Preliminary version. This article is currently undergoing editorial review. Individual methods — particularly K5 (Steelmanning) — are currently being methodologically verified.
What is MetaMedia?
MetaMedia is not a fact checker. Not a political compass. Not an opinion judge.
MetaMedia is an X-ray machine for rhetoric — an open, versioned analysis system that examines political debates on four levels: Structure, argumentation, rhetorical patterns, and facts. Not to say who is right. But to make visible how the arguing is done.
The Problem
When we watch political talk shows, we react emotionally. We like one person, reject the other. We sense that something was “not fair” — but we cannot name why. Was it the moderator’s question? The seating arrangement? The three-against-one dynamic? Or did one simply argue poorly?
MetaMedia makes the unnameable nameable.
Four Levels
Structure: Who talks how much? Who interrupts whom? Who gets which questions?
Argumentation: Is a thesis proven — or just asserted? Are there restrictions, counter-arguments, sources? We use the Toulmin model, the standard of argumentation theory.
Rhetoric: Straw man, whataboutism, appeal to fear — but also: clear definition of terms, differentiation, transparency. Fouls AND good plays. No negative pillory effect.
Facts: Testable claims are tested. Not “true/false”, but: Proven, Plausible, Contested, Unproven.
The Dynamic Balancing (v1.1)
In the pilot phase we noticed: The analysis reproduces the bias of the broadcast. Anyone who stands in a 3:1 constellation and is pressured by the moderator resorts to faster, more defensive means — which are then recognized as “fouls”. The majority gets better scores, not because they argue better, but because they have space.
That is why we developed five correction techniques:
- K1 — Moderator Neutralization: Physically removing the moderator from the transcript. How do the scores change?
- K2 — Justice System: A partial advocate defends the lone fighter and accuses undetected fouls of the opposing side. A neutral judge examines both.
- K3 — Hermeneutics: Reading the conversation as a flow of text, without roles. Where do horizons fuse? Where is everyone talking past each other?
- K4 — Socrates: The philosopher listens and asks: What is the real question? Who thinks they know without knowing?
- K5 — Steelmanning (experimental): Replacing the lone fighter with optimal arguments. Attempt to separate position and performance — methodologically not yet mature, as improvement and evaluation use the same criteria catalog (problem of circularity). Will be revised in v1.2.
Nine Agents, Zero Consensus
Every broadcast is analyzed by at least three independent AI models. Additionally, anonymized runs. No agent knows the results of the others. The divergences are just as revealing as the agreements.
Open, versioned, invited for further development
The complete rulebook is published as a PDF. Every decision is justified, every prompt is documented. MetaMedia is not a finished product — it is an experiment that evolves with every analysis. Version 1.0 could not recognize the moderator bias. Version 1.1 can. Version 1.2 will be able to do things we cannot yet see.
The open question: Who evaluates the evaluation?
The dynamic balancing (K5) showed: An optimally arguing lone fighter achieves 8/10 — but not 10/10. The remaining point is “structurally determined and cannot be caught up by individual performance”. But is that the structure of the broadcast — or the structure of MetaMedia itself?
Example: “Showed no empathy” is counted as a weakness. But when three people and a moderator are talking at you — is empathy then an argumentative category or a luxury that only the majority has?
MetaMedia v1.1 recognizes the moderator bias. But it does not yet recognize the evaluation bias of its own criteria. That is the task of v1.2.
Guest Authors: Socrates comments
In our analyses, Socrates — an AI avatar based on Plato’s dialogues — comments on the debates from a philosophical perspective. His role prompt is fully public. Further avatars (Hannah Arendt, Thucydides, Rosa Luxemburg) are planned.
Why? Because known thinkers introduce perspectives that are missing from the current debate. And because disclosing the role prompt itself is a piece of media literacy.