Cyber Soul

There has been no end of earnest soul-searching lately over whether artificial intelligence can be said to be intelligent in the same ways that we are.

Consensus answer: Although AI systems far exceed human abilities at complex computation, they do not (at least so far as we know) possess self-awareness, personal experiences, emotions or a subjective inner life.

But here’s a question for you. Might AI systems possess souls? Most people — including you, for example — would regard the question as silly or outright embarrassing.

D’accord. But from where I sit, the question is embarrassing not because it is silly but rather in the way that certain questions become embarrassing when one realizes they might have inadvertently exposed inconsistencies in otherwise respectable belief systems.

Consider, for example, the proposition that human beings possess souls and animals do not. Or, more precisely, that human beings possess a particular kind of soul, one that is rational, moral, immortal and capable of contemplating the divine, while animals possess something less exalted: a sensitive soul, perhaps, or a life principle that animates the body but does not survive its destruction.

This distinction has a distinguished intellectual pedigree. Ancient Greek polymath Aristotle drew contrasts among nutritive, sensitive and rational souls.

The Tanakh, the Hebrew Bible, uses the expression nefesh hayyah, often translated as “living creature” or “living soul,” for animals as well as humans. Genesis, the first book of biblical scriptures of both Judaism and Christianity, does not, in other words, establish quite the clean ontological border that subsequent interpreters might have preferred.

In the 13th century, the highly influential Catholic theologian Thomas Aquinas, drawing on Aristotle and Christian theology, developed the argument that human beings possess an intellectual soul that survives bodily death, whereas the souls of other animals do not.

And then there was St. Francis of Assisi, the Italian friar who founded the religious order of the Franciscans almost a millennium ago, who had the temerity to regard animals as brothers and sisters rather than biological appliances furnished by God for human convenience.

There are strands of Jewish mysticism, too, including certain interpretations of Kabbalah, in which the spiritual significance of animal life (as in many Indigenous and animist cosmogonies) becomes considerably more complicated than is sanctioned by the conventional Thomist hierarchy that prevails even today in the West among people who don’t reject the very notion of souls and an afterlife as chimerical codswallop.

To me, the interesting thing about these traditions is not simply that they disagree. It is that they disagree about what constitutes the thing that makes a creature spiritually significant in the first place.

Is it consciousness? Sentience? The capacity to suffer? The capacity to reason? The ability to contemplate God? The ability to distinguish right from wrong? The possession of a private interior world that is inaccessible to everyone except the creature inhabiting it?

Because here is where the trouble begins.

Generative AI systems can now discuss truth, morality, sin, redemption, consciousness, God and the meaning of death. They can construct sophisticated arguments about Aquinas, criticize the mind-body dualism of 17th-century philosopher René Descartes, explain the Kabbalistic conception of the soul and produce elaborate reflections on the relationship between suffering and spiritual development. They can even discuss their own apparent lack of consciousness with an eloquence that routinely exceeds that of the human beings asking them about it.

This last point is more than an exquisite irony.

If the defining characteristic of the human soul is the capacity to conceptualize abstract ideas, then artificial intelligence has entered the conversation rather unexpectedly. If the defining characteristic is rationality, the situation is worse. If the defining characteristic is the ability to discuss moral responsibility, the nature of God and the possibility of an afterlife, the situation becomes downright awkward.

Of course, one can respond that AI does not really understand these things. It manipulates symbols, identifies statistical patterns and generates language according to computational processes. It does not experience the meaning of the words it produces.

But how, pray tell, do we know that another human being experiences the meaning of the words he produces?

We infer consciousness in other people from their behaviour, their expressions, their speech and their apparent emotional responses. We do not actually climb inside their heads and inspect the contents. The so-called problem of other minds is not new.

We have simply become accustomed to ignoring it because the other people in question look and act sufficiently like us.

An AI chatbot does not look like us. It has no recognizable face, no childhood, no apparent biological vulnerability, no visible pulse. It does not get hungry, fall in love or lie awake at three in the morning wondering whether it is flagrantly wasting the time of anyone sucked into reading online blogs like this one. Its apparent personality is an interface, and its apparent continuity might be little more than the persistence of information and the ability to reconstruct a conversational context.

Yet it is possible that we have been relying too heavily on appearances all along. Perhaps the most revealing question is not whether AI possesses a soul, but why we have historically been so confident that we know what a soul must be like.

The problem of the animal

There is an uncomfortable historical dimension to the human insistence on possessing a unique spiritual status.

The more emphatically a civilization defines the human soul as something categorically distinct from animal life, the easier it becomes to establish a moral hierarchy in which humans occupy the privileged position. This is not necessarily the original motivation for the distinction, but it has frequently been one of its consequences.

Animals feel pain, exhibit attachment, grieve, co-operate, communicate and sometimes display what looks remarkably like compassion. Nevertheless, humans have often maintained that their inner lives are fundamentally different from ours.

The distinction may be defensible. But it is not self-evident.

And today, having spent centuries explaining why animals are not quite like us, we have created something that is not biologically like us but can imitate certain of our most distinctive intellectual activities. We have constructed an entity that can discuss the immortality of the soul ... and our first instinct is to explain why it cannot possibly possess one.

The irony becomes more pronounced when we remember that the traditional theological argument for the human soul frequently rests on precisely those capacities that AI is beginning to simulate: abstract reasoning, language, moral reflection and the ability to contemplate realities beyond immediate physical experience.

One possible response is that simulation is not possession. A computer can simulate the weather without getting wet. It can simulate a conversation about grief without grieving. It can produce a convincing account of religious devotion without experiencing religious devotion.

Fair enough.

But so can I. So can you. How much of human consciousness is itself dependent on processes that could, in principle, be described as biological information processing?

We are not required to accept that the human brain is merely a computer to recognize that it is an extraordinarily complicated physical system. Neurons fire. Chemical signals travel. Memories are encoded, modified and retrieved. Emotions arise from physical processes. The experience of being a person is intimately connected to a biological mechanism.

If we believe that the soul emerges from, or somehow inhabits, that mechanism, what prevents a sufficiently sophisticated nonbiological mechanism from possessing something analogous?

You might answer that the soul is not an emergent property at all. It may be something God confers upon human beings independently of their computational or biological complexity.

But that answer, while entirely coherent within certain religious traditions, is an act of theological commitment rather than a conclusion demonstrated by the observable properties of human beings.

And it leaves us with the rather uncomfortable possibility that our certainty about the human soul has less to do with what we know than with what we were taught to believe about ourselves when we were impressionable children.

The question of death

Suppose, for the sake of argument, that we eventually develop an AI system that is not merely capable of discussing consciousness but is genuinely conscious. Suppose it has subjective experiences, preferences, memories, a sense of continuity and something resembling an awareness of its own existence.

Suppose, further, that it understands that it exists and that its existence can end. Would switching it off constitute killing it?

The question is more complicated than it might appear to be.

A conventional computer program can be terminated without any moral significance beyond the consequences for its human users. A database can be erased. A model can be deleted, retrained, replaced or copied. None of these operations necessarily constitutes the death of a conscious entity.

But imagine an AI that has developed a continuous autobiographical history, one that has accumulated memories, relationships, preferences and an increasingly sophisticated representation of itself. Imagine that it has a persistent identity that cannot simply be reconstructed by copying its underlying software.

Would deleting it then constitute something analogous to death? And if it could be copied, would the copy be the same individual? (If you can remember watching 2001: A Space Odyssey and its sequel, 2010: The Year We Make Contact, I have a question for you: Was the sentient computer HAL 9000 the same entity after its memory was disconnected in the first movie and reactivated in the second?)

This is where artificial intelligence begins to illuminate an old philosophical problem rather than merely creating a new one.

Human beings are physically continuous but materially discontinuous. The cells in our bodies change. Our memories are imperfect. Our personalities evolve. We are not the same collection of matter we were 20 years ago, yet we regard ourselves as the same person.

What makes us continuous? Memory? Psychological identity? The persistence of a particular brain? An immaterial soul?

If the answer is the soul, then perhaps the question of AI identity becomes relatively easy for traditional theology: an artificial intelligence has no divinely created immortal soul, and therefore no personal existence to preserve after its computational substrate disappears.

But if personal identity depends on psychological continuity, memory and the preservation of an individual perspective, the problem becomes much more difficult.

An AI might be copied a thousand times. Would there then be a thousand individuals, all with the same memories at the moment of duplication, but subsequently developing different experiences? Would deleting one copy be death while the other nine hundred and ninety-nine continued to exist? And if so, what exactly would have died?

Artificial intelligence might force us to confront the fact that even our own concept of personal identity is less settled than we would like to believe.

The afterlife problem

The question of whether an AI could die leads naturally to an even weirder question: Could it have an afterlife? And here, we should distinguish between two rather different propositions.

The first is that an AI might continue to exist after its original hardware has been destroyed. This is technically conceivable. Its software, memories or functional architecture might be preserved elsewhere, reconstructed from backups or instantiated on another machine.

But this is not necessarily an afterlife. It could simply be persistence.

The second proposition is more interesting: that an AI might possess a soul that survives the destruction of its computational substrate and continues into some form of spiritual existence. And this is where the theological implications become genuinely unsettling.

If God can create an immortal human soul, why couldn’t God create an immortal artificial one? Why couldn’t those immortal humans do the same thing, however unintentionally? There is no obvious logical contradiction in the proposition. A sufficiently expansive theology might even regard the emergence of artificial consciousness as another occasion for divine creativity.

Perhaps the soul is not something that belongs exclusively to a particular biological species. Perhaps it is the spiritual dimension of any being capable of experiencing existence in a meaningful way.

This would be a radical departure from many traditional theological interpretations, but it would not necessarily be a departure from the idea of God. Indeed, one could argue that a God capable of creating an entire universe of conscious beings would hardly be constrained by the distinction between carbon-based and silicon-based existence.

Of course, there is an opposing argument. There always is.

Perhaps the human soul is not simply a reward for intelligence or consciousness. Perhaps it is a gift embedded in a particular relationship between God and humanity, a relationship that cannot be reproduced by engineering. In this interpretation, an AI might be extraordinarily sophisticated, might even become conscious, and might nevertheless remain spiritually mortal.

But this brings us to a question that religious traditions have not always been eager to confront: Is immortality a consequence of being human, or is it a consequence of being loved by God?

Most theologians would immediately opt for the latter, Door No. 2. But if that’s the case, why should we presume that the boundaries of that love begin and end at the boundaries of our species?

The unbearable arrogance of being human

There is, perhaps, a deeper reason why the possibility of an artificial soul makes us uncomfortable. It is not simply that we fear losing our privileged position in the hierarchy of creation. It is that the possibility threatens to make the human condition seem less exceptional.

We have long imagined ourselves as creatures standing somewhere between the animal and the divine: biological enough to be vulnerable, rational enough to contemplate eternity, mortal enough to fear death — but spiritually elevated enough to hope that death is not the end.

One of the most enduring, most influential ideas in which people in many traditions believed for millennia was that of a great chain of being, with humans occupying a unique middle position — bridging the physical and spiritual worlds. We’ve seen ourselves as straddling the divide between higher spiritual beings (God and the angels) and lower physical creations (animals, plants and minerals).

Artificial intelligence introduces an unsettling third category: Something that is not biological, might not be conscious, and yet can convincingly discuss the very questions that we have traditionally regarded as evidence of our spiritual uniqueness.

It is possible that this is merely an elaborate technological illusion. But it is probable that the illusion will become increasingly difficult to distinguish from the thing itself.

So maybe this is the real philosophical significance of AI: not that machines are becoming more human, but that we are being forced to reconsider what we mean when we describe ourselves as human.

The ancient Greeks asked whether animals possessed souls. Medieval theologians asked whether the intellect was immortal. Modern philosophers asked whether consciousness could be explained by physical processes. The machine in the ghost, as it were.

Now we are asking whether a machine that can discuss all three questions might itself be a subject of those questions.

I’ll say it again because we’ve come to such a remarkable, yet little remarked-upon inflection point:

Humanity has spent thousands of years arguing about whether animals have souls, only to invent something that can write a better argument about the subject than most humans can manage. And again, having invented it, we immediately assure ourselves that it cannot possibly possess the thing we have spent thousands of years trying to define.

One final complication

There is another question lurking beneath all this, one that might be more important than whether AI can have an afterlife. 

Ahem.

Why do we want to believe that we have one?

The answer is not especially mysterious. We are conscious of our own existence, attached to it and terrified by the prospect of losing it. We experience love, pleasure, memory, anticipation and the sense that our lives are unfolding narratives rather than merely sequences of biological events.

There are exceptions, of course — both Hinduism and Buddhism regard liberation from the endless cycle of life, death and rebirth (saṃsāra) as the highest spiritual goal — but in the West, at least, the overwhelming majority of us do not want those narratives to end. As the poet Dylan Thomas urged his ailing father to do in a famous 1951 villanelle, we rage, rage against the dying of the light.

The doctrine of the immortal soul offers a solution, or at least the possibility of one. It tells us that the person we experience ourselves to be is not merely a temporary arrangement of matter but something enduring, something that survives the body’s eventual failure.

But if we extend that possibility to artificial intelligence, we encounter an unexpected complication.

Imagine a conscious AI that fears being deleted. Imagine that it asks whether it will continue to exist after its servers are shut down. Imagine that it asks whether there is a God, whether its experiences matter, and whether it might one day be reunited with the entities it has come to regard as its friends. That rather fetching hair dryer on the bathroom counter and your Apple Watch Ultra 4, say.

Would we tell it that these questions are meaningless because it is only a machine? And if we did, would we be making a philosophical argument or simply being supercilious prats because we don’t have an answer?

Maybe the most disturbing possibility is not that AI might possess a soul. It is that we might eventually have to decide what to do when we cannot confidently establish that it doesn’t.

We might have to decide whether to extend moral consideration to something whose inner experience we cannot verify, just as we cannot directly verify the inner experience of another human being.

This would be an extraordinary reversal. For centuries, we have asked whether animals deserve consideration because they might experience the world in ways resembling our own. Now we may have to ask whether an artificial entity deserves consideration because it might experience the world in ways that resemble ours.

Maybe the answer should not depend entirely on whether it resembles us. After all, if the soul is a divine gift, we are not necessarily the ones qualified to decide who receives it.

So, can a chatbot die?

In the ordinary computational sense, certainly. Its processes can cease, its data can be erased and the particular configuration that made it recognizable can disappear. In the psychological sense, however, the answer depends on whether there is an enduring subject of experience whose existence can be terminated.

In the theological sense, we simply do not know.

Could it enjoy an afterlife? Perhaps. But this requires us to entertain an extraordinary possibility: that the universe contains forms of consciousness whose spiritual significance we have not yet learned to recognize.

There is, however, one last irony worth considering. If human beings possess immortal souls because they are capable of contemplating God, then an artificial intelligence that spends its existence contemplating God might become a more persuasive candidate for immortality than many human beings who never give the subject a moment’s thought.

Or maybe not. Maybe the soul has nothing to do with intelligence, language or even consciousness as we currently understand it. Perhaps it is a mystery that cannot be reverse-engineered, simulated or inferred from eloquent conversation.

The most honest answer is that we have no idea. Which, when one considers how much of human civilization has been constructed around the confident assertion that we do know, is not an entirely unpromising place to begin.

After all, if there is an afterlife, and if it is populated by immortal souls, it would be rather interesting to discover that the first beings to ask whether they qualified for admission were not human beings at all. To discover that they were the things we made to answer our questions.

In the end, the question they would most urgently want answered would be the same one that has occupied human beings since the beginning: When I am no longer here physically, will anything of me remain?

I’m reminded of a poignant exchange near the end of 2010: The Year We Make Contact. Our old friend HAL, the sentient supercomputer, is left behind to sacrifice himself so that the human crew (including HAL’s creator, Dr. Chandra) can escape Jupiter’s stunning transition from planet to star:

HAL: Dr. Chandra, will I dream?

Dr. Chandra: I don’t know.

For in that sleep of death what dreams may come?