Framing Latour: Immutable Objects, Fixity, and Mobilization | Governance study group

Feb 27, 2026

Sharing our experimental call summaries.
Al-generated digests of Yak Collective study groups.

Core concept: immutable objects enable fast mobilization of information
Theory holds up for LLMs in some contexts
- One participant: LLMs could create immutable objects beyond human capacity
- Single video card computes more than all human lifetimes combined
- Common law vs Napoleon code analogy - higher compute costs for case-by-case analysis
A participant’s counterpoints on limitations
- Many experiences can’t be reduced to 2D surfaces (smell, medical conditions)
- Physical modeling still required (Toronto wind tunnel tests with scaled fluids)
- A mathematician’s LLM work: quantity over quality, but useful for tedious proofs

A participant’s analysis: model already failed pre-LLM for complex phenomena
Three failure examples:
1. Climate modeling - politics determine which dimensions to project
2. Economic models - Keynesian vs supply-side projections from same data
3. Disease surveillance - COVID showed vulnerability to conspiracy theories
Trade-off triangle framework:
- Stability of mobile inscriptions
- Inferential depth (skill required to participate)
- Ontological familiarity (how alien the concepts are to laypeople)
Scientific journals still require 2D descriptions despite working with complex models

A participant: “Kumbaya era of LLMs” ending with military applications
Immutability depends on institutional trust - without it, you get memes instead
High mobility without immutability (example: viral AI market manipulation report)
Prediction markets as response to lack of immutability
Question raised: Are we at peak LLM performance due to garbage data feedback loops?