Does AI Really Have a Personality?

The doubt that an experiment like this could be filed under pseudoscience or divertissement is very real: asking an LLM whether it gets moved by a film might not make much sense — or certainly makes less sense, or a different kind of sense, than asking a human. But in every case, giving a definite answer to a question like that is the outcome of a process which, today, however much it may be a simulated process, can still involve a fairly articulated chain of conceptual processing — once the exclusive province of humans, but no longer.

In the age of agentic AI, moreover, a not-fully-deterministic conceptual process, typical of natural language, can correspond to actions that lead to real events (concrete, not simulated). If an AI has no character, it can simulate one, and act concretely on that basis. Monitoring, downstream of an LLM’s production process, the emergence of a given simulated personality (and not just its advance in semantic and cognitive capabilities) might not be such a foolish idea — also in relation to alignment problems.

Most likely, tests like these will need to be adapted for LLMs to understand what tendencies they have toward objective functions (once they become agents) and toward aggregate work dynamics for when they act in swarms.

Perhaps, as always, this too comes down to finding the right questions rather than the right answers.