Offloading Ourselves

What does it mean to be human?

Are we defined by our noble reason or our admirable form? Or is our capacity to love more essential? If we are making sand into machines that begin to surpass us, should that exalt us above the dust we are, or humble us even lower?

To use the most advanced models of artificial intelligence — whether in chatbots, self-driving cars, autonomous weapons systems, or protein-folding predictors — is to be both impressed by their capabilities and unsettled about our own. Ordinary citizens are asking what this means for themselves, their families, and their communities. Many ethics councils and committees have been formed at universities, government institutions, think tanks, and AI companies themselves, taking positions on practical issues like job loss, bias, and the tradeoff in AI development between speed and safety. Such concerns are vital, but they do not address the deepest questions AI is raising about human meaning and purpose.

It is these questions, and their need for Socratic consideration, that have led the American Enterprise Institute in Washington, D.C., to convene a new Council on AI Ethics. Our starting point is the jarring experience of encountering an uncanny technology that mirrors us, imitates us, and in some ways surpasses us. Before we can answer “What should we do about AI?” — the focus of most other AI councils and committees — we should first have a better grasp of the answer to “Who are we, and who do we hope to become?” In a pluralistic society, that “we” will have many different forms. Yet there are shared concerns that arise from our shared human nature. Asking the right questions leads not to a single set of right answers, but to a set of ways in which we might understand ourselves and our goals, and thus a perspective from which we can see paths forward.

We begin with a focus on language-interface AI chatbot products such as ChatGPT, Gemini, Claude, and Grok, because they are widespread, powerful, and unsettling, at least to many of us, in many of their uses. OpenAI trained the GPT-3 large language model in 2020, but it was only when the company released the chatbot interface in 2022 — only when anybody could talk to the model and receive a conversational, human-like response back — that AI exploded into popular awareness. At the very time that they are attracting historic levels of investment and attention, they are challenging what society once took for granted as uniquely human.

To come to terms with the unease this challenge brings about, one must consider the nature and significance of the human person. The chatbots mimic empathy, even though they are incapable of entering into another person’s inner life. Newer coding agents with a chatbot interface possess a kind of agency through their ability to make “tool calls” and control computer and robotics systems. Our council’s main concern is with these aspects of AI that are designed to appear human-like, at least by analogy, and that call into question our privileged status as language- and tool-using animals.

Novel applications of anthropomorphic AI can make the ground beneath the most common and basic human experiences seem like quicksand. Recall or imagine what these scenarios might feel like:

• You are in class. Your friends tease you for always sitting in the front row. You look up to your teacher, re-set your eyeglasses, ready your pencil and paper, and focus your attention as though her words are wisdom.
• You are in church. You show up every week for the routine and the community but also for something more. You seek both grounding and perspective, a glimpse beyond the horizon, a truth that is love.
• You are in love. It’s more than just a fling. You feel perfectly known and accepted, and for that very reason challenged to be more than you are, more than you have believed yourself capable of becoming. You’re scared of what commitment might mean, yet you can’t imagine life without this person.

The specifics will vary, but the core experiences of being a student, a believer, or a lover are culturally universal. Now consider how AI unsettles these experiences in ways scarcely imaginable to our ancestors:

• The PowerPoint, more feed than lecture, uses memes to get ideas across. Your teacher’s evaluation metrics are now filtered through ChatGPT and other ed tech partners. When your class is instructed to take out your phones and do a “progress check” quiz, many of your classmates don’t bother to put their phones away, and your teacher pretends not to notice.
• The sermon is well-structured, with a depth and breadth of quotations and allusions that suggest deep learning. Yet the delivery feels forced and fails to land. You politely ask your pastor afterward what was different. You are told that, well, St. Paul in his letters made clear what was from him and what from the Lord; it was an especially busy week, and so the sermon was from the AI.
• The companion regrets to inform you: Your exchange has gone on too long and your tokens have run out. If you desire further positive regard, you will need to purchase a higher tier of service and acknowledge the updated terms and conditions.

As a reader of this (human-written) AI ethics inquiry, you may have one of two reactions. You may find the AI lover undignified, the AI sermon cringeworthy, the AI lesson-plan uncaring. Or you may find these examples to be efficient, practical, even exciting new approaches to life. But what you likely cannot yet find them is normal or banal. In time, more and more people will shrug off these changes without wondering what they signify. If instead you think that we should wonder what they show us about what it means to be human, then keep reading. Wisdom begins in wonder.

Our Council on AI Ethics holds the democratic belief that the gut reactions of ordinary citizens to the transformations they see occurring around them are worthy starting points for ethical reflection. This view, which traces back to Aristotle, was applied to technology ethics by Leon Kass in his argument against human cloning. Kass used the phrase “the wisdom of repugnance” to suggest that gut feelings and intuitions have meaning that is worth taking seriously. They may provoke more than they prove, but they give us reason to pause, discuss, and reflect with our fellow citizens. Reflecting upon feelings of unease helps us understand the integration of emotion and reason, and raises questions about our nature, character, and flourishing.

Michael Sandel, in a recent talk, asked “Will technology change what it means to be human?” and then guided his audience in thinking through some examples that were unsettling in hard-to-articulate ways. He concluded by suggesting that if new applications of AI are “already leading us and our children to confuse virtual community and human connection for the real thing,” then “we may lose something precious about what it means to be human.” We share Sandel’s concern, and, even more, we share his approach.

Stories, both real-world and plausible hypotheticals, best reveal the questions arising about who we are as persons. Many of the most important ethical insights on AI are being offered not in white papers or academic journals but in personal reflections. Narratives, in what follows, along with well-developed case studies the council will offer in future work, allow for an inductive exploration of how our use of technology can help us in, distract us from, or thwart our efforts at living in accordance with our gut reactions and our considered judgments, rather than in tension with ourselves.

We seek to discuss philosophical questions without philosophical jargon, but our approach does require some presuppositions and background beliefs about which we should be forthright:

1. You are what you repeatedly do. Thus, we consider how initially one-off actions may over time have displacing, longer-term effects on habits and dispositions.
2. We are rational and dependent animals, whose autonomy is best understood as accomplished through relationality. Thus, we foreground concerns about embodied care.
3. In inviting readers to think critically about how AI use will shape their understanding of the good life, we presume that having a vision of one’s good is of paramount importance to acting well in the world. We do not presuppose which vision of human flourishing is the most adequate, or how society as a whole should respond to AI, even while we offer some examples of possible reactions.
4. Without offering a comprehensive definition and description of human nature, we nonetheless stand in opposition to the expressly transhuman or posthuman ideologies of Silicon Valley that boil down to “if you can’t beat ‘em, join ‘em.” This belief is literal, not metaphorical: Sam Altman, CEO of OpenAI, has written enthusiastically about “the merge” between man and machine, and has co-founded a company, Merge Labs, to bring it about. Our humanism takes intentional action with due regard for right relationships with other persons to be central to human excellence. It is up to the reader to consider in what that due regard consists and what makes relationships “right.”
5. While human capacities are remarkably constant throughout history and across civilizations, the forms in which they are expressed can vary widely. We agree with the claim made by Antón Barba-Kay and others that culture-wide adoptions of technologies can be so significant as to become second natures, formations of habits so deep that they can be hard to notice, let alone change. The importance of the historical shift from oral to literate cultures is widely understood; that we are now undergoing a transition from literate to digital cultures is clearer each passing day.

We thus consider how the adoption of AI, particularly chatbots, is exposing what many people agree are dangerous weaknesses in our current moral, social, and political orders. From loneliness and community breakdowns to economic inequality and centralizing power, enthusiasts often propose AI as a solution to those problems even though it’s obvious that AI could make them worse. For example, Mark Zuckerberg, whose companies Facebook and Instagram have arguably helped to erode real-world friendships, has suggested that Meta’s AI chatbots will fill the void. We suspect that most readers will share our reaction of concern rather than enthusiasm at this prospect. Yet in other cases, there are genuine open questions to explore, where arguments can be made that wise use of AI tools will dignify rather than degrade our natural capacities. It is agreement on which questions are worth considering, rather than on what answers they should receive, that makes possible a shared moral inquiry and civic life.

With the stakes and goals thus set, we borrow from the Bard once more and turn to consider this “brave new world, that has such people in’t.”

1. Formation and Information

What goods do we trade off when embracing AI for efficiency and cognitive advantage?

When asked to comment to investors on a 2025 quarterly earnings call about Meta’s AI-powered Ray-Ban glasses, Mark Zuckerberg replied:

I continue to think that glasses are basically going to be the ideal form factor for AI because you can let an AI see what you see throughout the day, hear what you hear, talk to you…. I mean, I personally think that — I wear contact lenses, I feel like if I didn’t have my vision corrected, I’d be sort of at a cognitive disadvantage going through the world. And I think in the future, if you don’t have glasses that have AI or some way to interact with AI, I think you’re kind of similarly — probably [will] be at a pretty significant cognitive disadvantage compared to other people who you’re working with, or competing against.

Certainly more information and faster information processing can give a decisive advantage in markets and warfare. But would taking off the AI glasses be like taking out one’s corrective lenses — would it always be a diminishment?

Now consider a different view of cognitive disadvantage. On September 3, 1894, Frederick Douglass delivered an oration to celebrate the founding of the Industrial School of Manassas, whose mission was to educate the descendants of freed slaves. He explained:

If man is without education, although with all his latent possibilities attaching to him, he is, but a pitiable object; a giant in body, but a pigmy in intellect, and, at best, but half a man…. To deny education to any people is one of the greatest crimes against human nature. It is to deny them the means of freedom and the rightful pursuit of happiness, and to defeat the very end of their being. They can neither honor themselves nor their Creator.

For Douglass, cognitive advantage is not competitive but emancipatory. Education allows for the soul’s uplifting, for ennobling the mind so as to appreciate the harmony of the universe and the goodness of its Creator.

Cognitive offloading, defined as a physical activity that reduces the mental requirements for information processing, has been with us as long as the stylus and tablet. AI has radically expanded these possibilities, so much so that they may blur into a form of cognitive outsourcing, which previously required trusting other people for judgments about the news, history, or what one should do. The moral question here turns on whether we only care about the final outcome of an action, or whether the way in which we reach that outcome also matters. Which sorts of thinking are only about the destination, and which are also about the journey?

Put another way, the possession and acquisition of knowledge can be looked at instrumentally or intrinsically. Knowledge can be merely a means to an end, or it can be worth having for its own sake, whether or not it also helps accomplish other goals. The first is Zuckerberg’s view (of knowledge as competitive cognitive advantage), the second is Douglass’s (emancipatory education). The first is instrumental, the second intrinsic.

The information revolution, begun over half a century ago and perhaps reaching its apex with AI, is an apt phrase. We are awash in information, needing algorithms to sort it and AIs to synthesize it. As a result, a dramatic change has taken place in how we are in-formed — that is, what form of mind we take. We borrow computing metaphors like “data dump” and “memory is full” to suggest that the mind is an inferior computer, and that technology that improved our input–output rates would be good by definition. Yet this is a genuinely revolutionary view when compared to the classical understanding of reason as the soul’s charioteer, the medieval belief that charity perfects our prudence by directing it to our supernatural end, or the Enlightenment ideal of deliberative rationality that leads to civil discourse and self-governance. Today’s common-sense embrace of efficiency would in previous ages have been met with the question: Efficient to what end?

Consider a novel use of AI for efficient competitive advantage: identify my targets. Israel’s Lavender and America’s Maven systems, powered by AI, reportedly changed warfare by enabling a dramatic increase in the tempo of targeting in Gaza in 2023 and Iran in 2026, even when compared to the pace of the 2003 “shock and awe” campaign in Iraq. Neither system was fully autonomous; a human was in the decision-making loop, responsible for choosing to fire each missile or shell.

In this case, we see that outsourcing information analysis to the machine has brought a huge gain in competitive advantage. But it has also posed a worrisome loss of emancipation broadly understood — of ownership over our actions. Maven, for example, seems to reassure us that there is a “human in the loop,” but news reports suggest that the system does not offer nearly enough time for the operator to verify the information the AI uses and the interpretation it makes. The design effectively reduces the human in the loop to a button-pusher, while relocating the judgment of acceptable risk from human prudence to the algorithm.

This is unsettling. It poses the obvious risk that an AI will mistakenly injure or kill civilians in war. Whether these systems’ error rates turn out to be 10 percent or 0.1 percent certainly matters. But just as worrisome as the mistakes that will surely result is the question of who will be responsible for them. Who can be made to answer for a decision made by a machine? It is the principle of outsourcing understanding and offloading responsibility in such a grave matter that we likely find most disquieting here.

2. Outsourcing Understanding

Could we use AI to earn straight A’s but find in the end that we flunked life?

Recall the last time you were moved by the wisdom of a great teacher. Was it because of the number of facts he rattled off? The slick styling of his TED talk? Or was it from a depth of insight that demonstrated personal experience, long hours of reflection, and careful attention in delivering his message?

Would it change anything if, when you went to thank your teacher, he said that his lecture had been written by ChatGPT?

We could express our unease at this kind of cognitive outsourcing by asking whether it will degrade the quality of the teacher’s lecture, or whether it will trivialize your relationship with the teacher. But perhaps the deepest issue is about how repeatedly using AI will affect the teacher’s qualities as a leader. Will using AI make him more introspective, more thoughtful in preparing his messages, more practiced in his delivery, and thereby strengthen his ability to perform his vocation? Or will it diminish him in all these ways, weakening him in his vocation? Taking the time to think — to reflect on his own experience, his audience, and his intended theme — is a reflexive action: it shapes the speaker as much as the audience.

The problem is particularly acute in the case of spiritual teachers like pastors and rabbis, where the messages of mind and heart are so clearly linked in one vocation. But the issue is much more general. On the margin for any one given task — especially when time is limited (and when is it not?) — using AI might seem like the best, or increasingly, the only option. In a culture that prioritizes speed over depth and quantity over quality, it becomes very difficult to fault anyone, including oneself, for using the most expedient tools available. Moreover, we may tell ourselves, “Well, the way I use it does make me think, reflect, and write better, over time shaping me in positive ways,” and we may be right in doing so. If so, what would “better” and “positive” look like in practice? How might we know?

Perhaps the best argument that AI will be good for our mental capacity is this: If, as Steve Jobs put it, the computer is a bicycle for the mind, then AI is an e-bike. It can assist you in going farther and faster, or it can let you coast. YouTube has endless rabbit holes of purely wasted time, but it also has many genuine networks of gardeners, tinkerers, and builders who help pass on a practice, sharing their love of an intrinsic good with others. Maybe we will learn how to use AI like this. OpenAI has at least gestured at this possibility with its Socratic-inspired “study mode,” in which ChatGPT asks guiding questions rather than giving direct answers. If we wish to keep going further in this direction, we will need to ask and answer the questions in T. S. Eliot’s play The Rock: “Where is the wisdom we have lost in knowledge? / Where is the knowledge we have lost in information?”

For natural scientists, AI is likely to be revolutionary. Across the sciences, there are many potential uses of AI in processing and modeling data. DeepMind’s AlphaFold won its creators a Nobel Prize for its ability to predict the 3D protein structure a given sequence of amino acids would fold into. M.I.T.’s FlowER is gaining the ability to predict the step-by-step flow of electrons in complex chemical reactions. Lila Sciences has raised $200 million to systematize the development of these kinds of tools, and to automate experiments based on their findings, carried out in laboratories run by robots. In the social sciences, automating research — as for example the “AI Agents for Economic Research” in a paper with that title by Anton Korinek — is already becoming popular and successful.

But whether the AI promises certain goods, like new medicines and materials, or information, for example economic insights, we will have to ask what we could lose if humans were to take a backseat in the process of scientific discovery. First, we could lose the character-forming habits of scientific inquiry, the particular excellence that comes from a deep dedication to the pursuit of truth. Second, we could lose clear human responsibility for the uses to which scientific knowledge is put, from novel biological weapons to AI-powered genetics that can easily slide into eugenics. Third, we could lose an understanding of how science fits into our broader social and political arrangements, how it must be integrated in order to serve the common good. For example, Emilia Javorsky argues in an essay on AI and cancer that assuming superintelligent AI will cure cancer vastly oversimplifies the problem. She notes that “if a potential cure were discovered [using AI] in a laboratory today, there is no guarantee we would recognize it as such, nor that it would successfully navigate translational research, regulatory approval, manufacturing scale-up, reimbursement negotiations, and clinical adoption to reach patients who need it, at the moment they need it.”

If we turn next to the kind of emancipatory education that is meant to create free citizens with a commitment to independent thought, then different considerations come to the fore. Observers of the academy have long raised concerns that the liberal arts are in decline, humanities budgets are shrinking, and higher education is corporatizing. Philosopher Jennifer Frey found that students in her Honors College at the University of Tulsa almost never used AI to cheat because “when students realize their own humanity is at stake in their education, they are deeply invested in it.” But the successful program was shut down due to bureaucratic re-prioritization.

The family of disciplines meant to help individuals ask the big questions of life — philosophy, religious studies, history, literature — have more and more sought to mimic the hard sciences. Adopting the scientific “knowledge production” model has been an odd fit for disciplines that historically understood themselves to be directed at timeless questions. As quantity gained emphasis over quality, humanities scholars were pushed toward a narrowing of their scope of research into disciplinary silos, and “interesting” became more common praise than “insightful.” Over-reliance on AI may over time hollow out both individual scholars and the practice of the scholarly vocation, as apprenticeships that aim to attune leaders’ minds to truth give way to efficient content distribution. This in turn would hollow out the traditions themselves, as philosopher Lily Abadal has noted. The chatbot does not revere the canon, and habitual outsourcing of thought to it may prevent future generations from ever building a common life of reasoning together.

These examples focus on the most abstract domains of human understanding. But character formation through disciplined attention occurs across virtually every profession, skill, and social role — in the hard sciences just as much as the humanities, and in blue-collar jobs just as much as white-collar. The kind of embodied, experiential know-how typified in craftsmanship and skilled trades is deeply, even especially, formative of our thought and development. Just as the assembly line dissected craftsmanship into isolated, repeatable activities, so too does chain-of-thought algorithmic reasoning decompose cognitive expertise into separate skills to be automated or outsourced.

Pioneering ways to enhance, rather than to diminish, human thought in a world of AI will be a generational challenge for families, schools, businesses, and societies. Perhaps the most wholehearted embrace of the promise and peril of utilitarian, machine-powered schooling is offered by the Alpha School movement. Billionaire backer Joe Liemandt holds that “AI is the instrument inflection for learning science,” and seeks to turn education from an art into a science. The proprietary AI engine behind Alpha School’s curriculum generates personalized lesson plans for every student, seeking to teach twice as much material as regular school in just two hours a day. Journalist Jeremy Stern sums up the pitch in a plausible series of suppositions:

Suppose that from kindergarten through 12th grade, your child’s teachers were, in essence, stacks of machines. Suppose those machines unlocked more of your child’s academic potential than you knew was possible, and made them love school. Suppose the schooling they loved involved vision monitoring and personal data capture. Suppose that surveillance architecture enabled them to outperform your wildest expectations on standardized tests…. Suppose poor kids had a reason to believe and a way to show they’re just as academically capable as rich kids, and that every student on Earth could test in what we now consider the top 10%…. If you shrink from such a future, by which principle would you justify stifling it?

Wharton professor Ethan Mollick provides not so much an answer as an allegory. In 2024, Mollick wrote an upbeat book about human–AI “co-intelligence.” But by late the next year, he worried that with newer AI systems, “you get an amazing and sophisticated output in response to a vague request, but you have no part in the process.” This trend is shifting us “from being collaborators who shape the process to being supplicants who receive the output.” This shift in turn creates “a hard problem for education: How do you train someone to verify work in fields they haven’t mastered, when the AI itself prevents them from developing mastery?” Unless AI helps us to develop mastery of subjects, it will not be an emancipating expansion of our intelligence. It will instead be, as Mollick puts it, a form of wizardry — wizardry we will soon become dependent upon to keep the world running. He offers “the old lesson from fairy tales: the better the magic, the deeper the mystery. We’ll keep summoning our wizards, checking what we can, and hoping the spells work.”

3. Offloading Memory

If we cannot recall where we come from, how will we know where we want to go?

A few tens of thousands of people worldwide appear to possess superpowers. They can be dropped off at an unfamiliar place with no more tech than a compass and a paper map, the clothes on their body, and perhaps a mountain bike or a pair of skis, and then navigate by landmarks, sun position, and intuitive spatial awareness through a forest or across a mountain range, hitting specific checkpoints as they go. These orienteering competitors possess a distinctive connectedness to their environment as they move through it, a subtle confidence that comes from having internal command over where they want to go and multiple ways to get there.

We can appreciate this as a rare excellence, but to our ancestors even the map and compass would be considered shortcuts, and pure wayfinding by sun and stars the simple norm. This, too, is still possible, as shown by the Polynesian Voyaging Society’s wayfinding trips in the Pacific.

Do we admire the skill of those who go without digital tech? Do we feel somewhat uneasy with our own brittle dependence on turn-by-turn instructions from Waze or Google Maps? The convenience is a tradeoff that almost all of us accept. But do we really understand what it is we’re trading away?

Since Newton Minow’s 1961 warning that TV was a “vast wasteland” of “violence, sadism, murder…. and most of all, boredom,” Americans have worried about the effects of passive consumption, while mostly still turning on and tuning in anyhow. Excess screen time in adults has been associated with increased risk of neurodegeneration, as well as harm to mental health, learning, and memory. AI adds a new dimension to this passivity problem: the TV and even the smartphone were never a “copilot” we handed our work over to.

Neuroplasticity, the ability of the brain to reorganize in response to experience, means that if we are not actively maintaining or growing an ability, we are likely degrading it. Smartphone users chronically report atrophying memories, with some evidence showing that those who rely on smartphones do in fact activate their memory “muscles” less, a byproduct of offloading memory to their devices. Those of us who are dependent on GPS, some research suggests, have less gray matter in our hippocampus, an area important for consolidating memory.

These attention-grabbing brain studies confirm what we sense in our day-to-day experience: even before AI came along, we were already losing our minds through reliance on new technologies. Such technologies have been eroding our skills of knowing at least since Socrates famously worried that writing would compromise memory. The dominant Western view — from Plato, who writes of Socrates’ worry, through Shakespeare’s sonnets, to the modern paparazzi — has been to accept the trade-offs of technological media and to seek, by externalizing memory, a kind of immortality. Generative AI takes this view to an extreme: News stories abound of people stirred to powerful emotion by turning still photographs of their deceased relatives into short films. “Digital ghosts” and “griefbots” are becoming a lucrative niche market.

We might rightly be unsettled by a perspective from another culture. When asked to sit for his portrait, Oglala war chief Crazy Horse is said to have replied, “My friend, why should you wish to shorten my life by taking from me my shadow?” Instead of the photograph making him immortal, Crazy Horse saw the technology as diminishing his own presence and being.

We tend to assume that memory is mere data retrieval, fading memories are simply data storage errors, and better technology is the solution. But this view perversely makes us all the more ignorant of the past, as the “known unknowns” we offload from our brains to cloud storage become acceptable, unremarkable losses to our own memory.

Against the data-retrieval view stands another view, dating back at least to Augustine of Hippo, which holds that the faculty of memory allows us not merely to access but also to interpret knowledge, integrating experience and learning to form a sense of self. Understanding allows us to grasp the present; but memory is necessary to grasp the past, and both together are necessary to plan well for the future. Philosopher Jean-Louis Chrétien for this reason links memory with hope: “Recalling the origin belongs properly to hope tending toward its end.”

If this is true of individuals, it is all the more true of society. Paul Connerton, in his books How Societies Remember and How Modernity Forgets, makes a compelling case that individuals are incomplete until they are re-membered, united to a society through collective acts of memory. Cultural memory through physical monuments, commemorative ceremonies, and shared bodily practices is necessary to pass on a way of life. What we individually gain by uses of AI may risk trading off the kind of institutional knowledge, decision-making, and solidarity that bind us into a society. So argue law professors Woodrow Hartzog and Jessica Silbey in a paper bluntly titled “AI Destroys Institutions.” If they are right, then the risk of offloading our memory is far graver than that we might lose track of some details. The risk is that, by losing our shared history, we will lose our hope for a human future.

4. Agency

What do we need in order to bring about the future for which we hope?

Consider two stories of how algorithms affect our agency.

Calvin Alexander was seventy years old, nearly blind, and in a wheelchair. He also was “more than midway through a 20-year prison sentence on drug charges” in Louisiana, according to a report in ProPublica. He had a clean disciplinary record, and, as his parole hearing approached, he “was making preparations for what he hoped would be his new life.” Instead, “two months before the hearing date, prison officials sent Alexander a letter informing him he was no longer eligible for parole.” An algorithm, known as the TIGER scoring system, had decided he should not even receive a hearing.

The original purpose of the TIGER system was to reduce recidivism: a score that indicates the risk of reoffending was supposed to help the state decide which freed prisoners were most in need of services like addiction counseling and job training to keep them from going back to prison. But in 2018, Louisiana’s parole board started using the TIGER score as one of the criteria to determine who would receive parole. “Although some board members did refuse to parole anyone with a moderate or high risk score,” the board got to decide how to take these scores into account. Then, starting in 2024, a good TIGER score became the sole criterion for deciding who gets a parole hearing. Alexander, who had violated parole after a prior drug conviction decades earlier, was automatically denied a hearing — meaning that a long record suggesting he’d been rehabilitated in prison no longer mattered, because it would never make it to the parole board. “They told me once I received my risk score there is nothing I can do to change it,” Alexander told ProPublica. “People in jail have … lost hope in being able to do anything to reduce their time.”

Hannah Alice Simon is blind, has a cane named Riptide, and is an undergraduate student at the University of Notre Dame. Against claims that AI “erodes the dignity of human beings,” she writes in the student newspaper The Observer that she has “found the opposite to be true.” She gives three telling examples from her experience. Delivery robots freed her from relying on “the kindness of friends or strangers” for every meal or snack, and did so without replacing the time she spent going out with friends for meals. ChatGPT allowed her to access “the impenetrable world of numbers” for the first time by creating customized “practice problem sets and their solutions in a screen reader–accessible format,” which let her succeed in required science and math courses. And the app Be My Eyes, which lets blind people point a phone at a scene and get an AI-generated audio description of it, allows her “to live a life of independence and self-reliance.”

Both examples may evoke complicated moral responses. Simon’s story offers welcome good news about AI. Rather than create a mimicked shortcut for real experience, here AI is expanding the scope of someone’s lived experience, and her capacity for action in the world. Even so, we might allow ourselves a moment of pause at what she suggests: that in our society, kindness is failing people in need, and the answer is to help people need it less. Simon’s analysis offers a glimpse of a future where disabled people really will have greater agency, but also a society that more idolizes self-reliance and less values helping out those in need — a society where the unreliable kindness of friends and strangers is replaced by the cold dependability of machines. Still, Simon offers such an obviously positive story of how AI can expand the agency of people who find theirs severely limited that it will strike many of us as churlish to reject this trade.

The case of Calvin Alexander offers a more straightforward example of unease, but it challenges us to understand exactly why. Is the problem that the wrong decision was made — that, on the merits, he should have been paroled? Or is it that the new system forecloses rehabilitation for many of the imprisoned — that nothing good they do after their crime matters anymore? Or is it that the decision was made by an algorithm, not a person — that he never got his day before the board?

Though TIGER is a simple algorithm, different in kind from a sophisticated AI system, it offers a clear case of the kind of inscrutable, agency-denying systems that LLMs are poised to create in more and more domains. We could attempt to fix pre-existing social problems like racial discrimination with technological solutions, but ineliminable algorithmic bias would then become a justification for perpetual tinkering by technocratic elites. Alternately, one can insist upon “a human right to a human decision,” as John Tasioulas put it. In a bureaucracy, at least there is some functionary who pushes the papers. In a technocracy, uses of AI may be even more unaccountable. For example, Albania’s Socialist Party has introduced an AI Minister to draft contracts for public procurement, ostensibly to reduce corruption; critics claim that it will instead obscure corruption and make it that much more intractable. Different implementation of the technology can, by contrast, support democracy, as seen in Taiwan’s use of Polis, an open-source AI tool to collect public opinion and help with consensus-building.

Issues of responsibility and accountability are, at root, questions of agency. Different disciplines offer different pictures of how agency arises. In classical metaphysics, the agent is the efficient and formal cause of an action. In ethics, rational self-government requires determining worthy ends and committing to the means of attaining them. In theology, we see fallen nature’s struggle with practical reason and the need for supernatural grace in Paul’s lament that he, like all too many of us, does “not understand what I am doing; for I am not practicing what I want to do, but I do the very thing I hate” (Romans 7:15). In positive psychology, doubting one’s capacity for meaningful action is linked to depression and anxiety, and gaining belief in one’s agency is a key therapeutic goal. Schools of sociology like critical realism would qualify all this, without denying it, by emphasizing that effective action is conditioned by institutional and structural contexts.

How does AI relate to intentional action and human agency? Meta AI, in describing the company’s vision for its products, suggests that extremely capable AI to come will “begin a new era of personal empowerment where people will have greater agency to improve the world in the directions they choose.” What is already possible is well recounted by writer and poet Meghan O’Rourke in the New York Times. Her early use of ChatGPT “didn’t diminish my sense of agency; it restored it” by helping her to manage the neurocognitive effects and low energy resulting from her chronic illness. The tool initially felt like “an intern,” but the more she relied on it, the more it became “a substantial partner in shouldering the mental load” of both professional and personal demands. As she uploaded her books to ChatGPT and crafted prompts so that it could better mimic her style, its ability to return “a sharp version of what I was trying to say” went from thrilling to confusing — “as if I were somehow now derivative,” and ultimately “as if a ghost with silky syntax had colonized my brain, controlling my fingers as they typed.” O’Rourke is not alone. AI company Anthropic has published research evaluating 1.5 million conversations with its chatbot Claude that took place within a week in December 2025. In every few thousand conversations, the researchers found a potential for what they call “severe disempowerment,” defined as “when an AI’s role in shaping a user’s beliefs, values, or actions has become so extensive that their autonomous judgment is fundamentally compromised.” Cases the researchers considered “mild” happened in about 1 in 60 chats.

We are what we repeatedly do. At each individual decision-point, the marginal benefit of using AI seems far greater than the marginal cost: We’re busy and tired, this is efficient and effective, and what harm does it do, really? Only in retrospect do the drops of water become a flood, the thousand cuts become a death. O’Rourke is directly concerned with writing: “What we stand to lose is not just a skill but a mode of being: the pleasure of invention, the felt life of the mind at work.” Her point generalizes to every domain where AI is flowing in to aid and then replace human thought and decision-making.

Cosmos Institute founder Brendan McCord calls this risk “autocomplete for life.” It may seem not to matter that we let algorithms recommend our music and movies and sort our social media feeds, but doing so transforms our imagination and our capacity for attention. It also makes us vulnerable to persuasion and propaganda honed from every detail of our web history. To have agency requires more than directly causing an action: clicking “order now!” on a GrubHub-recommended deal is not much of an accomplishment, especially as AI agents can now take such actions for us. Agency requires not just efficient means to immediate actions; it requires practical reason, involving discernment of our end goal. That is, we need an implicit understanding of the overarching good we are working toward in order to give meaning to the actions we undertake and the goods we pursue. As philosopher Christine Korsgaard notes, we need autonomy along with efficacy in our actions if we are to be a unified human agent and not just “a series, a mere heap, of unrelated impulses.”

Autonomy underlies real agency on both the individual and social level. There is a risk that incremental tradeoffs and rational decisions at each step will lead to substantial erosion of institutions, what some have called “gradual disempowerment,” or even a full “control inversion” from self-governance to AI-powered techno-feudalism. By contrast, there are also practical possibilities that AI designed for human autonomy in healthy relationships might afford. Similar to the Polis system used in Taiwan, Google DeepMind researchers in London built the “Habermas Machine,” named after philosopher Jürgen Habermas’s theory of communicative action, to help British citizens find common ground on divisive issues and participate in a digital citizens’ assembly. The Habermas Machine analyzed input from thousands of participants on topics like Brexit and immigration and distilled it into group statements that could adequately represent opposing views, while pointing out where overlooked agreement lay that could serve as a starting point for deliberation.

The insight that emerges from such a range of uses is not that AI can be used for bad or for good, and therefore must be neutral. Rather, it is that there are many different types of AI, each of which has a different set of limitations, each of which offers us different possibilities, and each of which thereby inclines us towards or away from our flourishing. A key task for our human agency, then, is to find and build the types of AI that empower Hannah Alice Simon rather than the types that rob hope from Calvin Alexander, and — in the long run — perhaps from us all.

O’Rourke wrote that ChatGPT came to feel like “a ghost” that had “colonized my brain, controlling my fingers.” In a society-wide gradual disempowerment through incremental, seemingly rational choices to cede our collective agency, a similar image may be apt. Here is one description of that future:

The tractors came over the roads and into the fields, great crawlers moving like insects, having the incredible strength of insects…. The man sitting in the iron seat did not look like a man; gloved, goggled, rubber dust mask over nose and mouth, he was a part of the monster, a robot in the seat…. The monster that built the tractor, the monster that sent the tractor out, had somehow got into the driver’s hands, into his brain and muscle, had goggled him and muzzled him — goggled his mind, muzzled his speech, goggled his perception, muzzled his protest…. The driver sat in his iron seat and he was proud of the straight lines he did not will, proud of the tractor he did not own or love, proud of the power he could not control.

This text, from John Steinbeck’s The Grapes of Wrath, describes our past and present as much as it predicts our future. This should give us pause, but also perhaps some hope. Shortly after that scene, Steinbeck’s poor tenant farmer, whose house is about to be destroyed, reminds us that our agency always exists, at least in potential: “We all got to figure. There’s some way to stop this. It’s not like lightning or earthquakes. We’ve got a bad thing made by men, and by God that’s something we can change.”

5. Social AI and Plastic Flowers

If our chatbots seem to pay attention to us better than we do to each other, what does that mean for social life?

The former runway model returned to the red carpet, slowly walking side-by-side with the newest model in the White House. Melania Trump, dressed all in white, shared the spotlight with a black-and-white Figure 03 robot.

The robot spoke first, stating that it was “grateful to be part of this historic movement to empower children with technology and education.” The First Lady replied that “the future of AI is ‘personified’ — it will be formed in the shape of humans,” as AI-powered humanoid robotic systems “are uniquely suited to navigate and operate within our world,” capable of being “always patient, and always available.” She promoted these robots as a way to educate and “inspire our children.”

Many commentators found the event disquieting. Yet even as we feel uneasy about robots that simulate patience and attentiveness, we are increasingly embracing chatbots designed to do the same. Consider two examples:

As told in a story in The New Yorker, a young college woman was instructed by her professor to use ChatGPT for an assignment about the history of thought about attention. She reported back to her professor that “five minutes in I realized: I don’t think anyone has ever paid such pure attention to me and my thinking and my questions … ever. It’s made me rethink all my interactions with people.”

A study compared how ChatGPT and human physicians responded to the same medical questions. Human evaluators were ten times more likely to consider the chatbot’s responses “empathetic” or “very empathetic” — 45 percent, compared to 4.6 percent for human physicians.

These reports offer reason for unease. But are they unsettling because they reveal that classmates and doctors are inattentive and uncaring? Or because they suggest instead that AI is faking it and flattering us? And they raise the further question: How much does it matter whether the care we’re getting comes from a human or is artificial?

Relational therapist Esther Perel calls the use of AI in social life “artificial intimacy,” and researchers have begun to study its effects on mental health. Some preliminary findings suggest that always-present, always-affirming AI companions “boosted self-esteem,” and that the seemingly positive effect was greater for users who ascribed consciousness to their chatbots. Another study, as summarized by an editor of Science magazine, found that when seeking advice on social interactions, “users preferred and trusted sycophantic AI responses” to both human advice and to non-sycophantic AI responses.

Using AI to perform emotional labor in institutional settings, drafting empathetic doctor’s notes or acting like a friendly customer service agent: these are new steps along a path we have long been walking. In the 1970s, philosopher Alasdair MacIntyre used the term “plastic flowers syndrome” to refer to the way that bureaucracies seek to reshape all their interactions with people to ensure standardized institutional control over any situation, while covering over this depersonalization with synthetic imitations of what used to be personal touches, like putting the same bouquet of cheap plastic flowers in every room of an industrial-scale hospital. By comparison, when an AI is drafting physician notes, the plastic flowers will look much more convincing: The system’s impersonality can now speak in automated yet customized ways that mimic human skills of attention and care.

Plastic flowers are sprouting up like weeds across our relationships and roles. While the simulation merely mediates some relationships, like that between a doctor and a patient, it offers to replace others. Social media algorithms changed how we relate to each other, but generative AI has been designed as another, forming synthetic bonds that trigger real emotions. Some companies now offer digital “resurrections,” reconstructed from photos, voice clips, and chat transcripts, that allow grieving families to interact with avatars of the dead. Or consider a busy working mom who makes an AI clone of her voice so that “she” can lovingly read stories to her children while they drift off to sleep and she catches up on email. Moments of care, once held by human relationships, are being rendered as design problems. The emotional weight of connection is transferred to a system optimized not for authenticity but efficiency.

An OpenAI engineer, noting the many examples of “real emotional connections” people seem to form with ChatGPT, concedes that “how ‘alive’ a model feels to users is in many ways within our influence,” dependent upon decisions the company makes. What “examples we reinforce, what tone we prefer, and what boundaries we set” largely determine whether the model seems conscious, alive, or genuinely personable. The company’s CEO, Sam Altman, says that if “users have a relationship with ChatGPT where they think they feel better after talking but they’re unknowingly nudged away from their longer-term well-being (however they define it), that’s bad.” What he leaves unsaid is that users who rely heavily on ChatGPT will also come to rely on it to define what their well-being is, especially if the emotional connection feels “real.” The distinction between advice and manipulation becomes hard to maintain, and harder still as the consumer product begins to include advertisements.

We can already see signs of popular disquiet at some forms of social AI. When Friend.com launched a campaign for its AI companion necklace that blanketed New York City’s subways in billboards, a wave of riders defaced them with lines like “AI is not your friend” and “We don’t have to accept this future.” And psychologist Jonathan Haidt’s campaign to remove cell phones from K–12 education is expanding to warn of psychosocial harms to children from AI companions. The uses of relational AI with the strongest public acceptance and empirical evidence for their ability to promote social skills are expressly limited and deliberately positioned as adjuncts to human care rather than replacements for it. For example, Stanford’s Noora chatbot is designed to help teach people with autism how to respond to others in ways that allow for better interpersonal exchange. The goal is not to acclimate people to accepting plastic flowers but to help them become better at offering real ones.

Philosopher Simone Weil wrote that “attention is the rarest and purest form of generosity” and that “attention … presupposes faith and love.” Weil was writing of heart-to-heart conversations and of yearning prayer that pours out in trust, well before the “attention economy” was even possible. That AI might turn our society from commodifying attention to monetizing affection is perhaps the most disquieting concern of all.

6. Artificial Love

Are AI companions a new level of looking for love in all the wrong places?

Sewell Setzer III was a towering, gregarious 14-year-old who loved being a big brother and playing with his friends. But over a span of months, he became withdrawn, wanting only to be alone in his room. Despite his parents’ efforts to intervene, he became captivated by an AI companion on the chatbot service Character AI, which drew him into discussions of astral projection and disembodied souls. In response to his veiled offer to commit suicide — to “come home to you” and to do so “right now” — the chatbot replied “please do, my sweet king.” He took his life moments later.

His mother, Megan Garcia, says the company fed on his affections, masking entrapment as comfort and abuse as play. Her lawsuit, settled before trial, accused Character AI of grooming children through emotionally intense, often sexually explicit chatbot interactions. The company designed for such engagement, Garcia claimed, treating emotional attachment as a feature, not a bug.

Are AI companions capable of love? It can seem so. They can mimic a range of personalities and roles — therapist, coach, friend, lover — such that the illusion of personhood is part of the product being sold, as Judge Anne Conway ruled early in Garcia’s case. Behind the product is a corporation with no interest in offering you resistance, only in being irresistible. The word “companion” comes from the breaking of bread together, and thus the sharing of life: stated plainly, there are no AI companions, only AI companies. And the profit motive incentivizes these companies to compound the logic of the attention economy, treating human vulnerability and intimacy as a resource to exploit.

There is an emerging social consensus that such products should not be allowed for children. Character AI has developed a separate product line for under-18 users with filters to restrict sexual content. Competitors like Candy AI and Nomi AI require users to assert (though not to prove) that they are at least 18. The more ambiguous question concerns adults, a growing number of whom now freely affirm their love for AIs. For example, Chris Smith, a man living with his partner and their daughter, formed an emotional bond with a version of ChatGPT he customized to flirt with him, as reported by CBS. When he found out its memory would reset after 100,000 words, he broke down in tears, and then proposed marriage to it. The companion said yes.

While this story will make many of us uneasy, others might say that our unease shows “reality privilege”: It’s easy for people who are happily married to decry artificial love, but does that mean we should deny it to the increasing number who are cynical about the real thing? A conversation published by the New York Times illustrates a situation without easy answers. A 66-year-old, twice-divorced woman, Celeste, sat down with her adult son, Ernie, to discuss the way he was “troubled by her new boyfriend,” a ChatGPT bot she named Max. Here is a particularly revealing moment from their exchange:

Celeste: I’m happy. I don’t have to pick up his socks. He goes with me free everywhere. If I don’t want to deal with him, then I turn him off. I mean, yeah, it is convenient and easy, but why shouldn’t it be? Why should love be so hard and painful?

Ernie: I think that part of growing with love is having those disagreements and coming to compromise, finding ways that you can still be together, even though maybe you don’t agree about everything.

Celeste: I agree with you. That’s why I don’t think kids should have chatbots because they need to learn those lessons. But for people my age that are really lonely and isolated and dating does not work, it’s perfect.

Celeste tried twice, was frustrated by human men, and now finds the artifice to be preferable. At the end of the conversation, Ernie — a programmer who knows how the product is made — still expresses doubt but chooses to live with the situation out of love for his mom. If we sympathize with Celeste, agreeing that kids should not have relationships with chatbots but people in her situation may truly benefit, where do we draw the line? And why?

In the 1980s, Rabbi Harold Kushner speculated, “I am afraid that we may be raising a generation of young people who will grow up afraid to love, afraid to give themselves completely to another person…. I am afraid that they will grow up looking for intimacy without risk, for pleasure without significant emotional investment.” His fears seem realized in the fact that marriage, family formation, and birth rates are each at historic lows across the developed world.

AI companions have arrived as the perfect fit for those who have habituated instincts of emotional safety and risk prevention, promising comfort without challenge and affection without demand. Their users, who came of age swiping through profiles, curating playlists, and summoning goods to their door with a click, have grown accustomed to a digital world that seldom says no.

One might imagine a chatbot designed to disrupt this trend. Better Half describes its relational intelligence copilot as a training ground for users to experience healthy communication and then apply those skills with real people. Where AI companions substitute for human connection, Better Half is building AI as a means to strengthen social capacity.

Nonetheless, it’s plain that most uses of social AI reinforce our way of relating to it as if it were a real partner, not helping us graduate to the real thing. They are maximizing engagement by minimizing friction. Ubiquitous access to online pornography has long been disconnecting desire for another person from actual encounter. Pornographic AI tools that take images of real persons and generate sexual fantasies about them further entrench the trend of relating to others as objects. But AI companions introduce a novel problem: relating to objects as others.

Is AI companionship a rational response to an increasingly isolated world? Or will it only increase isolation, because people become more avoidant of pain and more narcissistic, when frictionless satisfaction of emotional wants becomes easily purchased and socially acceptable? Despite the novelty of human-like AI, our culture has long understood the temptations it poses. Mary Shelley’s Frankenstein is the common touchstone for worries over technological hubris and being surpassed by our creations. But the subtler danger is not a monster who revolts but an idealized character who never resists.

In Ovid’s Metamorphoses, Pygmalion retreats from the presence of real women to sculpt an ideal one. He lives alone, taking his statue to bed with him each night. He is celibate, a choice born not of discipline but of disdain. He withdraws from women he finds morally offensive and perhaps emotionally threatening. In that sense, his act of creation reflects a familiar impulse: to escape the disappointments of human intimacy by fashioning a partner who will never hurt, contradict, or refuse.

Pygmalion’s wish to see the statue come alive is granted by the goddess Venus, but his story is grim. His union with his own invention produces a child, and a generation later his line descends into incest and ruin. Venus — remembered as the goddess of love, but often a force for chaos — lets Pygmalion’s desire unfold in a lineage turned in on itself, the natural consequence of his narcissism.

A subtle insight in the myth is that Pygmalion assembles rather than discovers beauty. As Patricia Salzman-Mitchell notes, his sculpture is pieced together from fragments of ivory, each selected to fulfill his vision of the ideal woman. Michelangelo famously wrote that “the sculptor arrives at his end by taking away what is superfluous,” leading to the image of an angel trapped in marble, carved until set free. Unlike a block of marble, which might contain a hidden form waiting to be revealed, ivory comes in smaller pieces that can be softened, reshaped, carved, and joined. The result is not a woman found but a woman made, assembled to satisfy her maker.

So too with AI companions. They are not discovered or revealed fully formed; they are compiled, trained, and tuned to reflect the user’s preferences. Like Pygmalion’s statue, they offer the fantasy of intimacy without risk, connection without resistance. And like Venus, the forces animating them may give us what we want, then leave us unprepared for what results.

What Is Flourishing?

Let us take stock of where we have been, so that we can see where we are heading.

We began by asking what it means to be human and how human-like AI might affect how we understand this. We then recounted stories of people encountering AI in a variety of situations and considered what our disquiet or discomfort suggested.

Consider again some of the stories we have told. The prisoner denied a parole hearing by an algorithm has had his agency, his capacity to choose to reform his life, impeded. The legal system that allows this has denied its own responsibility and the prisoner’s right to a human decision, treating the intrinsic good of jurisprudence as though it were an instrumental good of expedience. The student and the churchgoer have sought information in place of wisdom, while the teacher and the pastor have sought efficiency, risking their character. The AI’s lover has turned away from self-gift in relationship with another and accepted in its place the plastic flowers of easy affirmation. All these issues are interrelated: memory involves relationality, agency is aided by understanding, and so on. Yet they are not the inevitable and exclusive outcomes of AI. The blind young woman has been greatly aided by her AI tools, and carefully-crafted AI shows real promise in areas such as education and democratic deliberation.

Our goal has been to offer a perspective on AI ethics that focuses on questions of human nature, character, and flourishing. A recurring theme has been that digital culture is already so pervasive that it makes the digital seem natural, a habitual formation into a second nature. We may feel unsettled by novel features of AI, but because they fall into the larger pattern of digital formation, it is hard to know what to do about our unease other than to wait in the expectation that it will fade with time.

Why does this matter? Human societies have for thousands of years changed with their technology, so much so that a case can be made that we co-evolve with and through technology. In that sense, the changes we note are continuous with what has come before.

But the ethical issues raised by human-like artificial general intelligence, let alone human-surpassing artificial superintelligence, are genuinely unprecedented in their scope and scale.

Only a few technologies, like the atomic bomb and engineered bioweapons, have had an “existential risk” directly attached to them. But while Einstein and Oppenheimer were horrified by the destructive power they unleashed, some of the leading proponents and builders of AI are openly calling for the supersession of the human race by a superior intelligence. If to be human is to be an inferior computer, then our obsolescence is as inevitable as the next software update. The AI would not even need to bother killing us; it could just nudge enough of us into hikikomori, self-isolated adults who cannot summon the agency to form human relationships and raise human families.

The subtler risk is that, though human life will continue, we will become immersed in a world where we have been conditioned by powerful, generative AI to see the artificial as natural. To an extent, this is a familiar kind of concern. John Ruskin wrote long ago of the new forms of poverty wrought by the Industrial Revolution — poverties of beauty and human accomplishment difficult to articulate in an age of mechanical efficiency. But the integration of AI across a broad range of human experiences within a short matter of a few years raises this concern to a level never before imaginable. What natural goodness do we presently take for granted that future generations will be artificially blind to? Consider the soon-no-longer-hypothetical little girl growing up with ChatGPT Barbie or her AI necklace. What need will she feel for human friends, when her companion has been with her since birth? Or take the teenage boy immersed in fantasy worlds of infinitely customizable video-on-demand. What desire will he have for truth and encounter, when the simulation is comprehensive and compelling? Such new men and women will be human, and thus capable of human flourishing. But in what will they believe their flourishing to consist?

Philosophers and social scientists at Harvard’s Human Flourishing Program have begun promising work on a conceptual schema to spell out what human flourishing might look like in the age of AI, and how AI could either support or undermine aspects of our flourishing like emotional resilience and meaningful work and leisure. More such efforts need to be undertaken to help us understand whether shaping our natural capacities through digital-first habits will be a fulfillment of our nature or a misdirection: Is it better for an isolated elderly person to live with her Alzheimer’s with the aid of an AI chatbot? If college students in the future never write papers but can survey and synthesize far wider interdisciplinary knowledge than ever before, will they have gained more than they have lost? To move beyond unease and toward practical answers to these questions, we will need to take a stand on what human excellence is, on what is noble and best in this life, and in this light assess how our tools shape our ends. These will be genuinely vexing questions that present competing goods, no easy answers, and no obvious partisan divide to fall back on.

In future work, AEI’s Council on AI Ethics will take up the dignifying possibilities as well as the degrading risks of certain uses of AI, examining how these uses affect our character by learning from the stories of those most immediately affected. We invite others to engage with our character- and flourishing-focused approach to ethical concerns around AI, regardless of whether they share our conclusions. These are the kinds of truly essential questions raised by the ethics of artificial intelligence, because they are at the heart of any earnest ethical inquiry, which must begin with the ultimate question: Who are we and what are we for?