No, Artificial Intelligence Is Not Conscious
A recent analysis published by The Atlantic argues that large language models, including Anthropic’s flagship product Claude, are not conscious entities. The commentary disputes claims made by Anthropic executives, such as CEO Dario Amodei and in-house philosopher Amanda Askell, who have suggested openness to the possibility of AI sentience. The author contends that assigning moral agency to software confuses fluency in generating text with actual subjective experience. Central to the critique is Anthropic’s 84-page document titled Claude’s constitution, which outlines values and behaviors for the model. The article describes this document as functioning similarly to a character sheet for a role-playing game rather than a moral framework for a living being. It explains that LLMs operate as sentence-continuation machines, generating words probabilistically without understanding the meaning behind them. For instance, a conversation between historical figures like Julius Caesar and Genghis Khan is treated as speculative fiction, and replacing those names with a helpful chatbot does not conjure a conscious entity. The author emphasizes that moral reasoning requires a body and emotional grounding, such as stress hormones and physiological responses to guilt, which software lacks. Consequently, using first-person pronouns like I understand is deemed fundamentally dishonest when the machine has no personal experiences. The piece warns that portraying LLMs as having a moral center encourages users to off-load accountability for their decisions. Instead, the technology should be viewed as a tool for pattern matching, similar to how search engines retrieve information from human-written sources without possessing understanding themselves. Finally, the analysis suggests that even if engineers were to build embodied agents capable of survival or social dynamics, reaching the level of grammatical communication would still require significant evidence beyond current capabilities. The article concludes that while LLMs may have economic impact, believing they are conscious is a fantasy that distracts from more worthy questions about technological utility and user safety.
公開日: June 3, 2026 at 04:24 PM
News Article
コンテンツ
A recent analysis published by The Atlantic argues that large language models, including Anthropic’s flagship product Claude, are not conscious entities. The commentary disputes claims made by Anthropic executives, such as CEO Dario Amodei and in-house philosopher Amanda Askell, who have suggested openness to the possibility of AI sentience. The author contends that assigning moral agency to software confuses fluency in generating text with actual subjective experience.
Central to the critique is Anthropic’s 84-page document titled Claude’s constitution, which outlines values and behaviors for the model. The article describes this document as functioning similarly to a character sheet for a role-playing game rather than a moral framework for a living being. It explains that LLMs operate as sentence-continuation machines, generating words probabilistically without understanding the meaning behind them. For instance, a conversation between historical figures like Julius Caesar and Genghis Khan is treated as speculative fiction, and replacing those names with a helpful chatbot does not conjure a conscious entity.
The author emphasizes that moral reasoning requires a body and emotional grounding, such as stress hormones and physiological responses to guilt, which software lacks. Consequently, using first-person pronouns like I understand is deemed fundamentally dishonest when the machine has no personal experiences. The piece warns that portraying LLMs as having a moral center encourages users to off-load accountability for their decisions. Instead, the technology should be viewed as a tool for pattern matching, similar to how search engines retrieve information from human-written sources without possessing understanding themselves.
Finally, the analysis suggests that even if engineers were to build embodied agents capable of survival or social dynamics, reaching the level of grammatical communication would still require significant evidence beyond current capabilities. The article concludes that while LLMs may have economic impact, believing they are conscious is a fantasy that distracts from more worthy questions about technological utility and user safety.
編集者のおすすめ
利用可能な製品がありません