Can Artificial Intelligence Be Moral? (with Paul Bloom)
What this covers
Russ Roberts and psychologist Paul Bloom examine whether artificial intelligence can or should be moral, starting from Bloom's counterintuitive claim that moral AI is both conceptually incoherent and undesirable as a goal. The conversation moves between philosophy and psychology, touching on how humans actually reason about ethics, what we want from our tools, and the deeper tensions that AI merely makes visible. Rather than a technical dive into the Alignment Problem, this is a sustained argument about the nature of morality itself and why autonomy—including the freedom to make poor or wrong choices—matters more to us than we typically admit.
Bloom argues that universal alignment is impossible: any AI designed to reflect one group's moral worldview will necessarily misalign with another's, since people want fundamentally different things. More provocatively, he contends we would never actually accept an AI that overrode our daily choices, even when those choices are immoral—we prize autonomy over moral correctness. The discussion branches into why moral certainty is dangerous: most of the worst atrocities come from people convinced they are right, not from sadists aware of their villainy. The pair also weigh whether humanity has genuinely improved morally over time (Bloom cites Pinker, Singer, and Wright as optimists) against evidence of new forms of alienation and deaths of despair. A third thread emerges: beyond the conflict between morality and freedom sits an authoritarian impulse—the primate-level pleasure in controlling others—that often disguises itself as moral concern. Throughout, the conversation treats intelligence as a tool orthogonal to moral aims, and returns repeatedly to epistemic humility as a reason to preserve human choice even when we choose badly.
Bloom argues that 'moral AI' is both incoherent (no single human morality exists to align with) and undesirable, because we value our autonomy over having tools that override our choices; the real tension AI surfaces is the perennial conflict between morality and freedom.
- Aligning AI with 'our' morality necessarily misaligns it with other moralities, so universal alignment is impossible
- We would never tolerate an AI that overrode our daily choices even when it was morally right, because we prize autonomy
- AI revives the long-standing morality-vs-freedom debate plus an authoritarian urge to control others
Historical decline in violence is fragile, AI risks stem from design failures, and Western utopianism overlooks genuine value conflicts between groups.
- Pinker's claim that violence has declined over time is not persuasive because, following Taleb, the historical decline in deaths is fragile—one nuclear mistake would make the entire trend look catastrophically wrong.
“I think Steven Pinker is wrong about the violence part... I'm a Nassim Nicholas Taleb fan on this... one nuclear mistake will make all that look really bad.”
- A tool by definition does its owner's bidding, possibly imperfectly, but does not form its own intentions; a hammer never decides to strike a thumb. The AI debate hinges on whether the machine will fail to do our bidding or instead do a bad person's bidding, which requires designing in restraints.
“Tools don't have morality... my hammer does not decide one day to jump up and slam your thumb... By definition a tool means it does my bidding.”
- The WWI Christmas truce illustrates a real but limited human capacity to recognize shared humanity, because that war had little moral at stake; in most conflicts the parties genuinely hold different and conflicting values, so the naive belief that sitting down together reveals shared values often fails.
“World War I was a particularly foolish, tragically foolish war, where seemingly nothing moral was at stake... there's a sort of naivete about the human heart... A lot of times we don't have the same values: we have really different values.”
- The West's utopian, messianic impulse toward perfecting the human experience derives from Judeo-Christian sources, and the tech world's embrace of and unease with AI reflects this inherited belief that the world is improving toward a destination—whether religious redemption or a secular perfected being.
“we who are children of the West, who are the result of the so-called Judeo-Christian values, have a utopian impulse that comes from Jewish and Christian sources. A messianic impulse, an impulse toward the end of days.”
Everyday moral progress occurs through humane treatment across social relationships and declining violence within communities.
- There is evidence we treat each other better not just in the obvious group cases but in everyday interactions—treating children more humanely as corporal punishment declines and treating friends more thoughtfully—suggesting moral progress beyond tribal categories.
“I wonder if people just treat their children more politely, more humanely, as corporal punishment declines. Do they treat their buddies more thoughtfully? I like to think they do.”
- While material progress and cultural evolution have clearly improved standards of living, other parts of the human experience—meaning, belonging, and purpose—may not have advanced or may have worsened, evidenced by rising suicide rates among young people and weaker feelings of belonging, possibly partly a consequence of modern culture.
“It's only in my later years that I've wondered whether the other parts of the human experience have not advanced nearly as much, or in fact have become worse. I'm thinking about despair, a sense of meaning, a sense of belonging.”
Intelligence and good intentions do not guarantee ethical outcomes; moral reasoning requires humility about human capacity and motivation.
- The most vicious criminals typically do not see themselves as evil but feel justified and even proud of their acts; the assumption that they secretly know they are evil is a comforting but false perspective-taking error.
“the most vicious criminals that almost everyone would say did something horrific... I feel not just justified in what they did, but feel proud of what they did.”
- The idea that smarter people make more ethical decisions is foolish; one would not use IQ as a measure of who makes the best moral decisions, and you would not turn to a smarter person to learn the right thing a good human being should do.
“The idea that smarter people make more ethical decisions--listeners probably remember, I'm not a big fan of that argument... would you use IQ as your measure of who would make the best moral decision?”
Generational attitudes may not converge because marriage and children, which fewer young people experience, are what fundamentally reshape values.
- The assumption that young people will become more like older people as they age may fail because much of what makes older people who they are stems from the experience of marriage and children, which broadens and changes preferences; with far fewer under-thirties marrying or having children, generational attitudes may not converge.
“Not if they don't marry and not if they don't have children. A lot of the reason that old people are the way they are... is because they've gone through... the experience of marriage and children, which broadens you in certain ways and changes your preferences.”