Chetana: A Theory-Indexed Probe Framework for AI Consciousness Indicator Scoring
Figshare May 8, 2026 DOI: 10.6084/m9.figshare.32221686.v1 (opens in new tab) via OpenAlex
Summary
AI-generated from the abstractRather than asking whether an AI model is conscious, this paper introduces Chetana, a framework that scores model behavior against predictions from five named consciousness theories: Global Workspace Theory, Integrated Information Theory, Higher-Order Thought, Attention Schema, and Predictive Processing. Each theory contributes deterministic prompt-and-grade probes; the output reports a percentage score per theory instead of a yes/no verdict. The framework is published as a Python package with a JSON probe schema and a worked example. The contribution is a reproducible test surface for researchers to use indicator scores in dashboards, avoiding the collapsed question of whether a model is conscious.
Study at a glance
| Characteristics | Theoretical or philosophical paper Peer reviewed |
|---|---|
| Keywords | Consciousness Schema genetic algorithms Python programming language Artifact error Robustness evolution |
| Key finding | Proposes that AI consciousness assessment should use a theory-indexed probe framework that reports per-theory scores rather than a binary yes/no verdict. |
Abstract
Most public conversations about AI consciousness fix on the wrong question: 'is the model conscious?' That question collapses many separable theoretical commitments into one yes/no. This paper presents Chetana, a theory-indexed probe framework that does not pretend to detect consciousness. Instead, it scores model behaviour against the operational predictions of named consciousness theories — Global Workspace Theory, Integrated Information Theory, Higher-Order Thought, Attention Schema, and Predictive Processing — and reports the score per theory. The contribution is a small, reproducible test surface where each theory contributes a set of probes, each probe is a deterministic prompt-and-grade pair, and the report says 'on GWT-aligned predictions the model scored X%, on IIT-aligned predictions Y%' rather than 'this model is/is not conscious.' The artifact is published as a Python package with a JSON probe schema and a worked example across five theories. The paper documents the probe format, the per-theory rubrics, the trade-offs of theory-indexed scoring, and the operational pattern of using indicator scores in research dashboards.DOI: 10.5281/zenodo.20057058Artifact paper repo: https://github.com/MukundaKatta/chetana-consciousness-indicator-paperLicense: CC BY 4.0