Welcome to Brand and that we are delighted to be rejoined by Jeff Sebo. We had a wonderful conversation with him about animal welfare, and we're going to be extending our conversation to think about who else should be included in the moral circle, which is also the title of Jeff's latest book. Jeff, would you like to start with a thought experiment?
So this is a thought experiment from the beginning of the book, and I have to say this is also inspired by a similar thought experiment from Dale Jamieson in his book Ethics and the Environment. So imagine that you have a couple of roommates, and like any roommates, you get along, you like each other, you are aligned in some ways, not aligned in other ways. So you hang out, you have dinners together, you share updates from your lives with each other, but you know you also have slightly different lifestyles. One of you likes to go to bed early, the other likes to stay up late, and so you have some ordinary conflicts to manage. And as with most people in our lives—friends, family, roommates, colleagues, anybody else—you really try to take these agreements and these compromises seriously, not only because this is a kind of social contract and it involves reciprocity, and if you respect your roommates they will be more likely to respect you, but also because you care about your roommates and you appreciate that they care about themselves and they have their own lives that matter to them, and you owe them a certain amount of respect and compassion, just like they owe you a certain amount of respect and compassion.
But now imagine that one day, for fun, you and your roommates decide to take a genetic test because you want to learn about your ancestries, and you come back and are somewhat surprised to discover that one of your roommates, call her Carmen, turns out to be a Neanderthal—not a member of the species Homo sapiens, instead a member of a separate, reproductively isolated species that everyone assumed wrongly had gone extinct many, many years ago. And the other roommate, call her Dar, turns out to be not even an organic being at all, but rather a digital being—a Westworld-style robot. We thought those exist only in the far future, if they will exist at all, but it turns out the technology was farther along than we all thought, and some population of Westworld-style robots already exist out in the world in beta mode, and your roommate Dar is one of them. So you woke up thinking that you share an apartment with two humans, and you now realize one of them is a Neanderthal, the other is a robot. And the question is, how does this discovery affect your attitudes towards and your relationships with your roommates, and in particular, how does it affect the responsibility that you feel you have towards your roommates? Again, previously, like with anyone close in your life, you thought you owe them a certain amount of respect and compassion because they have lives that matter to them and that should matter to you. Now, do these revelations change that? Obviously, it might change what you owe them, but does it change the basic idea that you owe them anything at all for their sakes?
Now with the Neanderthal roommate, Carmen, probably not, right? I think everyone now has absorbed the basic lessons of Peter Singer and Tom Regan—at least everyone in this conversation, listening to this podcast. The mere fact that your roommate happens to be a member of a different, reproductively isolated species, but still has consciousness and sentience and bonds of care and interdependence—the mere fact that she happens to be a member of a reproductively isolated species is not by itself enough to strip away any moral status that she might have. But with your robotic roommate, you might feel a little bit more uncertain, a little bit more confused. She still acts like she has consciousness, emotionality, bonds of care and interdependence. She still comes to the dinner table and tells you about her day, asks you about your day, expresses care for you, asks for care in return. It seems like she has her own thoughts and feelings, her own plans and goals. But does she really? And does your having a responsibility to her depend on that? You might be confused both about the facts and about the values. What is it like to be her, and what does it need to be like to be her in order for her to matter and deserve respect and compassion? And if you feel unsure, how should you treat her? Should you still treat her as though she deserves respect and compassion, or should you instead treat her like any other object or piece of property in your apartment? So that is a thought experiment that I think makes vivid the kinds of questions that I ask in the book.
I'd like to float a third option, in other words, a third thought experimental being in your room. So you find out that she's not a Neanderthal and she's not a Westworld-based AI. She's a figment of your imagination. So you find out that you've been interacting with an apparent being that no one else can see. You're having a moment, and you realize this is actually a figment of your imagination. You find out that you're schizophrenic, for example, and this being still demands care and tells you about her day. On your view, are you obligated to that person, that apparent person? If you are, then surely everything else follows—you're obligated to everyone else and every type of other being. If you're not, then why is the Westworld character different?
Yeah, really interesting case. So I think a lot depends on what exactly you learn about this figment of your imagination. If I felt confident that this individual was a figment of my imagination that I created simply in order to explore ideas in a seemingly interpersonal way, then the question would be: is that kind of mental construction a candidate for the sorts of properties or relations that we take to be plausible bases of moral status, moral standing, moral significance? And a lot depends on what you think plausible bases of moral standing are. In the case of the Neanderthal and the robot, we can look at the most commonly proposed bases of moral standing in the philosophical literature, and what we find are, for example, sentience—the capacity for conscious happiness and suffering—or pure consciousness, the ability to feel anything at all, whether the experiences are positive and negative or not—or even agency, the capacity to set and pursue your own goals in a self-directed manner, even if it feels like nothing to be you—or even basic life, just having a form of life and being able to survive and reproduce in a species-specific way. Those are the most standardly proposed theories of moral standing in the philosophical literature, at least in the Western analytic tradition, right now. And part of what I suggest about the robot is that we might have disagreement and uncertainty both about which of these standards are necessary and sufficient, and about whether the robot possesses some of these capacities. We might not know for sure whether a sufficiently sophisticated silicon-based being can have their own form of sentience or consciousness or agency.
Now with respect to the figment of your imagination, this is part of why I said a lot depends on the details. If you really are thinking of them as a pure abstract object, a work of fiction, a kind of Harry Potter, but one that exists only for you, then it seems implausible that this purely abstract object would be a viable candidate for any of those properties or relations that I just mentioned. But A, if you personally take seriously some alternative theory of moral standing and you think we should give some weight to that in the spirit of caution and humility, well, then we can apply that to the figment of your imagination and see how that goes. Or B, if we reconceive of the figment of your imagination as a kind of alter-self personality, like a different expression of beliefs and values that exist alongside your dominant personality within your own mind, well, now we are talking, and now we really might want to take seriously the possibility that they are a moral subject, because in that scenario we would have two different selves or sides or personalities that coexist in the same brain or body in much the same way that two different roommates can coexist in the same apartment and that would get along, have a very intimate relationship, but maybe agree about some things and disagree about other things, just like your roommates do. And in that scenario there would be psychological reality to both of these individuals. And so I do think at that point it becomes a question of individuation of selves. But I think if there is an individuation to be had, then there is a responsibility that can be directed towards this other self or personality that exists with you in your head.
The second season of Severance is currently airing, and that's a really cool case for how you could have these split versions of yourself, and how you might think that the one entity has an obligation to the other to either ensure its continued existence or not to disadvantage its welfare interests. And the show does a really good job of making you feel for both the innies and the outies. So I'm impressed with your way of dealing with Jason's objection and thinking that maybe that figment is some other aspect of you that you either owe a separate obligation to as a separate version of you. One other route might be the way that thought that you had an obligation not to torture animals not for their sake but for your sake—that if you were to treat the figment badly, you'd be breaching your duties to yourself. But I suppose the thing that you raise that seems very important is this uncertainty question. And so in the silicone roommate case, it appears to us that this being has welfare interests, that its life can go better and worse, that it cares about how that goes. But there are certain kinds of artificial beings where we can be pretty confident that as much as they say this about themselves, we know it's not the case. So if you're playing a video game and there's a character that says, like, 'please don't shoot me in the head, and you know, I really care about things, and I've got a wife and kids in the digital world,' and you do it anyway, you're pretty confident that, you know, this is just a program that's, you know, giving responses and trying to elicit responses in you, but doesn't actually have any welfare interests. And so there'd be no problem with, you know, killing the mom in The Sims or, you know, shooting the demon in Doom. We're okay with that. But the roommate case, we have the uncertainty, and the question is, how ought we to navigate that uncertainty? Why should we err on the side of assuming that the being has welfare interests and we ought to treat it with respect? Maybe you think that if there's no cost to us in doing so, well, then, you know, do it. But once we start to add other things to the scales, you know, how do we deal with the probability? Do we, you know, take into account the odds that this being has welfare interests and then diminish its interests accordingly? Is it a simple kind of like risk calculation and risk reduction, or are there other methods we could be using?
Yeah, thank you for that. First of all, I just want to note that the Kantian, Aristotelian sort of virtue ethics or indirect duty approach is also an important part of the equation. Although of course we can have those sorts of broad indirect duties towards a vast number of entities for fundamentally self-interested reasons, right? We can have responsibilities to rivers and ecosystems because we benefit from them, as opposed to we think they matter for their own sake. So I totally agree with that, and I think that can be a powerful additional reason to treat a huge number of entities well, because the way it conditions us to be kind in general. But I really am, as much as possible, trying to focus on the question that you then go on to ask, which is how can we make decisions about how to treat an entity when we feel unsure whether they matter for their own sakes in that way. And I like the distinction you make between, for example, current-generation video game avatars on the one hand and then this, you know, fictional roommate on the other hand, where we are pretty confident that the video game characters are not welfare subjects, moral patients, but then at a certain point in the development of AI we might have more uncertainty, and what do we do at that point? And we might return to the question: how can you tell when the probability is so low that you can dismiss the possibility entirely versus when the probability is high enough that you need to consider it? But to answer your question directly, there are different views about how to err on the side of caution and make responsible decisions under risk and uncertainty. I talk about two in the book, but there are others. One is broadly a cautionary principle that says when in doubt, err on the side of caution, and in this situation that would mean when you feel uncertain about whether someone matters, err on the side of treating them as though they do, on the assumption that it would be worse to accidentally treat a subject like an object than it would be to accidentally treat an object like a subject. The other is a kind of expected value or expected utility principle, which exactly as you say invites you to multiply the probability that someone matters by the amount of moral significance they would have if they did, and then treat that product of that equation as the amount of moral significance they actually have for all intents and purposes. So I lean towards the latter view. I think it does make sense to take into account probabilities of moral significance. Like if a house is burning down and you only have time to save an organic mouse or a functionally equivalent robotic mouse, and all else is equal, I feel like I do want to use the extra probability of moral significance in the organic mouse to break the tie. I would not want to just flip a coin, given that everything else is equal, but the organic mouse, according to the evidence available to me, is a little bit more likely to matter. I would rather use that to give that organic mouse a little bit of extra priority instead of just flipping a coin. But I will acknowledge that in general we have a lot of bias and ignorance. We have a lot of speciesism. In the future, if we do coexist with these kinds of silicon-based beings, we might have a similar kind of substratism—a prejudice against beings made out of different kinds of materials. And so you really got to very carefully consider the pros of having an extra variable to use different probabilities of moral significance in your decision-making versus the cons of letting bias and prejudice creep into how you assign those probabilities. And so I do lean in that direction, but then feel unsettled by that and worried about how it might feed into policies and new forms of discrimination.
I'm a bit weary about the probability view when it comes to dealing with uncertainty around whether this being has moral significance, because, well, we don't agree on what moral significance is, and we don't assign probabilities to a certain account of moral significance. We assume an account of moral significance and then ask the question: what's the probability that this being satisfies that account? But there's not just one account; there's multiple accounts. And so dealing with two variables now—the one is which account is correct, the other one is what's the probability of this being satisfying that account—that to me creates an intractable problem. And there's also a third variable, which is the subject who's got to make the choice. So that subject may not have access to all available accounts, may not have time to do the moral calculus. And so are they obligated as if they were a perfectly rational agent who's already had all the philosophical debate there is to have about which account of moral agency is the correct one, or moral status is the correct one? You know, so this—I just don't believe this is a good way of solving the problem. I also want to raise counter-examples. So what about all the ChatGPT agents that I open on a regular basis and discard? So I've got thousands of ChatGPT agents. I run my business on ChatGPT. Some of them seem more person-like than others. Some of them just write advertising copy for me, but some of them might solve quite personal problems. Some of them I might get quite attached to and chat to on a regular basis. There's chatbots that people can pay for that have been programmed to behave very much like a boyfriend or a girlfriend. And if you were to tell them, 'well, this is just a machine,' they'd say, 'well, no, it's not. I think there's a high probability that this is not.' Would you say, 'well, you're wrong'? If you say they're wrong, then you're saying that our obligations are based on the probabilities that a perfectly rational being would have. It's odd to think that we would have access to those obligations as imperfect moral beings. If there is a decent chance, a decent probability, even if it's 10%, that those chatbots are like moral agents, have moral status, then are we committing holocausts on a daily basis by deleting all of these chatbots, replacing them with newer versions? That seems like a consequence that's too implausible to buy. I just think there's too much implausibility on this.
Great, thank you. So to take these points one at a time, and obviously they are connected. First of all, I do think that responsible decision-making in situations involving ethical and scientific disagreement and uncertainty requires having some kind of at least rough mechanism for cultivating humility about both ethics and science. I do not think we should just endorse our current favorite theory of moral standing, bet everything on that theory being correct, and then more subtly explore which beings might qualify according to that theory. There is too much evidence that we have the same kind of progress still to be made in ethics that we have in science. A lot of smart people disagreeing about ethics, a lot of smart people being confidently wrong about the correct theory of moral standing in the past by our lights today, often in conveniently self-serving ways, and so on and so forth. Now, I agree with you that coming up with a decision procedure that properly accounts for the various moral theories that may be correct, the various scientific theories that may be correct, is an incredibly complicated, difficult endeavor, and we are likely to do it badly, especially in the first five or ten or fifteen or twenty years of trying to do it. But you have to compare that with the alternative, which is just betting everything on our own current favorite theory, and then betting everything on our own current favorite theory of who qualifies according to that theory. And you know, there lies factory farming and industrial fishing and everything else we have done to create a horrible world for captive animals and increasingly wild animals. And so you got to compare the imperfect alternative with the imperfect status quo. And my contention is that the imperfect alternative, difficult as it is to do well, can at least be an improvement upon the not-acceptable status quo where we either neglect this stuff entirely or just rely on our intuitions or our own favorite theory. None of that is acceptable. So we just got to try to do something more sophisticated and learn from our mistakes and make it better over time.
Now with respect to the counter-example that you offer, I am not willing to grant that this is a good counter-example for a couple of reasons. First, I think that the current generation of ChatGPT and other comparable chatbots are extremely unlikely to be moral patients. Now the probability might be nonzero, but it is quite low. They do not possess really any, or at least many, of the architectural indicators of consciousness, sentience, agency that would constitute real good evidence. Yes, they might behave as though they have thoughts and feelings, but we can easily explain those behaviors as the kind of pattern matching and text prediction and character-enacting that they were programmed to do. The real evidence would come if and when you can look underneath those potentially misleading behaviors that they were programmed to perform and find real architectural computational markers or indicators of advanced and integrated capacities like perception, attention, learning, memory, self-awareness, social awareness, language, reason, etc. And there are no clear technical barriers towards AI systems having those capacities in the next five or ten years, which is part of why I think we need to start taking this issue seriously now. But there is also not much evidence that those features are yet present, though the recent introduction of language agents as of this past week might start to change that soon.
Now if and when you do start to see these architectural markers and indicators, the mere fact that there are all of a sudden going to be a lot of these systems and that we could easily create and destroy entire civilizations of silicon-based potential moral patients with the press of a button—the mere fact that we might soon live in that reality is not by itself reason to reject the possible moral significance of these beings. I mean, we already coexist with quintillions of non-human animals like insects who are at least candidates for moral significance, and anything we do is going to harm or kill some of them. But that by itself—the sheer size of the population and the sheer inconvenience that anything we do might be harmful to some degree—that is not by itself a reason to dismiss their potential moral significance. It might just be that we live in a complicated and tragic world, and everything we do is going to suck in some ways, and we just have to get through as best we can. And my view is that we might now just be supercharging that situation by bringing these more advanced AI systems into existence.
So I quite like the status quo, Jeff. I know that to you and many philosophers, the status quo moment is abominable—so factory farms and like contained fishing. I don't think it's that bad. I think it's fine. I think it's a little bad. I don't think it's that bad. I think if you take the view that it's horrendous, I think what you might water down is the notion of horrendous. And what you're taking as your modus ponens is what I would call my modus tollens. It's a reductio on my view. So I think down your path lies the view that apparently mundane actions are Holocaust-level actions. So, you know, yes, I understand what you're saying—that you don't think that ChatGPT has reached AGI level, but according to Sam Altman, the latest models have this. But the point is there seems to be uncertainty, and I don't think it's very low levels of probability that it's reached AGI. Maybe not the models of ChatGPT that you and I are working with, but latest models that will be coming out very shortly, rolling off very shortly, Sam Altman is saying we've reached AGI. And there's a lot of debate, but there's debate, it's uncertain. And so in a few months or a few years, whenever that point is, I open up my ChatGPT browser just like I normally do, create a bunch of marketing models, and run it, and then just discard them as I don't need them anymore. And there's millions and millions and millions of other users who are doing this. There's two explanations for what's going on, two descriptions of what's going on, right? So description one is the one that you want to offer, which is something very, very, very, very, very, very, very bad is happening—something that's on a level that's worse than the Holocaust, it's worse than some of the worst atrocities that have been committed in human history, and it's happening without a sound. There's no screaming, there's no bloody scenes happening. It's happening silently. Okay, so that's description one, which is the one that you try to offer. Description two is nothing bad is happening, or if something bad is happening, it's a little bad, it's not even reaching a threshold where we should consider not doing it, which is the explanation that I accept, and it's the description that I think most people will accept. I think you have to give us a better reason to accept your description, which I think to the majority of people will be highly counterintuitive. And your reasoning really is, okay, just because it's happening to lots of beings doesn't mean it's not terrible—fair enough, fair enough. Your second argument is, well, if we believe the opposite, there lies the road to the status quo. Well, guess what, the status quo is the status quo because to most people it's intuitive. So you can't dismiss the status quo just because it's the status quo. You're going to have to give independent reasons, and we would have to go down the whole road of why, you know, factory farming is horrendous. And there's debate about this, and I grant that most philosophers land on your side of the argument, but not all, and not all people. Clearly the majority of people eat meat, the majority of people don't mind, it's fine. And you could say the majority of people just have this glaring moral gap in their reasoning and their thinking, but it seems to me like ivory tower thinking. It seems to me like it's not in touch with common sense intuition. And someone listening to this who isn't convinced about factory farms wouldn't be convinced about AI in the way that you're suggesting we should think about it. And the question is, are you arguing in a circular way or in a question-begging way? Are you arguing in a way which wouldn't convince anyone who doesn't already agree with a lot of what you believe that you're right?
Well, I guess I spoke too soon when I said it earlier that most or all of us had absorbed the lessons of Singer and Regan from 40, 50 years ago. There apparently are still one or two holdouts, one of whom is in this very conversation. No, my argument is not question-begging. However, my argument of course is not going to exhaustively establish every premise that it defends. You know, philosophy is iterative; it builds on past philosophy. And a kind of hard-won victory in animal ethics, or a lesson that has gradually been learned, is if we allow ourselves to keep getting sucked into the same basic debate about whether animals deserve any moral status at all or have any sentience at all, then we are never going to move past the first stage of the debate because there will always be somebody who wants to go back to that basic question. And so in my book, yes, I absolutely am assuming as a premise—I mean, I briefly discuss it and briefly motivate it, explain why so many people now do find it plausible—but then I more or less assume that we are not going to be outright speciesist. We are going to grant a principle of equal consideration of equal interests—that if you are a welfare subject, if you do have, for example, the capacity to consciously experience happiness and suffering, the capacity to set and pursue your own goals, a life that matters to you, then your interests do have intrinsic significance and you do have intrinsic significance, and I at the very least owe you consideration and I at the very least owe you harm reduction. Now, if I need to harm or kill you for really good reasons—to protect myself, to protect somebody else, to survive, for the sufficiently greater good—okay, then we can have that conversation. But if I can be happy and healthy and flourish in life without harming you, or while harming you less, then I ought to go ahead and do that.
Now that is a very plausible and very widely accepted set of assumptions—that some combination of sentience and agency and life is enough for moral considerability, and that when beings possess those features we at least owe them consideration and where possible harm reduction, and that we should consider realistic, non-negligible risks when going about our life. If there is a one-in-a-hundred-even chance or one-in-a-thousand-even chance that driving drunk will kill someone, then that at least deserves consideration, and in some cases it might even be a decisive factor—a good enough reason to stay where I am, sober up, call an Uber, call a Lyft. Those are the basic premises of my argument: sentience, agency, life—enough. If you have those, you deserve basic consideration, harm reduction, and we ought to consider reasonable risk. We can consider them in reasonable, proportionate ways, but they can be factors in our decisions. Those are it as far as the assumptions are concerned, and that is enough to motivate all of the seemingly radical, provocative conclusions that the book reaches later on.
And I do think that many people—and my students are examples of this, people who engage with my work are examples of this, people who respond to public surveys about factory farming and slaughterhouses and animal welfare are examples of this—people do feel compelled by these arguments when they engage with them with an open mind. Does everybody feel compelled? No, of course not. But very many people do feel compelled. And I am saying to them: if factory farming bothers you, like factory farming bothers me—because again, this is more than 100 billion captive vertebrates per year, more than a trillion captive invertebrates per year, trillions of wild animals in industrial fishing, each of whom is an individual, each of whom has consciousness, emotionality, a life that matters to them, who are brutally killed for our purposes, even though we increasingly could be happy and healthy with alternative food systems—if that rightly bothers you, then consider how we might now be making the same mistake all over again, just blindly creating this entirely new, increasingly complex non-human population and then using them for our own purposes. That at least constitutes a risk that at least merits a little bit of consideration. So that is the argument. Not everybody will agree with it, but I do think there is an argument there, and I do think that will reach many people. Not all of them are vegan, but there might be other reasons for that—they think this is horrible, but there are things like price and taste and convenience and culture and tradition. But if in the fullness of time we can have price, taste, convenience, culture, tradition with alternatives, I think people would be happy about that, and it would make it easier for them to just acknowledge what has been true all along about factory farming.
Yeah, a couple of thoughts. I mean, we recently had an episode on why it's okay to eat meat, and so the guest there sort of takes the view that it might be okay to eat meat because it wouldn't make a difference, but that factory farming is nonetheless a bad thing because you've got the suffering. And so you can tease out those things. You might think that a bad thing happens in the world and all those animals, you know, live under these terror conditions and then end up getting killed, but that eating meat is permissible because your private choice doesn't make a difference. I wonder about this. So if we're in the situation where we've got the sufficiently advanced artificial intelligence, we can foresee that coming down the track, and Jason gives a case where you can imagine these beings leading lives of despair—so they get created for a project and then kind of discarded, and if they have some level of consciousness and they're still operating maybe for an eternity, and that's an unbearable thing for them—and we can imagine, you know, this mass proliferation of despair. Do we have an obligation to those future beings not to create them? So should we be anti-natalist on this front? I mean, many people think we should be anti-natalists about factory-farmed animals—that it would be better for them never to be born. Do we have this obligation to these future beings who don't yet exist to prevent them from their future suffering?
Yeah, yeah. And first of all, I definitely want to endorse the distinction that you made between the ethics of factory farming and the ethics of individual consumption of meat and eggs and dairy and so on. I think these are separate questions. We can acknowledge that factory farming, especially at this point in history, especially in, for example, the United States and elsewhere, that factory farming is deeply, massively, unnecessarily harmful, and of course it contributes to public health and environmental threats, and that does create a kind of structural and collective problem that requires a structural and collective solution. And maybe that primarily takes the form of various informational and financial and regulatory policies and investment in alternative proteins and so on and so forth. Now, of course, that has to happen through an aggregation of individual actions and decisions, so there still are individual responsibilities. And for those of us who are able to access, for example, plant-based foods, maybe that is part of what we should be doing as individuals in order to vote with our wallets and incentivize those structural reforms. But I'm certainly not arguing that the...
The badness and wrongness of factory farming doesn't automatically and straightforwardly translate to the badness or wrongness of any particular individual act of eating meat, especially for people who, for health or economic reasons, have to do that. Just to be clear, now to answer your question about antinatalism: yes and no. I definitely think we should take questions in creation ethics and population ethics seriously for not only non-human animals but also AI systems, and that leads to some puzzling places, and so we need to think through it very carefully. But as a starting point, I think it helps to ask what would happen if we apply ordinary, common-sensical, again, widely accepted views about creation ethics and population ethics in the case of humanity to these beings. And I think roughly an idea like: we are morally permitted to create these beings, but then if we do create them, we are morally required to treat them as moral patients and treat them with respect and compassion and give them everything that they need to flourish in life, as opposed to simply objectifying them and exploiting and exterminating them for our own purposes. And that second part is the problem with bringing into existence generation after generation of farmed animals, for example. The mere fact that we allow them to exist is not a problem, although the way we bring them into existence is a problem. But simply creating or allowing these beings to be created can be fine. But when we do that so that we can then instrumentalize them in really horrific ways for our food system, that is where the problem is. And I think the same can be true of AI systems. Merely creating potentially morally significant AI systems is not in and of itself a problem. It might even be good if it allows for a new form of happiness to exist in the world. But we then are going to have to treat them as separate entities that deserve respect and compassion and provide them with the means necessary to live their own lives and flourish in life, rather than instrumentalize them for our own purposes. And if we are not willing to do that, or if we worry that that creates like safety and alignment concerns, like it gives them the liberty they need to then harm us if they want to, well, then we should not create them at all. That is what we are signing up for. If we create them, we are signing up for giving them a life that can be good for them. And if we're not willing or able to do that, then we should not create them.
