A Framework for Evaluating AI Consciousness Across Five Levels Proposed
- What
- The authors organized various theories dealing with AI consciousness into a five-tier hierarchical framework and proposed a method to evaluate current Large Language Models (LLMs) using indicators for each level.
- Key result
- Current LLM consciousness assessments varied widely from under 0.01 to about 0.8, depending not only on how evidence is interpreted but also on how much weight is placed on which theory.
- Why it matters
- Given that consciousness-related indicators heavily overlap with structures necessary for general intelligence, this implies that as AI becomes more capable, it may become a stronger candidate for consciousness. However, this is not a definitive conclusion, and the abstract makes no direct mention of investment.
Background
The issue of AI consciousness is considered one of the most urgent prior tasks in philosophy and computer science, but progress has been slow due to many conflicting theories. The authors suggest that by separating the fundamental question of how experience arises (the hard problem) from the problem of mapping system structures to theories (the mapping problem), we can examine the issue while setting aside deep metaphysical debates.
Method
Expanding on three-stage analyses, they created a five-level functional description hierarchy of behavior, computation, intrinsic causal structure, organism, and organism-environment. Major theories of consciousness were placed in this hierarchy based on which level they consider central, and current AI systems were evaluated by setting measurable indicators for each level. Afterward, a Bayesian model combined the confidence in theories and evidence from indicators to calculate an overall probability.
Main results
In the example evaluation, the verdict for current LLMs depended as much on where theory credences were placed as on how evidence was read. Varying assumptions and credence distributions resulted in broad outcomes ranging from under 0.01 to about 0.8. They also reported that consciousness indicators at each level largely overlap with structures required for general intelligence.
Limitations & open points
The abstract presents this as an example evaluation and does not disclose definitive scores or verified conclusions for specific models. Specific measurement methods for the indicators and the validity of placing theories into levels cannot be confirmed from the abstract alone.
Glossary
- Supervenience (support basis) — Refers to the relation where if the internal organization of a system is the same, the experiences are also determined accordingly.
- Coarse-graining — A method of omitting details and describing in larger units.
- Multiple realizability — The concept that the same function can be implemented by different materials or structures.
- Hard problem — The philosophical question of why physical information processing gives rise to subjective experience.
- Credence — The degree of conviction about the likelihood of a claim being true, expressed as a probability between 0 and 1.
- Bayesian model — A method of calculating probabilities by updating prior beliefs whenever new evidence emerges.
This summary was written by AI from the paper's abstract and reviewed by the editor. Summaries on this page were translated by AI from the editor-reviewed Korean version. Check the original for exact figures and context.
This page compiles published papers, official announcements and press reports for information only. Paper summaries are written by AI from the abstract and reviewed by the editor; they may differ from the original, so check the paper for anything important. Nothing here is a recommendation to buy or sell any security or investment advice. Career moves cite official announcements or major press reports; some career records come from Wikidata (CC0). Summaries on this page were translated by AI from the editor-reviewed Korean version.
