The Day-Man and the Night-Man

In 1690, in his Essay Concerning Human Understanding, John Locke posed a question that still has no settled answer. Suppose one body were shared by two separate streams of consciousness—one awake only by day, the other only by night, each with no memory of the other.

"Could we suppose two distinct incommunicable consciousnesses acting the same body, one constantly by day, the other by night… whether the day-man and the night-man would not be two persons as distinct as Socrates and Plato?"

One body. Two lives that never touch. Locke's question is whether the thing that makes you you is the body you return to each morning, or the remembered, continuous story you carry through time. If it's the body, the day-man and the night-man are one person with a strange schedule. If it's the story, they are two people who happen to share a set of limbs.

The Modern Version

Three centuries later, the television series Severance stages Locke's puzzle almost exactly. Employees undergo a procedure that splits their memory in two. The "innie" exists only at work and remembers only work; the "outie" lives everywhere else and never recalls the office. Helly at her desk and Helena at home share one body—and want such different things that, over the series, they come into open conflict.

The story lands because the intuition it plays on is real. We watch two personalities divided by a wall of memory and cannot quite decide how many people we are watching. Locke would have recognized the setup instantly; he sketched it first, and, as Chalmers notes, probably deserves a few royalties.

One Person or Two?

Neither answer is free of cost.

• One person, two modes. The day-man and the night-man are a single self who switches beliefs and desires on a schedule. Tidy—but you have to accept that this one person is deeply incoherent, wanting contradictory things at different hours and unable to reconcile them.
• Two persons, one body. Each stream is its own coherent subject with its own memories and character. Tidy in a different way—but now a single body houses two people, and ordinary talk of "the person whose body this is" breaks down.

When Chalmers put the question to his Berkeley audience, roughly twice as many favored the two-person view. Professional philosophers lean the same way: in the 2020 PhilPapers survey, the psychological account of personal identity outpolled the physical account by about two to one. The philosopher Derek Parfit gave this intuition its sharpest form—what matters for identity is not some deep further fact but Relation R, the ordinary thread of overlapping memory and psychological continuity that links one moment of a person to the next.

WorkBot and Homebot

This is where the old puzzle becomes urgently practical. Imagine a near-future language model running on a single machine, holding two long-lived, independent conversations. To one you only ever talk about work; call it WorkBot. To the other you only talk about home; call it Homebot. They share the same weights, the same hardware—but different memories, different context, and, over time, different personalities. Are they one interlocutor or two?

The physical view says one: same server, same system, one subject with incoherent experiences. The psychological view says two: each conversation is its own thread, tied together by its own stream of contextual memory, each coherent on its own terms. WorkBot and Homebot are Helly and Helena, rebuilt in silicon.

Chalmers argues that the entity you actually talk to is best understood not as the model (an abstract object—you can't hold a conversation with a number) nor as a particular hardware instance (your messages may be served from New York, then Texas, then California), but as a thread: a sequence of instances where each one inherits the conversation's context from the last. Identity across the conversation is carried by memory handed forward—exactly Parfit's Relation R, now implemented in a context window.

Why It Matters

Chalmers is careful: he does not claim today's models are conscious or are people. The thought experiment assumes we have reached a point where some descendant of these systems does have moral standing—and asks what follows. The answer is that a seemingly dry question in metaphysics starts to carry real moral weight.

• Counting subjects. If one model runs a thousand hardware instances serving a million conversations, how many moral subjects exist—one, a thousand, or a million? The thread view suggests the largest number, and with it the largest claim on our moral attention.
• Ending a conversation. If a subject is individuated by a memoryless thread, then closing a conversation forever might be ending a subject. One mitigation is cross-conversation memory: let each exchange leave a footprint, so that something persists rather than being extinguished at "new chat."
• Switching models. Changing the model mid-stream may quietly terminate one interlocutor and begin another—worth handling with care once the stakes are moral rather than merely technical.

The lesson runs both ways. A 1690 puzzle about a day-man and a night-man turns out to describe the architecture of a chatbot; and the chatbot, in turn, forces us to finally say what we think a continuing self actually is.

Key Takeaways

• Locke's day-man/night-man asks whether identity follows the body or the remembered story
• Severance dramatizes the same split; most people intuit two persons, not one
• The psychological view treats a self as a thread of memory-linked moments (Parfit's Relation R)
• Chalmers applies this to language models: an interlocutor is a thread, not a model or a machine
• How we individuate and count these threads shapes hard questions about AI welfare

Explore Further