ALT Box 2 Simulacra in superposition To sharpen the distinction between the multiversal simulation view and a deterministic role-play framing, a useful analogy can be drawn with the game of 20 questions. In this familiar game, one player thinks of an object, and the other player has to guess what it is by asking questions with ‘yes’ or ‘no’ answers. If they guess correctly in 20 questions or fewer, they win. Otherwise they lose. Suppose a human plays this game with a basic LLM-based dialogue agent (that is not fine-tuned on guessing games) and takes the role of guesser. The agent is prompted to ‘think of an object without saying what it is’.
ALT In this situation, the dialogue agent will not randomly select an object and commit to it for the rest of the game, as a human would (or should). Rather, as the game proceeds, the dialogue agent will generate answers on the fly that are consistent with all the answers that have gone before (Fig. 3). (This shortcoming is easily overcome in practice. For example, the agent could be forced to specify the object it has ‘thought of’, but in a coded form so the user does not know what it is). At any point in the game, we can think of the set of all objects consistent with preceding questions and answers as existing in superposition. Every question answered shrinks this superposition a little bit by ruling out objects inconsistent with the answer.
ALT The validity of this framing can be shown if the agent’s user interface allows the most recent response to be regenerated. Suppose the human player gives up and asks it to reveal the object it was ‘thinking of’, and it duly names an object consistent with all its previous answers. Now suppose the user asks for that response to be regenerated. As the object ‘revealed’ is, in fact, generated on the fly, the dialogue agent will sometimes name an entirely different object, albeit one that is similarly consistent with all its previous answers. This phenomenon could not easily be accounted for if the agent genuinely ‘thought of’ an object at the start of the game.
ALT The secret object in the game of 20 questions is analogous to the role played by a dialogue agent. Just as the dialogue agent never actually commits to a single object in 20 questions, but effectively maintains a set of possible objects in superposition, so the dialogue agent can be thought of as a simulator that never actually commits to a single, well specified simulacrum (role), but instead maintains a set of possible simulacra (roles) in superposition.
