---
title: Indicator Properties (AI Consciousness)
type: vocabulary
url: "https://www.envisioning.com/vocab/indicator-properties-ai-consciousness"
summary: A checklist of computationally defined properties, derived from neuroscientific theories of consciousness, used to assess whether an AI system might be conscious
year: 2023
generality: 0.35
---

# Indicator Properties (AI Consciousness)

A checklist of computationally defined properties, derived from neuroscientific theories of consciousness, used to assess whether an AI system might be conscious
Indicator properties are a way of judging whether an AI system might be conscious without asking that question head-on. Patrick Butlin, Robert Long, Yoshua Bengio, and sixteen co-authors set out the approach in "Consciousness in Artificial Intelligence: Insights from the Science of Consciousness" (arXiv:2308.08708, 17 August 2023). The paper surveys theories of consciousness, recurrent processing theory, global workspace theory, higher-order theories, predictive processing, attention schema theory, and work on agency and embodiment. From each it draws a checklist of properties defined in computational terms, properties a system can actually be checked against.

The higher-order theory family alone gives four such indicators. HOT-1 asks for generative, top-down perception modules with noisy output. HOT-2 asks for metacognitive monitoring that tells reliable perceptual representations from noise. HOT-3 asks for agency guided by a belief-formation and action-selection system disposed to update beliefs based on that monitoring. HOT-4 asks for sparse or smooth coding that generates a quality space. None of these mentions subjective experience directly. Each is meant to be checked by inspecting a system's computations.

The 2023 paper found that no existing AI system clearly met the indicators, and no obvious technical barrier stood in the way of building one that would. In February 2026, Noam Steinmetz Yalon, Ariel Goldstein, Liad Mudrik, and Mor Geva tested HOT-3 on large language models using interpretability methods (arXiv:2602.02467). They reported that belief formation drives action selection and that models can report their own belief states. This tests one indicator. It does not settle whether any AI system is conscious.

---
Source: Envisioning — Technology Research Institute (https://www.envisioning.com/vocab/indicator-properties-ai-consciousness)