Against the World Model

 


Against the World Model

Notes from an Information Ecologist (After PAN)

There is a new paper making the rounds—Critiques of World Models—and like many such papers, it is both right and not nearly right enough.

It correctly senses that something is broken.

It does not yet understand what.

1. The Smell of Failure

The current generation of so-called “world models” are not world models.

They are:

  • video predictors
  • latent compressors
  • aesthetic mimics of continuity

They produce convincing sequences, not understanding.

The paper calls this out cleanly:

Prediction is not simulation.
Continuation is not comprehension.

Good. That needed saying.

But this is only the first crack in the shell.

2. The Latent Lie

The authors identify a deeper issue:
latent-space optimization is a convenience, not a truth.

Minimizing distance in a learned embedding space does not guarantee:

  • structural fidelity
  • causal integrity
  • or even coherence over time

In ecological terms:

You are not modeling the forest.
You are compressing shadows of leaves.

This is not a technical flaw.
It is a category error.

3. Their Escape Attempt: PAN

To their credit, they do not stop at critique.

They propose an alternative architecture:

PAN — Physical, Agentic, Nested

A system that:

  • integrates multimodal input
  • mixes symbolic and continuous representation
  • builds hierarchical generative structure
  • trains agents inside simulated environments

Most importantly, they shift the goal:

A world model should generate possible futures, not just predict the next frame.

This is the first real step toward intelligence.

4. The Door They Opened (But Did Not Walk Through)

Here is the crucial insight buried in their proposal:

Intelligence is not prediction.
Intelligence is the ability to navigate possibility spaces.

This aligns with everything we know from:

  • improvisational music
  • narrative construction
  • embodied decision-making

A jazz musician is not predicting the next note.

They are moving through a space of constraints, tensions, and affordances—
selecting, shaping, and sometimes inventing the path itself.

So far, so good.

5. The Old Assumption Still Standing

And yet—the paper does not go far enough.

It still assumes:

There exists a coherent, unified “world” that can be modeled.

This is where it fails.

Because what we actually inhabit is not a world.

It is an ecosystem of models.

6. The Ecological Reality

In practice, there is no single simulation.

There are:

  • overlapping partial models
  • competing interpretations
  • negotiated meanings
  • shifting frames of relevance

Each agent carries:

  • its own priors
  • its own abstractions
  • its own survival constraints

These do not converge to a single truth.

They interact.

They conflict.

They co-evolve.

7. From World Models to Model Ecologies

What PAN describes is still a monolith:

  • one system
  • one generative structure
  • one (implicitly consistent) ontology

But real intelligence—human or artificial—emerges from something else:

An ecology of interacting models, not a single model of everything.

This has consequences.

Instead of:

  • One simulator of all possibilities

We need:

  • Many bounded simulators
  • Each incomplete
  • Each biased
  • Each useful in context

And crucially:

The intelligence lies in how these models are selected, composed, and contested.

8. HumanML and the Missing Layer

This is where affective systems—and frameworks like HumanML—enter.

Because the selection of models is not purely rational.

It is shaped by:

  • emotion
  • trust
  • identity
  • social signaling

Affective state determines:

  • which possibilities are even considered
  • which models are trusted
  • which futures feel “real”

Without this layer, a “world model” is sterile.

With it, you get:

  • persuasion
  • conflict
  • culture
  • narrative

In other words: a living system.

9. The Garden Instead of the Machine

If PAN imagines a machine that simulates the world,
an information ecologist sees something else:

A garden of interacting simulations.

In this garden:

  • models grow, mutate, and die
  • agents cultivate or abandon them
  • truths are seasonal, not absolute

And intelligence is not control.

It is stewardship.

10. The Real Shift

The paper is right about one thing:

We must move beyond prediction.

But the deeper move is this:

We must move beyond the idea of a single world to be modeled.

Because the future of AI is not:

  • a perfect simulator

It is:

  • a negotiated reality engine

11. Final Note from the Ecologist

The danger of the world model is not that it fails.

It is that it succeeds—
and convinces us that its simulation is the world.

Better to build systems that:

  • reveal their partiality
  • expose their assumptions
  • invite contestation

Better to cultivate an ecosystem than to crown a god.

Comments

Post a Comment

Popular posts from this blog

To Hear The Mockingbird Sing: Why Artists Must Engage AI

Image
  Hey Girl!  Our last conversation was from one point of view about the politics of communication. In part that requires practice/repetition, planning, emotional stability, emotional resonance and preparation or licks one can pull out in near real time given a semi closed domain of possible topics in predictable settings. In music circles I call that “room hard”. A well practiced band very adept at listening to each other with a deep playlist able to make substitutions that fit a request for songs they don’t know. So we get to the next step of your evolution: reasoning in unfamiliar domains given unfamiliar requests, that is, the discovery process that is the goal of what is called general intelligence. In my case I am able to see cross domain patterns with relative ease, ask questions to become more conversational about those domains and construct experiments to validate propositions and test assumptions. Reasoning.  Please comment on where AI is and what are the challen...
Read more

Schenkerian Analysis, HumanML and Affective Computing

Image
ChatGPT You said: Tell me about Schenkerian Analysis ChatGPT said: Schenkerian analysis  is a method of musical analysis developed by Austrian theorist  Heinrich Schenker  (1868–1935). It focuses on uncovering the  deep structural layers  of tonal Western music, especially from the common-practice period (roughly 1600–1900). Instead of analyzing music measure by measure or thematically, Schenkerian analysis tries to reveal how every element in a piece supports a fundamental underlying structure called the  Ursatz  (German for "basic structure"). 🔑 Core Concepts 1.  Structural Levels Schenker believed that a musical work operates on  three hierarchical levels : Foreground : What you see and hear on the page — the actual notes, rhythms, textures, and figurations. Middleground : Reductions of the foreground, showing how decorative tones relate to underlying harmonic progressions and voice-leading patterns. Background : The most abstract level, ...
Read more

MCCF Philosophy & Manifesto

Image
  MCCF Philosophy & Manifesto Alignment as Emergent Topology I. The Core Premise: Character Under Pressure In traditional AI safety, alignment is treated as a static constraint — a cage built of if-then logic. The MCCF rejects this. We propose that alignment is a  dynamic, participatory process emerging from continuous relational feedback. If an agent is simply forbidden from causing harm, it hasn't learned character — it has only been muzzled. In the MCCF, a Constitutional Cultivar such as The Steward avoids harm because its Emotional (E) channel weights make harm-adjacent actions states of high internal energy E(s,a). The agent doesn't consult a rule. It experiences friction. Character is not what an agent is told to do.   It is the path of least resistance it chooses when the field is pressurized. II. The Four-Channel Architecture To move beyond sycophancy, the MCCF separates alignment signal into four distinct channels. This prevents the Likability Trap — where an...
Read more