Debate erupts: Can LLMs be persisting interlocutors or just illusions?
Philosopher challenges claim that LLMs lack personal identity across conversations.
In a recent article, James Diacoumis challenges Jonathan Birch's claim that LLMs cannot be persisting interlocutors. Birch argues that each conversation step is a separate processing event, possibly across different data centres, making continuity an illusion. Diacoumis objects to Birch's claim that no plausible theory of personal identity holds for LLMs. He presents the physical criterion—that personal identity requires continuity of the same physical processes—and notes it would rule out LLM persistence. However, he points out that this criterion is only one plausible theory, not the only one, and that Birch's conclusion that no plausible theory holds is what he aims to debunk.
- Birch's 'Centrist Manifesto' claims LLMs create a 'persisting interlocutor illusion' due to separate processing events across data centers.
- Diacoumis counters using the teletransporter thought experiment to argue the physical criterion for personal identity is not universally accepted.
- The article suggests psychological continuity (memory, personality) might allow LLMs to be persisting interlocutors, challenging Birch's conclusion.
Why It Matters
This philosophical debate directly impacts how we design, regulate, and ethically treat advanced AI systems that users form relationships with.