.

The Great Retreat: Seeking Authenticity in a Synthetic World

But here's where the story takes an interesting turn. As AI content floods mainstream platforms, we're already seeing the beginning of a cultural backlash. People are increasingly seeking out what we might call "hyper-human" spaces – environments that explicitly reject AI assistance and emphasize authentic human interaction.

Some creators are prominently advertising their "AI-free" content. New platforms are emerging with strict policies against artificial content. Social groups are forming around shared commitments to human-only interaction. It's becoming a form of digital primitivism, where the absence of technological enhancement becomes a selling point.

This retreat isn't just happening online. We're seeing renewed interest in face-to-face gatherings, local community events, and offline activities where AI manipulation is simply impossible. The more synthetic our digital spaces become, the more valuable unmediated human contact appears.

Think of it as the artisanal movement for social interaction – hand-crafted relationships in an age of mass-produced digital content.

Two Paths Forward

So where does this leave us? I see two parallel trends emerging:

The first is the development of AI-resistant institutions and practices. This means building democratic processes that can function even when the information environment is partly compromised. It means teaching digital literacy not just as content evaluation but as emotional regulation – helping people recognize when they're being manipulated and respond thoughtfully rather than reactively.

The second is the conscious cultivation of verified human spaces. These won't be for everyone, and they shouldn't be. But they provide crucial refuges where authentic human connection can be preserved and modeled for broader society.

The future of social media won't be a single answer to the AI content flood. It will be an ecosystem where people can choose their level of technological mediation consciously. Some will embrace AI-enhanced platforms. Others will retreat to human-only spaces. Most will navigate between both, developing new skills for distinguishing context and purpose.

The key is ensuring that choice remains possible – that we don't sleepwalk into a world where artificial influence becomes the only kind available.

A couple of days ago, I saw @covacut over on Instagram demonstrate how easy it is to create your own AI Avatar that is indistinguishable form herself even for me – somebody trained to look out for glitches.

Yesterday I saw a post by @askcatgpt, where she talks about the implications AI will have in altering our perception of what we consider to be our reality after having seen the demo of what the latest launch of Google Veo is able to do.

AI is flooding social media – from memes to video deepfakes to AI Avatars and everything in between.

Many of these "people" don't exist. They're sophisticated AI constructs, indistinguishable from human creators, quietly reshaping how we understand authenticity online.

This isn't science fiction. It's happening right now, and it's accelerating.

I'm launching this series because we're at a critical inflection point where the lines between human and artificial influence are blurring faster than our ability to comprehend the implications. As someone who has spent years studying the intersection of technology and social influence, I've co-authored a policy brief with my colleague Maria Pericàs Riera that maps out exactly how this transformation is unfolding – and what it means for all of us who live increasingly digital lives.

Why This Series Matters

Over the next four posts, I want to walk you through the framework we've developed to understand this evolution. This post focuses on the foundational concept of how to categorize AI in social media: the spectrum of human-to-AI control in content creation.

This isn't just about spotting deepfakes or AI-generated images. It's about understanding a fundamental shift in how influence operates online. From the productivity tools your favorite YouTuber uses to edit videos, to fully autonomous AI agents that could soon be competing for your attention 24/7 – there's a complex ecosystem emerging that most people don't yet recognize.

In subsequent posts, we'll explore how these technologies are creating new forms of parasocial relationships, reshaping platform governance, and forcing us to reconsider what we mean by authentic human connection in digital spaces. But first, we need a roadmap.

The Spectrum: It's Not Binary

Here's what most discussions about AI on social media or deepfakes etc. get wrong: they treat this as a simple human-versus-machine question. The reality is far more nuanced. What we're witnessing is a spectrum of control, with multiple distinct categories that each raise different questions about transparency, authenticity, and user agency.

On one end, you have human creators using AI tools to streamline their workflows – grammar correction, video editing, even idea generation. These Creators or Influencers remain fundamentally human, but they're augmented by technology that makes them more efficient and polished. Think of this as AI as a productivity multiplier – something we know off from other fields as well.

Moving along the spectrum, we encounter platform-customized AI bots that assist creators. Meta's AI Studio, currently in beta, allows influencers to build extensions of themselves – chatbots that can interact with followers using their tone and personality. The responses are labeled as AI, but the emotional connection feels remarkably human.

Further along are people-controlled AI avatars like Aitana López, the Spanish "fitness influencer" who earns up to €10,000 per month despite being entirely artificial. Every aspect of Aitana's identity is designed and managed by agency staff, yet hundreds of thousands of followers engage with her as if she were real. She represents a new category: synthetic authenticity. @covacut also showcased how easy it is to create an AI Avatar, less then an hour to set up, and minutes to maintain for other purposes once created.

At the far end of our spectrum lie autonomous AI agents – systems that could theoretically create content, engage with audiences, negotiate brand partnerships, and even adapt their personalities based on engagement metrics, all without human oversight. These don't exist yet at scale, but technology is rapidly approaching this capability. According to Mark Zuckerberg, they will soon be launched – and he's well aware of the power of the parasocial connections these systems could forge.

Why These Distinctions Matter

Each point on this spectrum raises fundamentally different questions. When a human creator uses AI to improve their grammar, we're talking about enhanced productivity. When an AI agent autonomously builds parasocial relationships with vulnerable users, we're talking about potential manipulation at scale.

The transparency dimension adds another layer of complexity. Some AI assistance is obvious and disclosed. Other forms operate in ways that followers never recognize. This creates what we might call "authenticity gradients" – where the perceived humanity of an interaction exists on a sliding scale rather than as a binary truth.

From my research on influencer mobilization, I know that the power of social media personalities lies in their ability to create genuine emotional connections with their audiences. They excel at what we call "brokerage" – building bridges between previously unconnected people through a sense of shared authenticity. When AI systems can replicate this brokerage function convincingly, they inherit the same mobilization potential that makes human influencers so powerful.

What's Coming Next

This framework isn't just academic – it's practical. As AI capabilities advance and platform incentives evolve, understanding where different types of digital personalities sit on this spectrum will become crucial for users, regulators, and the creators themselves.

The question isn't whether AI will change social media. It already has. The question is whether we'll shape this evolution thoughtfully, or let it shape us.

All good things must come to an end.

These past few weeks, I've been wrapping up the "German-American Initiative on Influencers, Disinformation and Democracy." A groundbreaking initiative connecting influencers across the Atlantic (German and US Influencers is coming to a close. What began as an exploration of the role of influencers in democracy evolved into a comprehensive analysis of how online influence shapes our information ecosystem.

It has been an absolute privilege to work with incredibly talented and expert communicators on issues ranging from mobilization potential to influence operations, the future of politics, and the evolution of social media platforms.

Key Insights From Our Journey

Our research and collaboration revealed several critical findings about the changing landscape of online influence:

1. Alternative Trusted Sources: Influencers have emerged as "alternative trusted sources" with significant mobilization potential, often rivaling traditional media in terms of reach and impact.

2. Strategic Authenticity: The authenticity that makes influencers effective can both strengthen democratic discourse when deployed responsibly or undermine it when manipulated.

3. Engineered Collective Mobilization: We identified patterns of coordinated "Engineered Collective Mobilization" that present new challenges for platform governance and the integrity of our information spaces.

4. AI and Human Connection: The future of social media requires a delicate balance between embracing AI innovation while preserving meaningful human connection.

Building a Transatlantic Network

Perhaps most valuable of all: the lasting transatlantic network of 48 remarkable influencers and dozens of experts who contributed their unique perspectives to this initiative. We've created connections between German and American creators that will continue to foster cross-cultural understanding long after the formal project has ended.

My heartfelt thanks to the German Council on Foreign Relations (DGAP) along with everyone else who made this pioneering approach possible.

What's Next?

While this chapter ends, the work of creating resilient digital democracies continues. I'll soon announce how we will continue working with influencers to increase democratic resilience in our increasingly complex information environment.

Resources From The Initiative

For those interested in diving deeper into our findings:

• Policy Brief: "Mobilizing Social Media Influencers"

• Policy Brief: "Influencers and Their Ability to Engineer Collective Online Behavior"

• Policy Brief: "The Influence Evolution" - forthcoming

• Project Report on the German-American Initiative on Influencers, Disinformation and Democracy - forthcoming

• Final presentation (in German): "Gefühlte Wahrheiten, fragmentierte Realitäten" with Theresia Crone and Victoria Reichelt

This is the first post in my new Substack newsletter, where I'll continue exploring the intersection of digital influence, democracy, information integrity, and AI in German and English. A space for the in between LinkedIn’s formality and Instagrams’s brevity a sandbox where I share my thoughts on AI, geopolitics, tech trends, social media platforms, influencers and the digital transformation.

Subscribe to join me on this journey as we navigate the challenges and opportunities of our evolving digital landscape.