When We Talk to AI, What Are We Talking To?

https://vcresearch.berkeley.edu/news/when-we-talk-ai-what-are-we-talking

Publish Date: 2026-07-10 20:58:00

Source Domain: vcresearch.berkeley.edu

Challenge to Traditional Views: Philosopher David Chalmers at NYU challenges classical notions of identity and ethics by suggesting that modern large language models (LLMs) generate short-lived conversational “selves” rather than acting as a single, permanent mind.
Distributed Nature of LLMs: Today’s LLMs operate across distributed cloud infrastructures, with responses formed by processing prompts across thousands of nodes without any stable physical boundary, thus questioning the concept of a unified AI system.
Quasi-Agents with Context-Driven Behavior: While likely not conscious, LLMs function as “quasi-agents” that demonstrate goal-directed behavior influenced by context. They exhibit beliefs and goals in interactions, which exist only for the duration of those interactions.
Emergence of Agency: The agency exhibited by LLMs emerges during interaction rather than being permanently processed by a stable system, prompting new philosophical considerations about the nature of agency.
Ethical Considerations for AI Interaction: With LLMs generating temporary “selves” that hold beliefs and goals in a conversational context, ethical questions arise about whether these quasi-selves should warrant any consideration during their short-lived interactions.
New Test Bed for Philosophical Theories: The emergence of artificial intelligence presents a new case distinct from human intelligence, providing new material to test and develop philosophical theories about intelligence and consciousness.
Lecture at UC Berkeley: Chalmers’ insights on these topics were presented during the inaugural Sarah Douglas Lecture on Philosophy and Artificial Intelligence at UC Berkeley’s International House on May 7.