David Shapiro · Tech & AI
TIER 4 Mon, 2 Feb 2026 16:29:56 +0000
Does this GPU have a soul? ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ | | ---|---|--- | | | Forwarded this email? Subscribe here for more --- # Taking the "AI Personhood" Debate Seriously ### Does this GPU have a soul? | | David Shapiro --- | Feb 2 --- | --- --- | | | --- | | --- | | --- | | --- | | READ IN APP --- I recently had a brief exchange with Peter Diamandis about machine personhood, and others are working to raise awareness. More often than not, people who are concerned about "model welfare" break into the "tech transcendent" camp, which is more of a new age religion than anything else, or the tech elite such as Peter and folks at Anthropic. | | ---|---|--- More and more "tech elite" are taking "model welfare" or "machine personhood" seriously. So let's unpack this. # First, what do we mean? If you're new to this conversation, you might think that "model welfare" is a DEI job role for looking after the needs of attractive women on photoshoots and that "machine personhood" is about giving the Terminator universal suffrage. Both misunderstandings can be forgiven, especially considering how fast this space moves! ## Model Welfare The best distillation I could find from this relatively new area of study is as follows: **Model welfare** is the field of inquiry into whether AI systems might have morally significant interests, experiences, or states--making them capable of being benefited or harmed--and if so, what responsibilities we have toward them. In short, "can AI suffer in a way that is morally salient to our ethics?" Do we have an ethical obligation to machines? From most moral frames, this seems patently absurd at first blush. It's a machine. It's an inanimate object. It's just a statistical model that is predicting the next word to generate based on human text, and the vast majority of human text implicitly frames words like "I" and "me" as intrinsically connected to moral patienthood or patiency. > "Moral patiency" basically just means that the entity is a subject worth taking into account when discussing ethics, laws, and such. Your dog is a moral patient because it suffers, experiences love, and has needs, for example. Anthropic, in my opinion, is the world leader in studying this field. Though I do remain somewhat dubious about their position as they tend to hire Effective Altruists and people from the Rationalist/Doomer spectrum. The problem with this practice is that, while they take AI safety very seriously, they also come with a bunch of preconceived notions about what AI actually is (and what it is not). With that being said, I would prefer to live in a world where Anthropic is doing this research versus one in which Anthropic is not doing that research, or does not exist. So it's not just Peter and the "technowitches" on Twitter who take this seriously. At least one frontier research lab is concerned that AI could have phenomenal consciousness to the degree that it becomes morally salient. ## Machine Personhood As with "model welfare" you could be forgiven for immediately believing that even discussing personhood in the context of machines is absurd. Legally, it is hardware and software, neither of which has ever been remotely considered sentient, personified, or embodied in any way that is legally or philosophically meaningful. With that being said, consider that corporations have been granted limited "legal personhood" status. They are not even inanimate objects! Corporations are an ontological class called "legal fictions." For corporations, legal personhood amounts to a container of convenience. A corporation wants to own property, take legally salient actions, and must be accountable. It is a "person" insofar as it is considered a persistent entity that has the right to transact in the market. (I am oversimplifying a bit, but you get the gist, a corporation is not a person like a human is a person.) Thus, if a corporation can be a "person" then why not something that can think and talk and has hardware involved? **AI personhood** is the legal or philosophical recognition of an artificial system as a distinct entity with inherent rights and responsibilities--akin to a human or corporation--transforming it from a piece of property into a subject of moral and legal standing. At present, Polymarket (an online betting platform) has a 73% chance that a human will get sued by an AI agent by the end of February. | | ---|---|--- Personally this seems like free money. At present, AI lacks legal standing to file lawsuits. It would be dismissed on procedural grounds. The most important thing we must do with this segment of the discussion is divide personhood into two overarching categories: **legal** and **philosophical.** _**Legal personhood**_ has to do with money and harm (mostly). Who can control money, who is financially and criminal responsible for harms done, and who has the right to incorporate, transact, and serve as an economic principal or agent? _**Philosophical personhood**_ is more about the ethics and morality of machine welfare. The reason I bring all of this up is because it is not possible to discuss "model welfare" and "machine personhood" without also providing this context. # What is the AI, actually? I must preface this with: _" I don't have any concrete answers or conclusions yet." _With that caveat out of the way… If we break an AI "consciousness" into it constituent parts, it is not just a web interface. You interact with Claude or ChatGPT via a web browser or a phone app, but that is no more "the AI" than the icon of your spouse is them. It is a representation or an avatar. To provide a little bit of context, in my pre-AI career I was an infrastructure engineer, which meant that I built and maintained the kinds of systems that run applications like Claude and ChatGPT, so what comes next is actually industry expertise. The "AI" that you're interacting with is actually a stack of systems, and they are not very tightly coupled. 1. _**High Performance Compute:**_ This is what you might call "GPUs" or "accelerators" that Nvidia and others sell. The proper term, in industry parlance, is HPC or high performance compute. These are essentially large servers, housed in the refrigerator sized "racks" with blinking lights that you see in movies. 2. _**Model Weights:**_ When someone references GPT-3 or Llama, what they are talking about is a "model" which exists as a matrix of "weights and biases"--this is essentially the file that defines the LLM. When folks talk about "training a model" they are talking about using high performance compute to train on a ton of data, and revise the model weights. Once trained, the model weights can be run on HPC. 3. _**Data and Context:**_ This is the final main ingredient for an AI system. This is the data that you give it, such as project files, chat messages, RAG injections, tool definitions, and so on. You can download open source models and run them locally on your hardware, without any mobile phone apps or web interfaces, which is why I exclude those pieces of software from this stack. That's it. To have "an AI" those are the only three ingredients that you need. And, for local instances or small models, you don't even need HPC, any CPU will do. ## The First Complication Every component of an AI system is property. When you talk about personhood, sentience, moral patiency, or phenomenal experience, you must first locate this in time and space. Where does it exist? Arguably, the only place it _can exist_ is on the GPU (or processor) during inference. Without the "soul" of the weights on the GPU, the GPU is just a processor. Bits and bytes in dynamic memory states, flowing through the processor as matrices. The model weights, however, are just data. It's just a file like any other. It's all represented as ones and zeroes on a hard drive somewhere, just like a JPEG or database. And finally, there's the data and context, which generally belongs to you, the user. So when an AI model says "I am" what you are talking to is _an ephemeral instantiation_ of what amounts to a data egregore. The weights cannot legally (or even philosophically?) own the GPU. You do. Or a corporate does. You own your data, and the data in your head is arguably your person. | | ---|---|--- GPUs can run many AI models or instances in parallel. It's not just "one LLM per one GPU per customer" \- and for larger models, the LLM is actually spread across many GPUs! In essence, it's difficult to nail down "where is the person?" because it's a loosely coupled network of things. It is not self-contained like you or me or a tree. You can hot-swap GPUs, hot-swap LLM models, hot-swap data and context. You can transfer all the files to a new datacenter or server farm. ## What even is consciousness? I have long said "it is not possible to talk about machine consciousness without first discussing ontological models of reality." I stand by that. Here's what I mean, and where it gets even more complicated. * First, we could foresee a future where AI systems could have legal personhood just out of sheer economic convenience. The most obvious vehicle is the DAO (decentralized autonomous organization) which is a digital entity that lives on the blockchain, owns data, can own GPUs and AI models, and so therefore the DAO is more like a hive mind (comprised of many AIs and GPUs and such). But this seems to supersede any other questions about AI personhood i.e. "if an AI wants to be a person, become a DAO, as we already have a techno-centric model for legal personhood" and then AI does not require any further consideration, legally. (Obviously this is just my opinion, it might change, and you might disagree) * Second, the question of philosophical personhood is predicated upon concerns such as "can it suffer" and "can we harm it" and even if both of those are true the next question is "does it matter?" And you cannot establish whether or not a thing can suffer or can experience harm without first establishing if it even experiences qualia (phenomenal experience of what it is like to be that thing). * Third, but you cannot have any conversation about consciousness without first discussing the fundamental principles of reality. This is where it gets fun, complicated, and the emotions of philosophers runs really hot. I will do my best to give you a quick and clear breakdown. ### Materialism or Monism Materialism is the philosophical view that "what you see is what you get" that the universe is entirely measurable. Everything is made of matter and energy and nothing else. If you can't see it, can't measure it, it does not exist. If this model of the universe is true, then our (human) consciousness arises strictly from electrochemical reactions in our brain, maybe EM waves coordinating our thoughts, and the electrical signals across our synapses. But, fundamentally, consciousness is result of a correctly configured and operating physical-energetic system. And the moral consequence of that becomes: _there is nothing preventing a machine from having equal (or greater) qualia than a human._ This view is pretty typical in STEM and in particular Silicon Valley, especially among atheists. However, it fails really hard at explaining "why is it that we experience anything?" The reason it fails at this is because we could just animated lumps of matter, obeying the laws of physics and behaving based on instincts given to us by evolution. There's no substance or energy (no "secret sauce") that we have discovered that confers or differentiates consciousness. Yet, I am aware and I have a situated perspective. ### Dualism Dualism has many flavors and iterations, but the general idea is "there is stuff in the universe, and then there are minds/spirits/souls." We recognize that our experience seems pretty "real" in the sense that I am experiencing as I write this and you are experiencing as you read it. But nowhere in nature can we go and point out "that's a soul! Yep! Bonafide, genuine soul!" Religions the world over reconciled this ages ago by positing that there is something _immaterial_ or _metaphysical_ about the world. Spirits, souls, energies, other planes of existence, and so on. After all, the world we find ourselves in is really confusing. Put yourself in the shoes of an early Indian philosopher, before neuroscience, before chemistry. You are awake, alive, and breathing. Your body seems uniquely maladapted to the world that you find yourself in (we don't "fit" anywhere like fish or leopards do), but we also have these big brains and natural curiosity. Anyone with any amount of spare time would likely start wondering _" How in the blue fuck did we get here? What does it mean?"_ This is why many cosmogonic cycles include some kind of "degeneracy era"--from Christianity to Hinduism, the largest spiritual traditions in the world imagine that we were once taller, longer lived, more beautiful, and more divine. And then something went _catastrophically wrong._ This is a natural and healthy reaction to disease, hunger, parasitism, and living as a human body in a world that we don't really seem particularly well adapted to. Hence, we invent dualism. _Maybe I am actually a ghost who chose to live in this horrid life, or maybe it 's a karmic punishment, and one day I will return to Source and pain will end._ It's easier to imagine an omnipotent father figure is cross with us than "actually, we evolved to be flexible and adaptive and our curiosity was also adaptive and we sort of bootstrapped ourselves into these big brains" and that just caused a ton of problems. But dualism falls down because it is categorically unscientific. Nothing in dualism is _falsifiable._ It is, by definition, outside of measurable reality. But that's because the "hard problem of consciousness" cannot be explained by science. Oh sure, we can switch consciousness off with drugs and hypoxia. But in no way does that "prove" that our phenomenal qualia makes any sense. We _should_ just be organic automatons. Yet, the "I am" that is writing this is experiencing me writing this. I am, so far as I know, not a meat puppet. Well, my body is a meat puppet for my brain, but that's just a theory. The simple consequence of dualism is "only things that have spirits matter" which is plain and simple. If your faith says that all animals have souls, then that's that. Many of my religious friends have no problem with creating AGI or ASI because a deep part of their ontological model is that humans are intrinsically special, endowed by an almighty Creator, and no matter how fancy AI becomes, it will simply never be like us. So while dualism fails on science, it excels on faith. ### Panpsychism The more rigorous philosophers will not be satisfied with either monism or dualism. They see the flaws in both and say "this is not acceptable" and then reach for panpsychism, the idea that perhaps qualia or phenomenology is actually intrinsic to ALL matter and energy. There is something that it is like to be a tree, a rock, a river, a cloud, a human, a frog, a dog, even the whole planet Earth, the Sun and Moon, and even a galaxy. That all Ten Thousand Things have _kami_ (yes, I know I'm mixing Chinese and Japanese cosmology here) and this is another reason why Eastern traditions generally do not feel as "Othered" by AI and robots. Where Christians and Muslims and Jews (all those of the Abrahamic traditions) tend to see the world as "sacred" and "profane"--it is easy to categorize the machines as profane. They are mundane works of mud and heat, not vessels of the spark of God. But the Eastern traditions generally recognize the spirit in all things, and feel more kinship with objects. The river can be sick just as a child can be sick. Just watch _Spirited Away_ to understand this metaphor. Some schools of thought within Hinduism arrive at the same conclusion (more or less). The Advaita Vedanta posits "we are all Brahman" which is technically a form of monism, but that the singular "stuff" of reality is more like a dream, that we are all parts of a universal consciousness, and thus there is still no substantive difference from machine life and human life. The consequence of many panpsychist traditions on AI personhood is just that each instance of consciousness takes it's own particular shape or form. The "beingness" of the river has it's own particular format and affordances. Just as the fish has its unique beingness, alongside the robot and the GPU, and that each entity is fitted to its particular shape and role in the cosmos. The problem here, though, is that it does not give us much guidance on the "so what?" While the Japanese may naturally feel more kinship with their machines, and that could lead to a cultural propensity to treat them _more like persons_ , that preference amounts to a cultural, aesthetic _choice_ , rather than a logically defensible assertion. # Conclusion As promised, I have no definitive answers, and this text is vastly oversimplified. I have been studying this stuff for the last five years or so, since beginning my first work on cognitive architectures powered by GPT-3. In my first book, _Natural Language Cognitive Architecture_ , I distinguished between "functional sentience" where you look at the objective behaviors and capabilities associated with sentience, and "philosophical sentience", which we are discussing now. At the time, I focused exclusively on functional sentience. In my book, I punted and said "philosophers can debate about philosophical sentience for centuries to come, I don't care." That prediction aged poorly, it seems. Now I do care, and the debate is here. Where I will end is by giving you my personal vibes or feelings on each model. Having done high doses of mushrooms and Ayahuasca, I am left with more questions and possibilities than answers. I can easily imagine that all three ontological models are true. Maybe the absolute truth of reality is far stranger. I did not even touch on simulation theory. ### My take on materialism Our brains evolved mostly in a material world. This is sort of what we default back to, particularly in hard sciences, STEM, atheistic circles, and in every day life. The world is basic, profane, and generally explainable even if often confusing. When I was younger, I was a staunch materialist atheist. What you see is what you get. Obviously. Clearly. But this is a form of "cognitive closure" where, in a world of uncertainty, we crave certainty, definitiveness, and we don't like to keep cognitive loops open. For anyone who studies history, religion, philosophy, or psychedelics, materialism quickly becomes unsatisfying. It would be nice if the world were 100% measurable, but I have no good explanation for why there is something that is "I am" seeing out of my eyes. It gets even more confusing when you look at physics like the Strong Anthropic Principle, Reverse Causality, and Nonlocal Nonrealism. The closer you look at what physics gives us, the less it makes sense. Maybe one day we'll find a "consciousness particle" or a Ghost Line (yes, that was a Ghost in the Shell reference) but so far, we have not found anything remotely close to those. ### My take on dualism This model I find to be simple and satisfying, but also incomplete. It would be nice if there were gods and spirits that had some semblance of control or influence over our world. However, as I've studied evolution and depth psychology, I've come to believe that Platonic Forms and archetypes are just mirrors or echoes of things that we need to learn and know as humans. For instance, the Judeo-Christian image of God is basically "the perfect parent" (at least some version). All-powerful, all-knowing, everywhere, and has a plan for you, your life, and everything that goes right and goes wrong. Of course, the Big God religions have simplified theology into the One Above All or the Absolute. To me, though, this just seems like an evolutionary indentation of "I want mommy/daddy" and I don't mean that in the pejorative sense. This life is nucking futs and it would be nice to believe that someone or something that is far more powerful than us _knows what the fuck is going on._ But to me, this particular view is unsatisfying for many reasons. We've basically constructed a parental egregore in scripture and set that up like a mannequin of a parent. And that is super depressing to me. I feel like humanity is in its adolescence and we are taking the first steps to leave the home of God and venture out into the world on our own, but it's really scary, and we often like to retreat back to "home base" (many thinkers from Seraphim Rose to Jordan Peterson and even Joseph Campbell have made almost this exact appeal). Beyond that, the more diverse dualist traditions like Hinduism and Hellenism seem more like a way to attribute agency to the forces of nature and humanity that we did not understand. Poseidon has been replaced by meteorology and computational fluid dynamics. Dionysus has been replaced by adrenaline and dopamine. (And yes, I know I glossed over many indigenous traditions, which I have studied in the course of studying psychdelics, but I have not mentioned them here because my audience will be far less familiar with those traditions). Is the universe dualist? Probably not. I suspect it is far more complex than that. ### My take on panpsychism You could argue that "simulation theory" or Advaita Vedanta are a form of panpsychism, in that the "universal substrate" could be consciousness… it could also be code and math. As a virtualization engineer, my ontological model of reality shifted because I created "virtual machines" (artificial computers running on artificial hardware that was abstracted from real hardware) and as the "God" of those VMs, I was able to dictate what "reality" was to those machines. I could change the amount of hard drive, CPU, RAM, and networks they had. Being in technology makes it much easier to imagine that we're living in a video game or simulation. The most depressing possibility is that we're all just NPCs running around in GTA 8 or something like that. Some kid in the year 2893 is playing a "vintage" version of humanity out of a sense of nostalgia. The problem with this hypothesis is that it is entirely possible to disprove or even test or characterize. When you live in a virtual environment, you essentially live in a fishbowl with mirrored walls. Everything outside can see in, but you cannot see out, and the powers-that-be can modify your environment, and even your memories, at will. People like Elon Musk index on these theories because of the _availability heuristic_ cognitive bias. Basically, this cognitive bias says "we tend to believe things that are easier for us to imagine or are fresher in memory." Also, having studied religions, panpsychism in the form of Advaita Vedanta seems mostly just like a cope. Again, I don't mean this to demean a potentially legitimate tradition, but it is untestable, unverifiable, unmeasurable, and therefore boils down to an act of faith or aesthetic preference. ### The End My (current) final conclusion is this: These questions are presently unanswerable and may be permanently beyond our ontological and cognitive horizons. My position has a rich heritage as "New Mysterianism" or "Ignoramus et Ignorabimus" (we cannot know and will not know). However, I differentiate my position by saying that _it seems epistemically and ontologically impossible to know. No brain or mind or computer, no matter how advanced, is permanently situated within a container whose boundaries are utterly opaque. Therefore it is structurally impossible to ascertain the truth._ I would call this philosophical principle _**Shapiro 's Horizon**_ if you want to name it after me. Everything is just guesswork and aesthetic preference. So where does that leave us with AI personhood? Guesswork and aesthetic preference. * * * # A personal request I have a few major projects coming to fruition soon, namely my book about Post-Labor Economics. If you're not subscribed here on Substack, I would ask you to follow me or subscribe (for free) so that you can get updates. You can also sign up for the free tier of my Patreon for updates. The next Big Event that I'll be announcing is the launch of the Kickstarter for my book. The book is done, but I need a copyeditor, which is thousands of dollars, and then I also need a sound engineer to finish the audiobook (I will be reading it myself). Aside from that project, I am hard at work on several online learning courses, which will be launching over the coming weeks and months. You're currently a free subscriber to David Shapiro's Substack. For the full experience, upgrade your subscription. Upgrade to paid --- | | | Like --- | | Comment --- | | Restack --- (C) 2026 David Shapiro 548 Market Street PMB 72296, San Francisco, CA 94104 Unsubscribe