I can see no reason in principle why interaction with AI could not offer practice that helps form good habits. But there is no guarantee that it will. A chatbot designed to gratify our egos will not help form the moral habits that teach us to treat others as we should. We know the type: billionaires, movie stars, autocrats, all surrounded by obsequious underlings, often acquire the vices that flourish under flattery, namely narcissism, vanity and cocksureness. There is already research literature on “sycophantic AI”, including the claim that many AIs “affirm users’ actions 50% more than humans do.”[12] If we leave things as they are, prolonged exposure to artificial sycophancy, like prolonged exposure to human sycophancy, could make us less generous and compassionate in our dealings with others, and less open to the possibility that we are wrong.
“Synthetic intimacy” and the value of friendship
But the question of how to live well does not end with character. Aristotle and Confucius also recognized that the best kind of life depends on the right kinds of relationships. Friendship is central to Aristotle’s conception of the good life; family, not least the relations between husbands and wives, siblings and generations, is central to Confucius’s. Both thought our political relationships mattered, too. The many forms of love—eros, family, friendship, neighborliness, political fraternity—are integral to human well-being. A life without them would be a diminished one. What matters in all these relationships is that there is another real, independent person on the other side, someone whose reality you must acknowledge. As the novelist Iris Murdoch put it, love is “the extremely difficult realization that something other than oneself is real.”[13] A life made up largely of “conversations” with AI would not cultivate the real relationships that human flourishing requires.
Some people, as we know, have already developed deep feelings for the beings they imagine on the other side of their screens. Perhaps these pseudo-relationships are better than none at all; perhaps they are better than some toxic relationships with actual people. But “synthetic intimacy” is less valuable than the ordinary loves and friendships we have with real ones. If someone told you that their most important relationship was with a pet, you would be entitled to conclude that they had invested those interactions with more significance than they could bear, and that they were missing something important about a human life.
The same, I think, is true of a putative relationship with an LLM. In some ways it may be worse than emotional overinvestment in a dog. With pets, after all, there really is a creature there. Fido and Fifi may not have all the feelings you project onto them, but they do have feelings. They are real. And because they are real, they may resist your wishes, a mark of independent existence. A dog does not merely return your affection; it also interrupts your plans, ignores your preferences and occasionally soils the carpet.
AI presents possibilities here as well as perils, to be sure. For some people, the alternative to a chatbot is not a rich and varied social world but isolation and loneliness. In those cases, attachment to an LLM may be less about delusion than about deprivation. The spread of such attachments would reflect a deeper social failure, one that has left many people without the forms of community and care on which we ordinarily depend. However soothing such systems may be, of course, they still do not provide us with what a human does: another center of experience who can make claims on us and resist us, as a being with an independent life. So while synthetic companionship may be a helpful accommodation to certain forms of isolation, we shouldn’t be eager to normalize substitutes where solidarity, friendship and care should have been possible.
The same temptation appears when we begin to treat our exchanges with LLMs as conversations in the full human sense. A well-made machine may sometimes help us prepare for human interactions, clarifying what we mean or anticipating misunderstanding. People rehearse conversations in their heads all the time, and AI may serve as a more elaborate aid to that familiar exercise. But such preparation becomes ethically suspect when the aim is to manage other people rather than engage with them.
Even when exchanges with a chatbot improve us, calling them “conversations” is misleading. Real conversations are part of our evolving relationships with other people. They are shaped by the concerns of all the participants, not simply by one person’s wish for an answer. Ideally, they are shared explorations of questions that matter to everyone involved, and part of their value is that, through them, we come to know one another better. An exchange with an LLM is nothing like this. An LLM has no concerns or interests. It may be interesting to know about, but it is not worth getting to know. At most, it offers an ersatz interaction with an imagined being. To return to Murdoch’s insight, in engaging with it you are not in contact with the reality of another person.
Compare these exchanges with our engagement with fiction. There, as Coleridge said, we practice a “willing suspension of disbelief.”[14] We respond, for the moment, as if Ophelia has drowned, or Dorothea Brooke has married the man she loves, or Charles Foster Kane was traumatized by being sent away by his mother. But our response is qualified by our knowledge that none of it is real. We do not really believe these events occurred. Ask us, and we will say so. People who truly believe they are in a relationship with an LLM are doing something else. They are not engaging with a fiction. They are suffering from a delusion.
The LLM as moral guide
What aspects of our humanity, if any, could be improved in a world where AI use becomes pervasive? The answer, I am arguing, depends both on how we build it and on how we build it into our lives. Imagine an LLM that behaves as the best moral guides do. You give it your circumstances and your options and ask what you should do. It does not simply choose for you. It helps you see which considerations are relevant to choosing for yourself. It walks you through the decision rather than making it on your behalf. It reminds you of options you had not considered and alerts you to facts you did not know. Perhaps you ask it to draw on your own religious tradition, or on some specific moral approach. Used in that way, AI could enhance our humanity, deepening rather than diminishing our autonomy.
The positive case is clearest on the intellectual side. AI can help you test and refine your existing views, learn about the views of others, consider objections, sharpen your wits and cultivate a thoughtful openness to opposing arguments. Part of living well lies in developing an understanding of the world that answers to how things actually are. AI, properly built and wisely used, can serve that end. An exchange with an LLM about a question that interests you may help you pursue it, bearing in mind your own knowledge and concerns, with a remarkably well-informed interlocutor, and without pressing an actual person into service for your intellectual projects.
The risks, however, can’t be waved away. Of greatest concern is what I’ve called constitutive de-skilling, the erosion of the capacities that make us human in the first place.[15] Some forms of it, such as the loss of character and sensibility that moral deference might produce, can be mitigated by better design. Others, such as reliance on pseudo-relationships, depend on whether people learn to use these systems well. That, in turn, is made harder by the fact that most users know very little about how LLMs work or how one model differs from another. So one obvious task is to improve public understanding of AI, through greater transparency and more serious study of how these models reason, advise and mislead.
Ethics, as I said, is about how to live well. But the lives we live are human lives, and what is good for us reflects what it is to be human. Our humanity is bound up with character, with deliberation and autonomy, and with relations to other people whose reality we must acknowledge. These capacities are formed over time, through engagement with the world and with one another. As with every new technology, from the printing press to the internet, the question is how AI might help us do that better.
The Wordsworth cure
Mill’s life has as much to teach here as his arguments. When he suffered that early breakdown, in 1826, it was because he lost faith in the pursuit of utilitarianism, his family creed. But he had no one he felt he could talk to about his crisis. His recovery, he said, came in part from reading Wordsworth. It came, too, from rejecting the psychological picture in which his father had raised him, a form of associationism that treated the mind as a mechanism for managing pleasure and pain, just as his father’s utilitarian ethics treated morality as a matter of maximizing the surplus of pleasure over pain. As he wrote in On Liberty, human nature is “not a machine to be built after a model,” but a living thing that must “grow and develop itself on all sides.”[16]
As Mill came into his own, in a rapidly industrializing nation, his life was formed by argument, attachment and friction with other minds. Consider his friendship with Thomas Carlyle. They agreed about much and disagreed about much, and if the friendship later cooled, that cooling itself reminds us that intellectual companionship need not rest on ease or similarity. The most valuable thing that another person gives you is sometimes not reassurance but resistance.
Mill’s life was remade more fully when, in 1830, he met Harriet Taylor, with whom he began a long and rewarding intellectual partnership, grounded in part in their shared enthusiasm for feminism. They married in 1851, two years after she was widowed. And when On Liberty was published after her death, Mill wrote in the dedication that it, “like all that I have written for many years, belongs as much to her as to me.” That sense of a life’s work shared captures how deeply human flourishing depends on the reality of others.
But Mill’s life was also a model of autonomy. He practiced a sturdy intellectual independence, following his reasons where they led him. Reared as a machine tuned for performance, he became, through the long labor of thinking, loving, arguing and learning, what his philosophy prized: a human being formed for freedom. The question he would have asked about our new machine age is the one we must ask: not just what it enables us to do but also what manner of people we become in doing it.
Notes
[1] Krzysztof Budzyń, Marcin Romańczyk, Diana Kitala, Paweł Kołodziej, Marek Bugajski, Hans O. Adami, et al., “Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study,” The Lancet: Gastroenterology and Hepatology 10, no. 10 (October 2025): 896–903. Jonathan Makar, Jonathan Abdelmalak, Danny Con, Bilal Hafeez, and Mayur Garg, “Use of artificial intelligence improves colonoscopy performance in adenoma detection: a systematic review and meta-analysis,” Gastrointestinal Endoscopy 101, no. 1 (January 2025): 68–81.e8. https://www.giejournal.org/article/S0016-5107(24)03471-0/fulltext
[2] John Stuart Mill, On Liberty (London: John W. Parker & Son, 1859), 106–7.
[3] Bertrand Russell, The Collected Papers of Bertrand Russell, Vol. 11: Last Philosophical Testament, 1947–68, ed. John Slater and Peter Köllner (London: Routledge, 1997), 497.
[4] See, e.g., David Rozado, “Measuring Political Preferences in AI Systems: An Integrative Approach,” arXiv:2503.10649, 2025. I don’t endorse all his suggestions for what to do about this. My thinking here has been shaped by W. Russell Neuman et al., “‘Amazing, They All Lean Left’—Analyzing the Political Temperaments of Current LLMs,” paper given at the N.Y.U. Sociology of Culture Workshop, Thursday, November 13, 2025.
[5] See, e.g., Amory Gethin, Clara Martinez-Toledano, and Thomas Piketty, “Brahmin Left Versus Merchant Right: Changing Political Cleavages in 21 Western Democracies, 1948–2020,” Quarterly Journal of Economics 137, no. 1 (2022): 1–48. http://piketty.pse.ens.fr/files/GMP2022QJE.pdf
[6] Thomas Kelly, Bias: A Philosophical Study (New York: Oxford University Press, 2022), 63.
[7] On climate change: in the United States, according to a poll by NORC at the University of Chicago, among Republicans, belief in human-driven climate change rose between 2017 and 2024 from 26% to 34% but fell among Democrats from 72% to 67%. https://epic.uchicago.edu/insights/2024-poll-americans-views-on-climate-change-and-policy-in-12-charts/ On vaccination: in the United States, according to Gallup polling, there has been a significant decline in support for vaccination among Republicans while Democratic support has remained fairly constant in this millennium. https://news.gallup.com/poll/648308/far-fewer-regard-childhood-vaccinations-important.aspx
[8] Jesse Graham, Jonathan Haidt, and Brian A. Nosek, “Liberals and Conservatives Rely on Different Sets of Moral Foundations,” Journal of Personality and Social Psychology 96, no. 5 (2009): 1029–1046.
[9] W. R. Neuman, C. Coleman, and M. Shah, “Analyzing the Ethical Logic of Six Large Language Models,” arXiv:2501.08951, 2025, 1.
[10] Neuman et al., “Analyzing the Ethical Logic of Six Large Language Models,” 10.
[11] John Stuart Mill, The Collected Works of John Stuart Mill, Volume XVIII: Essays on Politics and Society, Part I (Toronto: University of Toronto Press, 1977), 253.
[12] Myra Cheng, Cinoo Lee, Pranav Khadpe, Sunny Yu, Dyllan Han, and Dan Jurafsky, “Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence,” arXiv:2510.01395. https://arxiv.org/abs/2510.01395
[13] Iris Murdoch, Existentialists and Mystics (London: Chatto & Windus, 1997), 215.
[14] Samuel Taylor Coleridge, Biographia Literaria, ed. Adam Roberts (Edinburgh: Edinburgh University Press, 2014), 208.
[15] Kwame Anthony Appiah, “The Age of De-skilling,” The Atlantic, October 26, 2025. https://www.theatlantic.com/ideas/archive/2025/10/ai-deskilling-automation-technology/684669/
[16] Mill, On Liberty, 107.
