Theoretical Framework
Understanding AI Limitations Through Foundational Concepts
Overview
The AI Research & Philosophy framework provides a principled theoretical structure for understanding AI capabilities and limitations. Rather than treating AI failures as bugs to be fixed, the framework identifies categorical boundaries inherent to the architecture of current AI systems.
Core Framework Papers
AIDK Framework
The complete theoretical framework establishing structural epistemic limitations in Large Language Models. Introduces SEEE, AIDK, IDKE, and MAPT concepts.
Read FrameworkHCAE Deployment Model
Human-Curated, AI-Enabled: A tiered approach to AI deployment based on epistemic authority and verification requirements.
Read FrameworkResearch Program
The full theoretical program investigating AI through the origination-derivation lens. Includes research agenda, methodology, and open questions.
Read ProgramKey Concepts
Origination vs. Derivation
The fundamental categorical divide:
- Origination: Capacity to access reality, render judgments, set purposes, evaluate truth. Contact with what is.
- Derivation: Transformation of inputs according to learned patterns. Operations on representations producing representations.
These differ in kind, not degree. No amount of derivation produces origination.
SEEE: The Originating Error
Sentience Emergence Expectations Error: The categorical error of expecting sentience, consciousness, or understanding to emerge from systems whose mechanism (inductive symbol correlation) is not on the same ontological continuum as the expected outcome.
AIDK: The Structural Condition
AI Dunning-Kruger: The structural epistemic limitation of AI systems. Unlike human Dunning-Kruger, which is developmental and correctable, AIDK is architectural and permanent.
IDKE: The Interaction Effect
Interactive Dunning-Kruger Effect: When AI epistemic limitations meet human epistemic limitations, producing confidence amplification untethered from warrant.
MAPT: The Security Frame
Model Advanced Persistent Threat: Security framing for AIDK - treating structural AI limitations as an ongoing threat requiring continuous mitigation.
HCAE: The Deployment Model
Human-Curated, AI-Enabled: Framework for appropriate AI deployment based on epistemic authority requirements.
Connection to Logic Realism Theory
The AIDK framework shares foundational primitives with Logic Realism Theory:
- $I_\infty$: Infinite Information Space - all possible configurations
- $L_3$: The three classical logical laws as ontological constraints
- $A_\Omega$: The $L_3$-admissible subset that can be physically instantiated
Human cognition has access to both $I_\infty$ and $L_3$. AI systems are confined to derivatives of human-generated data, operating downstream of these primitives without direct access.
Archives
- Zenodo: AIDK Framework - Persistent DOI-minted archive
- GitHub: inference-stack - Source code and development