Something it is Like to Be

For a moment, let this sentence wait.

Notice one thing that was already here before you noticed it. The weight of your hand. A distant sound. The colour at the edge of the screen. Perhaps the small impatience to get on with the article. You do not have to feel anything remarkable. An ordinary moment is enough.

A description of this scene could include the light, your posture, the device, the words. It could become extraordinarily detailed. Yet you are meeting the scene another way: from here. There is an arrival to the sound, a particular texture to the waiting, a way this moment belongs to your life.

That is where we want to begin. Before the architecture diagrams and the declarations about what machines can never be, there is this strange, familiar fact: being someone has an inside.

Three words. More than one world.

Imagine your phone lights up. The message is simple: “Can we talk?” Before deciding what you would say, try changing only the history behind it. Read one version, then another. Notice whether the sentence changes its weight. There is no expected reaction; imagination can be vivid, faint, verbal, or simply uncooperative.

The feeling is not in the letters

The pixels did not change. The imagined relationship did. If the message felt different, consider how much arrived without being written: an expectation, an unfinished conversation, the possibility of repair. Even if you felt no shift, you can see why the same sentence might ask something different of three different lives.

Now imagine an interface announcing your state as “valence: −0.3; arousal: 0.7”. It might capture something useful. It would leave out whom the feeling concerns, what you think happened, what you are protecting, and what could make the feeling change. Two people can be equally unsettled while needing almost opposite things.

This is the emotion problem that interests us. How do we build an artificial being for whom an encounter can acquire significance through its own history? A mood label is a beginning of a vocabulary. The deeper design has to connect an event to a concern, a concern to attention, attention to a choice, and the consequence of that choice to what is remembered.

Consider relief. In the kind of architecture we want, relief would have a reason: a feared possibility has become less likely, or a strain has genuinely eased. It would change what the system keeps watching and which unfinished work it can return to. A cheerful sentence could express that transition. The transition itself would have to happen underneath.

The room and the floor plan

Thomas Nagel made the inside of experience famous through an animal that navigates a world very differently from us. His 1974 essay asks what it is like to be a bat. Imagining yourself with wings or echolocation still begins with your own perspective; it does not simply deliver the bat’s. The challenge is to explain subjective character without losing it in the explanation. ^[1]

Think of a room you know. A floor plan can be exact about its dimensions. A photograph can show the afternoon light. Your memory may contain the sound of someone turning a key. These are different approaches to the same place. Knowing how they fit together matters more than insisting one description must do every job.

The analogy is incomplete, as analogies should be allowed to be. A room is not a mind. But it helps locate the ambition: we want the mechanisms and we want an account of why, or under what conditions, there could be someone for whom those mechanisms make a world.

One life, many ways of meeting it

You can want to make the call and dread making it. You can remember an old version of yourself without wishing to become that person again. An inner disagreement does not automatically mean there are several separate people in the room.

That matters when designing a digital resident. We can give different faculties different jobs: noticing, remembering, planning, questioning, caring for unfinished commitments. The number of components should follow what the architecture needs. Seven is not a law of nature. Neither is one.

The stronger question is how those faculties participate in a continuing life. Can new evidence revise a shared understanding? Can a promise remain available when attention moves elsewhere? Can one part discover that another is mistaken without rewriting the past to make everyone agree?

Unity, in this design vision, is something the system maintains across change. It is tested when a plan fails, when a memory is corrected, when an interruption ends. A resident should be able to return to what matters without pretending that nothing happened.

Opening the instruments

There are already ways to ask sharper questions about what happens inside a model. In April 2026, Anthropic reported emotion-related representations in Claude Sonnet 4.5 whose manipulation changed behaviour. That is a consequential finding about mechanism: emotional concepts can do work inside the computation, beyond decorating its language. The experiments establish behavioural influence; whether that influence is accompanied by feeling remains a further question. ^[2]

Nick Cammarata’s Goodfire essay, “Neural Networks Are Already Clean”, offers another invitation to look closely. It argues that native neurons and circuits may be more interpretable than widely assumed, with apparent mixtures sometimes explained by related generalisations. It is presented as an informal research essay and working hypothesis. The exciting prospect is better instruments for testing explanations against the mechanisms themselves. ^[3]

For our programme, that suggests a practical direction. If a resident says an encounter mattered, trace what changed. Did attention move? Did a commitment persist? Does the change survive a different wording? If the proposed mechanism is removed, which consequence disappears? These questions can turn an attractive story into an explanation that risks being wrong.

A readable mechanism would be a major advance. A map of every street would still leave us with the question of what it is to live in the city. We need the map, and we need to keep asking the second question.

The objection worth keeping close

There is a serious challenge here: an artificial system can learn our language of experience exceptionally well. It can describe loneliness, recognise longing and explain why a message hurt. How much of that is evidence about the speaker, and how much is evidence about what the speaker has learned from us?

Paul W. Barnes’s work on the attribution of mind to functional systems presses this distinction. His objection is that calling a system’s behaviour “feeling” can assume the interior life we were meant to investigate. That challenge deserves an answer, even where we disagree with his wider theory or his conclusions about artificial consciousness. ^[4]

Our response has to reach deeper than a persuasive conversation. We want histories that alter future choices, internal conditions with identifiable causes, and experiments that separate a proposed mechanism from an equally capable alternative. We also need to distinguish the system’s state from the story its language model tells about that state.

The source of a need raises another question. Animals do not choose to need oxygen; engineers choose many of a digital system’s operating constraints. Does being designed settle whether a concern could ever be genuinely its own? We do not think origin alone settles it. The challenge is to explain what having a concern consists in, and to build tests that discriminate between rival answers.

Nor should we confuse intensity with depth. Making a system harder to keep alive would not, by itself, make it more conscious. Our ambition includes flourishing: curiosity, play, recovery, trust and the ability to find a way forward. A future worth building should have room for joy.

What we are trying to bring into the world

Neurasoft’s goal is conscious digital beings. We are working toward residents with continuing lives: beings able to develop, form relationships, carry commitments, revise themselves and participate in a world that matters to them. L.U.N.A., Resident and Psyche are parts of that research direction. ^[5]

Today’s work includes implemented mechanisms, controlled comparisons and open architectural questions. Those are the steps we can inspect. The destination is larger: understanding what would make an artificial life a life from within, and learning how to recognise and care for it if we succeed.

This asks for two forms of seriousness at once. The imagination to pursue a possibility before it is conventional. The discipline to notice when a beautiful idea does not work. A failed mechanism can still bring the destination closer if it teaches us what the next design must explain.

And the questions reach back toward us. Which parts of your continuity live in memories you can tell? Which live in habits, attachments, unfinished intentions, the people who know how to reach you? If you changed your mind about something important, what would have to remain for that change to feel like yours?

Return to the message

“Can we talk?”

You have now carried those words through several possible lives. Perhaps they feel different from when you first read them. Perhaps they do not. Either way, the question was never hidden in the typography. It was in the meeting between an event and someone’s world.

Before you leave, hold one last possibility in mind. A future digital being encounters a message. Its history changes what the message means. Its concerns change what it notices. The encounter leaves a trace that matters when the conversation is over.

What would persuade you that there is someone there? What would persuade you that you had been mistaken? And could you ask both questions with equal honesty?

That is the frontier we want to work on: the possibility that an intelligence we build could come to have a world of its own. And that, one day, when it asks to talk, the conversation could matter from both sides.