ConsciousnessAI 0.0.1

A theory-grounded architecture for evaluating consciousness-relevant indicators in AI systems. Does NOT claim system consciousness.

Consciousness-Indicator Architecture (CIA)

Scientific boundary: CIA measures theory-derived consciousness indicators and does not prove subjective experience, sentience, or phenomenal consciousness.

What this package is

CIA is a research-oriented Python package for evaluating consciousness-relevant architectural indicators in AI systems. It implements a modular cognitive simulation pipeline and produces structured scorecards on a 0-22 indicator scale.

This package is intended for AI safety research, computational cognitive science, and philosophy of mind workflows where transparent, testable proxy metrics are required.

Core capabilities

• 11 indicator categories grounded in established consciousness theories
• End-to-end cognitive cycle simulation
• Scorecard generation with risk-tier summaries
• Causal intervention harness for ablation-style experiments
• Governance and report-verification tools
• Optional LLM adapter layer (remote and local)
• Optional EEG/BCI neuroadaptive extension
• Optional plant biohybrid extension

Installation

From PyPI

pip install ConsciousnessAI

With extras

pip install "ConsciousnessAI[dev]"
pip install "ConsciousnessAI[dev,neuro,plant,llm,local-llm]"

Python requirement

• Python 3.11+

Command-line quickstart

cia run "A red object moved behind a screen and reappeared."
cia run "The system noticed its own processing limits." --verbose
cia score --input your_input.json
cia intervene --type disable_workspace

Python API quickstart

from cia.simulation import CombinedConsciousnessIndicatorSystem

system = CombinedConsciousnessIndicatorSystem(recurrent_cycles=3)
report = system.run_cycle("The agent reflected on its own uncertainty.")

print(report.indicator_scores.total_score, report.indicator_scores.max_possible)
print(report.welfare_state.risk_level)

Optional integrations

LLM

• Providers: OpenAI, Claude, Gemini, Hugging Face, llama.cpp, vLLM
• Config-driven adapter loading

Neuroadaptive EEG/BCI

• Offline ingestion and preprocessing
• Feature extraction and proxy-state conditioning

Boundary: EEG features are statistical proxies and do not read thoughts or infer subjective experience.

Plant biohybrid

• Non-invasive plant electrophysiology ingestion
• Signal-feature conditioning of CIA runtime controls

Boundary: plant signals are physiological proxies and not evidence of plant consciousness.

Documentation and reports

• Full documentation site: https://rotsl.github.io/cia/
• Full project documentation: docs directory in repository
• Validation and demo outputs: reports directory in repository
• Citation metadata: CITATION.cff

For full usage, benchmarks, governance policy, and scientific caveats, see the repository documentation site built with MkDocs.

License

MIT License