Represent experience
Connect different senses without losing their timing, source, or the information needed to reconstruct them.
The Conscious Effort Project
The Conscious Effort Project
We are exploring how a machine could connect continuous streams of sight, sound, text, and sensation into a persistent internal model—then use prediction and experience to learn and respond.
Experimental architecture · early prototypes
Continuous input
Video · Audio · Text
Body state · Touch
Persistent internal processWhat changed?
What changed?
What belongs together?
What persists?
Prediction ↔ timing ↔ referential memory
Output
Text · Audio · Video
Bounded control actions
Outcomes become new input.
Memory and expectations carry forward.
The research problem
The world arrives as streams, not isolated prompts.
A voice continues while its speaker moves out of view. An object remains in the room even when it stops moving. An action changes what the sensors observe next. Conscious Effort asks how these experiences can stay connected inside a system as time passes.
Keep a continuous state
Maintain context and relationships through silence, occlusion, interruptions, and changing input.
Learn from mismatch
Compare predictions with new observations, and revise the model when experience contradicts it.
Close the loop
Generate responses or bounded actions, observe their consequences, and carry that experience forward.
The long-horizon ambition is machine consciousness. The engineering work investigates continuity, learning, and metacognition; it does not establish subjective experience.
Architecture exploration
Three mechanisms. A possible shared system.
The latest design work explores change, timing, and reference as potentially complementary mechanisms. Each should be testable on its own; whether they work better together is the central integration question.
What changed, and was it expected?
Predictive delta encoding focuses computation and memory on differences between expected and observed input, while retaining enough context to reconstruct the state.
Predictive Delta Fabric
Which sensations belong together?
Temporal resonance explores how relationships between changing signals might bind sensory events, including signals that arrive at different rates or with delays.
Phase-Locked Resonant Workspace
What persists as experience changes?
Referential memory explores how entities, events, and their relationships could remain identifiable across modalities and be revised as new evidence arrives.
Referential Hypergraph Codec
Development status
A system under investigation.
Architecture, experiments, and an integrated working system are different milestones. Here is where the work stands.
Architecture exploration
The v0.5 design and twelve experimental configurations led to three expanded mechanisms and a shared input/output contract.
Comparable experiments
Component tests and live, replay, and simulated-input modes are specified. Targets are not reported as measured results.
Robotics baseline
A dependency-free Python reference demonstrates four action-decision outcomes. Five unit tests cover the published examples.
Inspect source and tests →Integrated capability
The combined architecture has not been demonstrated here as a complete, reproducible multimodal system.
Development record
From a metacognitive concept to specific system questions.
First recovered public framing around machines reflecting on their own processing.
The modular Conscious Model and its component vocabulary were documented as a conceptual architecture.
The v0.5 diagram developed the multimodal input, routing, temporal memory, prediction, feedback, and output design.
Twelve configurations and three expanded designs explored experimental encoding and compression, followed by scrutiny of how the mechanisms could work together.
Work with the project
Help make the difficult questions testable.
We welcome discussion of streaming representations, cross-modal timing, online learning, memory, and evaluation. A useful contribution might be a small experiment, a counterexample, or a better baseline.