TLDR: The Situated Epistemic Infrastructures (SEI) framework, introduced by Matthew Kelly, analyzes how knowledge gains authority in an era destabilized by Large Language Models (LLMs). It moves beyond traditional notions of ‘epistemic coherence’ to focus on ‘post-coherence’ conditions, where knowledge is mediated by hybrid human-machine systems. SEI identifies four core components (material coordination, institutional embedding, symbolic mediation, governance) and six infrastructure types (ritual, institutional, communicative, algorithmic, archival, experimental). The framework highlights how LLMs cause ‘infrastructural convergence’, disrupt ‘power signatures’, stress ‘symbolic compressions’, and make ‘breakdowns’ visible, ultimately revealing that legitimacy now stems from algorithmic-human coordination and demanding ‘infrastructural literacy’ for navigating this new, unstable knowledge environment.
In an era where Artificial Intelligence, particularly Large Language Models (LLMs) like ChatGPT, is rapidly transforming how we access and create information, our traditional understanding of knowledge is facing a profound challenge. A new research paper, “Situated Epistemic Infrastructures: A Diagnostic Framework for Post-Coherence Knowledge” by Matthew Kelly, introduces a groundbreaking framework to help us understand this shift.
The Shifting Landscape of Knowledge
For decades, we’ve largely operated under the assumption that knowledge is organized into stable, coherent systems, defined by shared vocabularies, disciplinary boundaries, and established standards of evidence. Think of peer-reviewed journals, academic conferences, and libraries – these have been the pillars of what the paper calls ‘epistemic coherence’. However, the rise of LLMs has exposed the fragility of these systems. These AI models generate seemingly authoritative responses from vast, often opaque, training data, bypassing traditional citation practices and even competing with human experts. This isn’t just a minor disruption; it’s a fundamental destabilization, leading to what the paper terms the ‘post-coherence condition’.
Introducing the Situated Epistemic Infrastructures (SEI) Framework
The SEI framework offers a vital lens to analyze how knowledge gains authority in this new, unstable environment. Instead of assuming stable communities, SEI focuses on how credibility is built across ‘hybrid systems’ – a mix of human and machine, symbolic and material elements. It acknowledges that knowledge now travels through diverse ‘infrastructures’ like search engines, chatbots, and academic databases, each with its own biases and ways of claiming authority. The framework also embraces ‘temporal ethics’, recognizing that our analytical tools must evolve and adapt, rather than claiming permanent authority.
Four Pillars of SEI: How Knowledge is Organized Today
The SEI framework identifies four core mechanisms that organize knowledge in our post-coherence world:
-
Material Coordination: This refers to the technical systems – databases, algorithms, and repositories – that store, search, and provide access to knowledge. These systems fundamentally shape what information is available and how it’s found.
-
Institutional Embedding: Knowledge is also shaped by organizational routines, funding models, and incentive structures within institutions like universities and journals. These dictate who produces knowledge and under what conditions.
-
Symbolic Mediation: We rely on ‘symbolic compressions’ – terms like “peer-reviewed” or “evidence-based” – that act as shortcuts to convey trust and quality. These symbols help us navigate vast amounts of information but can also obscure underlying complexities.
-
Governance and Normativity: Implicit norms about rigor, openness, and transparency are embedded in the design of our knowledge infrastructures. These norms subtly guide what is considered acceptable or incentivized.
Understanding Authority in a Hybrid World
The paper further breaks down how authority operates through ‘power signatures’ and ‘infrastructure typology’. It identifies six types of knowledge infrastructures:
-
Ritual: Like peer review or academic conferences, where legitimacy comes from established procedures and shared community standards.
-
Institutional: Formal organizations such as universities and funding agencies that exert authority through rules and procedures.
-
Communicative: Social media, mailing lists, and preprint servers where credibility builds through repeated sharing and network effects.
-
Algorithmic: Search engines and AI models that shape knowledge circulation through automated, often opaque, logics.
-
Archival: Libraries and databases that stabilize knowledge over time by preserving and curating information.
-
Experimental: Ad-hoc arrangements like hackathons or crisis dashboards, where authority is provisional and based on effectiveness in solving novel problems.
Each of these types has a distinct ‘power signature’ – a way it claims and enforces authority. For instance, algorithmic systems gain authority through their computational scale and sophistication, often operating as ‘black boxes’ where the underlying logic is hidden from users. The paper highlights ‘theoretical vulnerability’ as a key aspect of genuine theoretical contribution, where concepts are developed to be generative even if their future interpretations diverge from original intentions.
The Challenge of Symbolic Compression
In our information-rich world, we rely heavily on ‘symbolic compression’ – condensing complex processes into simple markers of legitimacy. “Indexed in Scopus” or “peer-reviewed” are examples of ‘legitimation shortcuts’. However, this can lead to ‘semantic overload’ (when terms like “open science” mean different things to different groups) or ‘symbolic drift’ (when terms lose their original meaning but retain their legitimating power, as seen with some commercial “open access” claims).
Breakdown as a Diagnostic Opportunity
A crucial insight of SEI is that ‘breakdown’ – when infrastructures fail – is not just a problem but a diagnostic opportunity. Failures in technical systems, erosion of trust, or misalignments in temporal rhythms (like slow peer review versus rapid pandemic response) reveal hidden assumptions and dependencies. These breakdowns force us to adapt and innovate, leading to ‘computational adaptation’ and new hybrid human-AI arrangements.
LLMs: A Case Study in Epistemic Transformation
The paper uses LLMs as a prime example of this ongoing transformation. Their integration has created a ‘polytemporal computational zone’ where human-paced deliberation clashes with machine-speed processing. This leads to ‘infrastructural convergence’, where systems like search engines and academic databases become functionally coupled despite incompatible logics. LLMs disrupt traditional power signatures, as their simulated coherence bypasses established validation rituals. They also intensify symbolic compression, with terms like “AI-assisted” or “comprehensive analysis” becoming high-velocity tokens that can obscure opaque algorithmic synthesis.
Also Read:
- Mapping the Human-AI Knowledge Connection
- Mapping the Inner Workings of Belief: A New Graph-Based Approach
Towards a Reflexive Epistemic Future
Ultimately, the SEI framework shifts our focus from simply classifying knowledge to understanding how it is coordinated across diverse, often conflicting, infrastructures. It argues that legitimacy now emerges from ‘algorithmic-human coordination’ rather than computational autonomy. Breakdowns are normal and reveal computational limits, prompting better human-machine collaboration. Power signatures hybridize, and symbolic compressions are amplified, demanding ‘computational semantic stewardship’ to maintain meaning. Effective use of LLMs requires ‘infrastructural literacy’ – understanding their biases, limitations, and synthesis mechanisms.
This framework provides essential tools for navigating the complexities of knowledge in the 21st century, emphasizing ongoing adaptation, ethical flexibility, and the cultivation of reflexive, justice-oriented epistemic arrangements. For more details, you can read the full paper here.
