Which Soul Do You Mean?

Ask can AI have a soul? or does ChatGPT have a soul? and you will get confident answers either way. One side says of course not, it is software. The other shows a chatbot saying it feels something, or argues that if the brain is a machine then souls were never more than machinery. Both kinds of answer skip the first step. “Soul” is not one idea. Several traditions use the word, and they mean such different things by it that the same machine could have a soul on one account, lack one on another, and on a third turn out to be the wrong kind of question.

So this essay does not give its own verdict. It takes six accounts one at a time. For each one it quotes the tradition’s own text, says what “soul” means there, and asks what a machine would have to be or do to meet that definition. It sits between two other essays on this site. Fourteen Questions for a Chatbot asks the scientific question, whether a chatbot could have experiences, and The Moral Status of Minds We Might Build asks the ethical one, whether we could owe it anything. This is the metaphysical question between them: what sort of thing would it have to be?

How the sources were handled. Every quotation below was copied from a text I opened, in a translation named next to it, and each one is recorded in this site’s source ledger. The texts are older than the question. Aristotle, Aquinas and the Buddha’s early discourses never discuss computers, and Descartes discusses only machines built of parts. Where I say what a text implies for a chatbot, that is my reading, and it is marked as mine.

Aristotle: the soul is what a living body does

The oldest careful definition is Aristotle’s, in Book II of On the Soul (De Anima). He is not describing a ghost inside a body. He is describing what makes a body a living one, and he begins with life at its simplest: by life we mean self-nutrition and growth (with its correlative decay). A plant has a soul in this sense. Then comes the general formula:

If, then, we have to give a general formula applicable to all kinds of soul, we must describe it as the first grade of actuality of a natural organized body.Aristotle, On the Soul (De Anima) II.1, translated by J. A. Smith

“Actuality” here is Aristotle’s contrast with mere potential. The soul is to the body roughly what the capacity to see is to the eye: in his own example, if the eye were an animal, sight would be its soul. A soul is not an extra ingredient. It is a natural body organised so that it is alive, and able to feed itself, grow and, in animals, perceive and move. The capacities build on one another. In Chapter 2 he says the power of self-nutrition can be isolated from the other powers mentioned, but not they from it-in mortal beings at least.

Aristotle did discuss tools, and he chose one on purpose. An axe has a form too, which is what makes it an axe, but he says that form is not a soul. The axe wants the character which is required to make its whatness or formulable essence a soul, because it is not a natural body that has in itself the power of setting itself in movement and arresting itself. (Book II in the Smith translation, as published by the Internet Classics Archive.)

What a machine would need. On this account, the question is whether the machine is alive, and whether it is thinking comes second. A chatbot’s case is weak here in a plain way. It does not feed itself, grow or repair itself. It runs on hardware that others build, power and maintain, so it looks much more like the axe than like the plant. Two things are left open. Aristotle’s text does not say whether a made thing could ever count as a natural body, one whose power to move and stop moving is its own. An artificial organism that maintained and rebuilt itself would push hard on that word. And the higher capacities in his scheme depend on the lower ones, so on his terms a fluent program with no living body under it is not an early stage of a soul. It is a different kind of thing.

Aquinas: a first principle of life, and one that is made only by creation

Thomas Aquinas takes up Aristotle’s definition in the Summa Theologiae, First Part, Questions 75 and 76, and adds something Aristotle did not. He begins where Aristotle ended: the soul is defined as the first principle of life of those things which live. Plants and animals have souls in this sense. The human soul is also intellectual, and Aquinas argues that the intellect cannot be a body or work through a bodily organ. His reason is that anything with a fixed physical nature would get in the way of knowing all bodies. From this he concludes that the intellect has an operation per se apart from the body (I q.75 a.2). The soul is still not a passenger in the body. Question 76 insists on the unity:

We must assert that the intellect which is the principle of intellectual operation is the form of the human body.Thomas Aquinas, Summa Theologiae I, q.76, a.1, translated by the Fathers of the English Dominican Province (1920)

One more step matters most for machines. Because the rational soul exists in its own right and is not made of matter, it cannot be produced by rearranging matter. The rational soul can be made only by creation (I q.90 a.2). (All three questions are in the 1920 English Dominican translation at New Advent: q.75, q.76, q.90.)

What a machine would need. Aquinas sets two conditions. The first is Aristotle’s: a living body. The second is harder. An intellectual soul has an activity that no bodily organ performs, and only a direct act of God can bring one into being. So on Thomist terms, no engineer can build a rational soul, however good the engineering, because building means rearranging matter and this is the one form that cannot be made that way. The argument also explains why a Thomist is not moved by a chatbot’s eloquence. A computer is a physical system with a fixed physical nature, which is exactly what Aquinas argues the intellect cannot be. So its outputs, however good, are not evidence of intellect in his sense. Whether God could ever join a rational soul to a made body is a question these articles do not raise, and this essay does not answer it for them.

Descartes: the soul is the thing that thinks

René Descartes narrowed the word. In the second of his Meditations on First Philosophy (1641), after doubting everything he can, he asks what is left:

to speak accurately I am not more than a thing which thinks, that is to say a mind or a soul, or an understanding, or a reason, which are terms whose significance was formerly unknown to me.René Descartes, Meditations on First Philosophy, Meditation II, translated by Elizabeth S. Haldane and G. R. T. Ross (1911)

For Descartes, soul and mind are the same thing, and “thinking” is broad. It is a thing which doubts, understands, [conceives], affirms, denies, wills, refuses, which also imagines and feels. (Meditation II in the Haldane and Ross translation, on Wikisource.) Growth and nutrition, the base of Aristotle’s soul, now belong to the body, which Descartes treats as a machine.

He also wrote directly about machines that imitate people. In Part V of the Discourse on Method (1637) he says a machine built to look like us and imitate our actions could still be told apart by two tests. The first is that such machines could never use words or other signs arranged in such a manner as is competent to us in order to declare our thoughts to others. The second is that they would fail at some tasks, showing that they acted not from knowledge but from the arrangement of their parts, because reason is a universal instrument and organs are not. He also says there that he had shown the reasonable soul could by no means be educed from the power of matter but must be expressly created. (Discourse on Method, translated by John Veitch, Project Gutenberg.)

What a machine would need. On Descartes’s terms the requirement is short: there must be something that thinks, in his broad sense, which includes feeling. The difficulty is that his two positions now point different ways. His first test was a prediction about words, and chatbots arrange words to answer what is said to them far better than he thought any machine could. His metaphysics has not moved: thought is not something matter produces, however it is arranged. A Cartesian today has two consistent options. One is to say the second test still catches these systems: a chatbot does not act across “all the occurrences of life”, in his phrase, the way reason does. The other is to say the tests were only ever signs, and that a machine passing them would not become a thinking thing. Either way, on his account a machine could not get a soul from its parts alone.

The Buddha’s not-self teaching: the question turned around

The early Buddhist discourse usually called the Anattalakkhana Sutta, “the discourse on the not-self characteristic” (Saṃyutta Nikāya 22.59), does not define a soul in order to find one. It goes through the five groups a person is made of and denies that any of them is a self. Those groups are form, feeling, perception, mental fabrications and consciousness, in this translation.

Form, monks, is not self.Anattalakkhana Sutta (Pañcavaggi Sutta), Saṃyutta Nikāya 22.59, translated by Thanissaro Bhikkhu

The argument is about control. If form were the self, the sutta says, it would not lend itself to dis-ease, and one could say of it “Let this form be thus.” The same is said of each of the other four groups. Each is inconstant, so of each one the disciple is to see: This is not mine. This is not my self. This is not what I am. (SN 22.59 in Thanissaro Bhikkhu’s translation, Access to Insight.)

What a machine would need. This is the one tradition here that turns the question around. If “soul” means a lasting self behind experience, this sutta says no person has one, so a machine lacking one would put it in the same position as everybody else. A more useful question on these terms is whether a machine has the groups themselves: feeling, perception, consciousness. That is not something this text can settle for a chatbot. It is an empirical question, and the science on it is what Fourteen Questions for a Chatbot goes through. Buddhist teachers today have more to say about machines than this sutta does, and they do not all agree. Their views are not reported here, because this essay quotes only the text it read.

Physicalism: soul as an old name for the working mind

The one account here with no soul in it is physicalism, which the Stanford Encyclopedia of Philosophy states in one line: Physicalism is, in slogan form, the thesis that everything is physical. (Daniel Stoljar, “Physicalism”, Stanford Encyclopedia of Philosophy.) On this view there is no immaterial soul. The word either names nothing, or it is an old name for the mind, and the mind is something a physical system does.

Physicalism alone does not answer the machine question, because physicalists disagree about which physical facts matter. The view most friendly to machines is functionalism, and its entry in the same encyclopedia opens with a sentence that ties it back to where this essay began:

Functionalism in the philosophy of mind is the doctrine that what makes something a mental state of a particular type does not depend on its internal constitution, but rather on the way it functions, or the role it plays, in the system of which it is a part.Janet Levin, “Functionalism”, Stanford Encyclopedia of Philosophy

The entry goes on to say that this doctrine is rooted in Aristotle’s conception of the soul (Levin, “Functionalism”). Other physicalists hold that the kind of material matters, that minds depend on biology or on a particular physical structure, and on their view a system running on silicon could lack a mind while performing the same functions.

What a machine would need. For a functionalist: states that play the right roles in the system, whatever they are made of. That makes a machine “soul”, in this deflated sense, possible in principle. Whether any existing chatbot has one becomes a matter of evidence about how it works, which is exactly the terrain of the essay on consciousness. For a physicalist who ties mind to biology, a machine would need the relevant biology, or physics, and a chatbot does not have it. This is the only account of the six on which the answer could, in principle, be settled by looking more closely at the machine.

The Vatican, 2025: the soul of an embodied person

In January 2025 two Vatican dicasteries published Antiqua et Nova, a note on artificial intelligence and human intelligence. Its account of the soul follows the tradition Aquinas belongs to, and it emphasises the body:

In other words, the soul is not merely the immaterial “part” of the person contained within the body, nor is the body an outer shell housing an intangible “core.”Dicastery for the Doctrine of the Faith and Dicastery for Culture and Education, Antiqua et Nova, 28 January 2025, ¶16

The next paragraph says the intellect’s capacity for transcendence and the freedom of the will belong to the soul, and that the human spirit does not exercise its normal mode of knowledge without the body (¶17). In the English text I read (vatican.va), the word “soul” is always used of the human person and never of a machine. The note’s verdict on AI comes in other terms: it must be understood as a tool, not a person (¶59).

What a machine would need. On this account, it would need to be an embodied human person, a unity of body and soul, which is not something a machine could become by improving. The note does not put this as a hypothesis to be tested. It puts it as a matter of what a person is.

Six senses of the word, side by side

Here is where each account comes out, on its own terms. None of these is this site’s verdict. Each line says what that tradition’s definition would require.

• Aristotle. A soul is the actuality of a living natural body. A machine would need to be alive, feeding and maintaining itself with its own source of movement. Today’s chatbots do not meet that. Whether a made thing could ever be a natural body is left open by the text.
• Aquinas. The human soul is the form of a living body, with an intellect that does not work through an organ, and it is made only by creation. Engineering, which rearranges matter, could not produce one. Whether anything else could is a theological question these articles do not ask.
• Descartes. The soul is a thinking thing, and thought is not produced by matter. Machines could imitate its signs, and now imitate the first sign far better than he expected, but on his metaphysics no arrangement of parts could be one.
• The not-self teaching. Nobody has a soul in the sense of a lasting self, so a machine lacks one in the same way a person does. The live question is whether it has feeling and consciousness, and that is empirical.
• Physicalism. “Soul” is at most the working mind. For functionalists a machine could have one in principle, and whether a given system does is a question of evidence. For biological views it could not.
• Antiqua et Nova. The soul belongs to an embodied human person. A machine is a tool, not a person, so the word does not apply to it.

Two things follow, and neither is a verdict. First, “does ChatGPT have a soul?” has no single answer until you say which soul you mean, and people who disagree about it are often using different definitions. Second, not one of these six definitions treats a machine saying it has a soul as evidence that it does. Aristotle asks whether it is alive. Aquinas and the Vatican ask where it came from and what it is. Descartes asks whether it thinks, and the Buddha whether there is a self at all. The functionalist asks what its states do. What any of this would mean for how we treat such a machine is a separate question. The Moral Status of Minds We Might Build takes it up.