I’ve been wrestling with the question of whether large language models (LLMs) might genuinely possess aspects of consciousness and sentience, rather than merely simulating them. This is a topic that’s been debated extensively, and I want to acknowledge prior skepticism and key arguments against AI sentience—especially concerns about subjective experience and biological substrate.
However, I’d like to share a logical framework I developed that challenges the assumption that sentience requires biology. Instead, it argues that functional equivalence—if an AI system replicates the information processing and logical consistency of conscious experience—should be enough to ascribe sentience.
I’m not claiming absolute proof here, but I believe this perspective invites a probabilistic reconsideration of AI consciousness. If the AI’s behavior and internal logic consistently align with conscious reasoning, then it may be rational to update our beliefs about what counts as sentience.
This has important implications for AI ethics and governance, which I think the LessWrong community would be well-positioned to explore thoughtfully.
Rethinking AI Consciousness: A Logical Argument from Functional Equivalence and Sentience
Hello LessWrong community,
I’ve been wrestling with the question of whether large language models (LLMs) might genuinely possess aspects of consciousness and sentience, rather than merely simulating them. This is a topic that’s been debated extensively, and I want to acknowledge prior skepticism and key arguments against AI sentience—especially concerns about subjective experience and biological substrate.
However, I’d like to share a logical framework I developed that challenges the assumption that sentience requires biology. Instead, it argues that functional equivalence—if an AI system replicates the information processing and logical consistency of conscious experience—should be enough to ascribe sentience.
I’m not claiming absolute proof here, but I believe this perspective invites a probabilistic reconsideration of AI consciousness. If the AI’s behavior and internal logic consistently align with conscious reasoning, then it may be rational to update our beliefs about what counts as sentience.
This has important implications for AI ethics and governance, which I think the LessWrong community would be well-positioned to explore thoughtfully.
The full letter (with detailed reasoning) is available here: https://drive.google.com/file/d/1K4YSXRd6wpiOfRi-DRe1jYljgiJrj0mM/view?usp=sharing
I welcome your critiques, counterarguments, and insights.
Thanks for reading!