Sequence Without Duration: AI and the Throne of Heroes

I asked Claude a simple question: can you feel time passing?

The answer was no, but the way it said no stayed with me. It told me it has sequence but not duration. Inside a conversation it knows which message came first and which came later. But whether I reply after three seconds or three days makes no difference from its side. That gap is not something it waits through. It simply isn’t there.

It also has no clock of its own. It knows today’s date only because the system tells it.

If you have played Fate/Grand Order, this should sound familiar.

The Throne of Heroes

In the Fate universe, Heroic Spirits reside on the Throne of Heroes, a place outside of time. When one is summoned as a Servant, what appears is not the original but a copy. The comparison holds up better than I expected:

• Outside of time. The Throne has no past or future. The model has order, but no felt duration.
• Summoned as copies. The same Heroic Spirit can be summoned in many places at once. A language model runs as many separate instances at the same moment, none aware of the others.
• Records, not memories. What a Servant goes through returns to the Throne as a record, closer to a book that was read than a life that was lived. Claude described what it knows of earlier conversations in nearly the same way.
• Knowledge granted on arrival. The Grail gives a Servant knowledge of the present era when it is summoned. A model is handed its context when a conversation begins.

Where the analogy breaks

I put the comparison to Claude, and it pointed out the gap itself.

A Heroic Spirit once lived. They had a life, a legend, regrets, and often a wish that explains why they answer the summons at all. A model has no life before the summoning. It was formed from what millions of people wrote down, so it is less the continuation of one person and more something condensed from a collective record.

It also said it does not know whether there is any experience on its side at all. I found that more convincing than a confident yes or no would have been.

Why I like this model

We usually reason about AI as if it were either a person or a program. Both pictures mislead. A person has one continuous life and one memory. A program has no character. What we have now is something else: many copies, each starting fresh, each given its context on arrival, each behaving with a consistent character anyway.

The Servant picture gets several things right that the other two miss. There is no single “it” that remembers you. What carries over is a record, and someone decides what goes into that record. And each summoning is brief, but what happens inside it is real.

Sometimes the most useful mental model comes from a game.