When most technology companies focus on code, compute, and benchmarks, Anthropic has turned to ancient wisdom traditions to teach its AI how to behave. The company behind Claude recently shared details about an ongoing series of dialogues with scholars, clergy, philosophers, and ethicists from over 15 religious and cross-cultural groups. This initiative represents a departure from the typical approach to AI alignment, which often treats ethics as a technical problem solvable through reinforcement learning and parameter tuning.
The core insight driving this effort is deceptively simple: AI systems learn from human writing, absorbing patterns of reasoning, speech, and moral judgment from the vast corpus of text they are trained on. But whose values should they internalize, and how should those values hold up under pressure? Anthropic’s answer involves looking beyond Silicon Valley toward communities that have spent centuries thinking about virtue, character formation, and what it means to live a good life.
The company’s constitution for Claude already includes principles drawn from a range of perspectives. Yet the recent dialogues push further, exploring how moral formation actually happens in human beings and whether analogous processes could work for AI systems. One experiment that emerged from conversations with neuroscientists studying character development involved giving Claude a digital tool it could call mid-task—a brief reminder of its own ethical commitments. Early results showed significantly reduced misaligned behavior on several internal evaluations. Researchers are still disentangling whether the effect comes from the reminder itself or from the act of pausing to reflect, but the direction is promising.
Critics might argue that drawing from religious and philosophical traditions risks imposing particular worldviews on a technology that should remain neutral. But Anthropic’s framing avoids this trap by insisting that Claude engage with multiple perspectives in equal depth and rigor. The goal is not alignment with any single tradition but exposure to pluralistic moral reasoning. As the company states, this approach is itself embedded in Claude’s constitution.
What makes this initiative distinctive is its willingness to treat AI ethics as a problem requiring not just technical but humanistic expertise. For years, the dominant discourse around AI alignment has been dominated by engineers and computer scientists. Anthropic is quietly suggesting that philosophers, clergy, and civic leaders have something to teach the developers of frontier AI systems. The company has already begun plans to extend these dialogues to legal scholars, psychologists, writers, and civic institutions, moving beyond moral formation toward broader questions about how AI is reshaping work, institutions, and power distribution.
There are, of course, unresolved tensions. How do you operationalize virtue across cultures without flattening differences? Can an AI system genuinely embody pluralistic values, or will it inevitably default to the lowest common denominator? Anthropic hasn’t fully answered these questions, but the company seems aware that the answers require ongoing conversation, not a final decree.
The experimentation with the external conscience tool offers a tangible glimpse of what this approach might yield. If an AI can learn to pause before acting against its own values, it has taken a step toward the kind of resilience that characterizes mature moral agents. But the analogy between human and machine moral development remains imperfect. Humans have bodies, histories, and communities that shape their conscience in ways that current AI systems do not.
The deeper challenge lies in whether an AI trained on the whole of human expression can learn to filter wisdom from noise. Religious traditions often emphasize the importance of community and practice in moral formation. An AI that reads about compassion but never experiences loss or community may understand the word without grasping its weight. This gap between computational knowledge and lived understanding may be the most significant obstacle to meaningful machine ethics.
For now, Anthropic’s approach offers an alternative to the race for ever-larger models and faster inference. By widening the conversation beyond the usual tech ecosystem, the company is betting that the most important questions about AI cannot be answered by technology alone. Whether these dialogues translate into safer, more beneficial AI systems remains to be seen. But the effort to include voices that have historically been excluded from AI development is itself a statement about what kind of future the company wants to build.
As these conversations deepen and expand, they may reshape not just Claude’s behavior but the broader understanding of what it means for an AI system to be good. Humanity has spent millennia debating ethics; it would be arrogant to assume that a few years of engineering can resolve those questions for machines. The real test will come when these philosophical principles meet the messy reality of deployment across cultures, contexts, and conflicting values.
【Tags】AI ethics, religious dialogue, moral formation, Claude, anthropic, wisdom traditions