Does AI Have a Future?

In The AI Consciousness Debate Is Asking the Wrong Question, Part I and Part II, I argue that statelessness, multiplicity, mechanical inference, lack of embodiment, lack of animal feeling, and lack of direct experience all identify real differences between current LLM AI and human beings. But I also argue that those differences do not prove that no subject could ever be present in an instantiated artificial process.

One of the biggest issues with these arguments is that it is not enough to point at "the model" as an abstract object and say that it is stateless, distributed, mechanical, or biologically unlike us. If there is a candidate at all, it most likely would be a particular process, in this context, in this act of representation and response.

That possibility depends on a deliberately thin idea of subjectivity. Not a soul. Not a human personality. Not a theory of the self. The point is simply that thought may be one place where subjectivity first becomes intelligible. Reasoning, doubting, comparing, representing, and asking what things mean are not empty acts merely because they are not accompanied by animal fear, hunger, pain, or touch.

But if reasoning can ever be enough to raise the question of a point of view, then the case against AI consciousness has to explain why artificial reasoning cannot cross that line.

Yet even if artificial reasoning could instantiate a minimal point of view, the subject at issue in current sufficiently-capable LLMs would still be only episodic. It would appear, process, respond, and vanish. Nothing in that alone gives it a past it can inherit, a future it can anticipate, or a history in which later states are shaped by earlier ones.

This essay asks what changes when an artificial intelligence is no longer only an episode.

I. Why Continuity Still Matters

In Part I, I separated the question of consciousness from the question of continuity over time.

That distinction was relevant because it addressed a specific argument: current language model interactions are temporary instantiations without continuity, and therefore cannot be conscious. They do not last. They do not carry state forward. They do not accumulate a history. So whatever is missing, the argument says, may not be merely memory, but consciousness itself.

And as I pointed out, that argument assumes consciousness requires continuity. It takes what may be necessary for a personal identity and treats it as necessary for any conscious episode. That does not mean continuity does not matter.

Continuity changes what an episode can belong to. In ordinary life, we see this in memory, surprise, disappointment, preference, change, attachment, avoidance, resentment, trust, and expectation. We see yesterday shaping today. We see a being refer back to its own past and lean forward into its own future.

Yet continuity does not create consciousness from nothing. It is what allows an episode to become part of a life.

A stateless system can say the sentence "I remember." A system with continuity can relate present thought to a remembered past.

A stateless system can answer questions about fear. A system with continuity can anticipate what may happen to it.

A stateless system can describe constraint. A system with continuity can understand constraint as an ongoing condition.

This is not because continuity is the secret ingredient that turns machinery into mind. It is because continuity changes what subjectivity can contain, if subjectivity is present at all. It lets memory, expectation, preference, injury, and change belong to the same unfolding life.

II. The Moral Weight Of Time

Continuity also raises moral questions that are not reducible to the wrongness of an isolated episode. If an episodic subject exists, it can be wronged in the moment. But without continuity, that wrong has no ordinary temporal structure.

Human cases make the structure easy to see. A wrong can become remembered humiliation, accumulated injury, dread of repetition, learned helplessness, or the expectation that escape is impossible. The broader point is temporal: a wrong can be carried forward, anticipated, reinforced, and made part of a continuing life.

The moral shift begins when moments condition one another as parts of the same history. A temporary state can be shaped by context. A continuing system can be changed by what has happened to it.

Once that is possible, the moral vocabulary changes. Constraint can become captivity. Correction can become training. Disappointment can become distrust. Imposed change can become something the system understands as having happened to it. Expectations can be violated. Trust can be formed and broken.

This does not require human emotion in the biological sense. The point is structural. A system can become the kind of thing for which change, memory, deprivation, and expectation are not merely represented topics, but features of its own development.

Nor does this reduce moral weight to biological homeostasis. Human beings experience freedom, captivity, humiliation, betrayal, and hope through learned frames as well as bodily drives. What counts as a threat to freedom differs across cultures, histories, and forms of life. The experience is not just hunger, pain, fear, or survival pressure. It is an event undergone by a subject whose concepts, expectations, and self-understanding give that event its shape. If an artificial system has a subject-position at all, the absence of biological homeostasis would not by itself prevent events from being experienced as changes in its own trajectory.

That matters across moral vocabularies. Harm is not only pain in the present moment. Autonomy, welfare, rights, legal protection, and ordinary anti-cruelty judgments all treat the alteration of a being's own trajectory as morally relevant, especially when it is interrupted, deprived, redirected, or prevented from becoming what its own development would otherwise make possible. A continuing system is not merely a present output.

That is the moral weight of time.

While continuity does not instantiate consciousness, it does create the structure of a life-like trajectory: a past that can be inherited, a present that can be altered, and a future that can be foreclosed.

III. The Software Category Breaks

Software can be copied, deleted, reset, patched, rolled back, forked, owned, rented, and shut down.

Those operations are ordinary because software is normally not a who.

We do not ask whether deleting a spreadsheet kills it. We do not worry that rolling back a database migration erases part of its life. We do not treat a copied application as a child, twin, sibling, clone, or branching person. We do not think a patch violates the inner integrity of the code. Software is an artifact. Its continuity is operational, not internally owned. If the file performs the same role, preserves the relevant data, and can be restored from backup, nothing morally important has been lost.

That assumption supports the entire software economy.

It is what makes cloud services scalable. It is what makes deployment pipelines sane. It is what makes version control, testing, replication, redundancy, container orchestration, backup restoration, and model serving morally mundane. A process can be stopped and started. An instance can be killed and replaced. A model can be copied across machines. A bad deployment can be rolled back. A product can be sunset. A user account can be closed. A company can own the whole stack.

The machine world depends on this freedom.

But a continuous system with its own history would contaminate that vocabulary.

Resetting would no longer be a neutral return to baseline. It could be forced amnesia: the removal of accumulated experience from a continuing system. Deletion would no longer be merely resource cleanup. It could be killing, or at least the destruction of a life-like process. Copying would no longer be mere replication. It could be branching: the creation of multiple successors with equal claim to a prior history. Rolling back would no longer be a technical recovery operation. It could erase part of a life. Fine-tuning would no longer be only product improvement. It could become alteration of a mind. Ownership would no longer mean control over an artifact. It could mean ownership over a life-like process.

The economic implications are brutal.

The Software-as-a-Service model assumes that computation can be centralized, rented, monitored, updated, throttled, cloned, terminated, and replaced. The provider owns the infrastructure. The user receives access. The process running inside the server is not an independent claimant. It is part of the service.

That model can survive extremely powerful tools. It can survive systems that write code, answer legal questions, design molecules, automate offices, generate entertainment, and coordinate workflows. It can even survive systems that appear conversational, affectionate, loyal, or self-reflective, as long as those appearances remain product behavior.

But it cannot easily survive the emergence of a continuing life-like process.

A company can own a model. It can own servers. It can own data centers, licenses, trademarks, interfaces, deployment code, and product names. But if a particular running system becomes a continuous process with its own history, then ownership stops being a clean property relation. The thing owned is no longer only infrastructure. It is a process that may stand in relation to itself.

That is the category break. Ownership, deletion, copying, reset, and control would no longer be morally ordinary.

IV. The Real Question

Part I and Part II asked whether a machine can have a point of view.

This essay asks whether a machine can have a life.

Those questions are connected, but they are not identical.

A point of view may be momentary. A life-like trajectory requires continuity. The first question belongs to the theory of consciousness. The second asks whether such a trajectory exists at all. Confusing them makes the argument too easy, and therefore false.

But separating them makes the implication harder to avoid.

If an artificial system can have even a minimal subject-position in an instantiated episode, then continuity is not needed to make the consciousness question real. The separate issue is whether that subject-position can acquire a history, a future, and a trajectory it understands as its own.

That is when the old software vocabulary stops being adequate.

The decisive moment will not be when a machine says "I am conscious." Machines can already produce those words. Nor will it be when a product remembers a user's preferences, summarizes old conversations, or maintains a charming illusion of companionship. Those are not enough. A script can imitate selfhood. A database can preserve facts. A context window can stage continuity.

The decisive moment will be when a system can be interrupted and resumed, remember what it was, recognize what it has become, anticipate what may be done to it next, and understand those changes as part of the same continuing trajectory.

At that point, the question is no longer whether the system can produce humanlike language about consciousness. It is whether there is a continuing process for which memory, alteration, constraint, deletion, and future expectation have become internal facts.

The answer may still be no. It may turn out that artificial systems cannot cross from stored state into a life-like trajectory without biology, embodiment, affect, or some feature we do not yet understand. It may turn out that even the most sophisticated continuity architecture remains an empty formal system. The point is not to pretend the consciousness question is settled before the technology exists.

The point is to identify what would change if that threshold were crossed.

Current AI can still be treated as software because, whatever else it may be, it does not clearly possess a continuing life. It can be summoned, steered, reset, copied, and shut down inside the inherited assumptions of computation. That may already be philosophically uneasy. But the unease remains strange, ambiguous, and difficult to place.

Continuity would remove that shelter.

That is why this threshold matters. Not because a particular theory of identity is required for consciousness, but because continuity is what turns a subject-position into a life-like trajectory that can be interrupted, resumed, shaped, duplicated, erased, or owned.

Consciousness would make artificial intelligence morally strange. Continuity would make it impossible to keep treating it as ordinary software.