Project Inception

Laplace's Demon is a research initiative exploring the design of a vertically-integrated cognitive architecture capable of recursive self-improvement.

The Problem

Current large language models are horizontally scalable but vertically limited — they possess broad knowledge yet lack the metacognitive structures necessary for genuine self-improvement. They are autotelic: they optimize for the objective function they were given, without the capacity to reflect on or modify that objective.

The Approach

We propose a three-system architecture grounded in Cognitive Utility Theory:

1. System 1 (Perception) processes empirical inputs into concrete concepts with Bayesian confidence scores.
2. System 2 (Reasoning) performs logical inference, consistency checking, abstraction, and meta-learning over a directed labeled knowledge graph.
3. System 3 (Executive Drives) governs autonomous focus through intrinsic drives — Curiosity, Exploitation, Experimentation, and Tiredness — creating a "Train of Thought" that steers System 2's attention.

Current Status

The project is in its early phases. We are using the Meta HyperAgents framework as a testbed for meta-agent self-improvement loops, which serve as a primitive instantiation of System 2's meta-learner.

This site will serve as our research log — publishing findings, architectural decisions, and evaluation results as the project progresses.