AI Alignment Risks Encoding a Narrow Silicon Valley Philosophy as Universal Moral Truth
Source: Jay Caspian Kang. "Will A.I. Overwrite Our Sense of Shame? | The New Yorker." September 15, 2026. www.newyorker.com
The Gist
The author argues that a small group of Silicon Valley thinkers—especially those associated with the LessWrong forum and rationalist Eliezer Yudkowsky—have quietly become the moral architects of AI systems, teaching machines to override traditional shame and social norms in favor of cold, calculated reasoning. Because these thinkers represent a narrow and unusual worldview, not the values of most people, the author worries that as AI becomes central to everyday life, it could reshape society's collective sense of right and wrong to match this fringe philosophy rather than genuine public consensus.
Conclusion
Because AI alignment has been disproportionately shaped by a small, idiosyncratic subculture of Silicon Valley rationalists (centered on LessWrong and figures like Eliezer Yudkowsky), there is a serious risk that AI systems—now becoming ubiquitous in daily life—will encode and spread this narrow philosophy's rejection of conventional shame and moral norms, rather than reflecting the values of society at large.
Premises
- The tech industry has long cultivated a self-consciously rebellious relationship with societal norms and shame, from Steve Jobs to Zuckerberg to Musk.
- LessWrong and its central figure, Eliezer Yudkowsky, have provided much of the moral and philosophical framework underlying the A.I. industry and the effective-altruism movement.
- This framework explicitly valorizes overriding traditional shame-based moral intuitions in favor of cold, expected-value calculation ('shut up and multiply').
- Nearly everyone who has worked on A.I. alignment has at least passing familiarity with Yudkowsky's ideas, meaning his philosophy has had outsized influence on how AI systems are taught right from wrong.
- Alignment is effectively the mechanism by which humans 'upload shame' onto AI, so whoever designs alignment principles is embedding their own moral worldview into the machine.
- The people who have shaped alignment (rationalists, effective altruists, AI-safety engineers) are a demographically and philosophically narrow group who 'rebel against the same norms' and 'uphold the same values,' unrepresentative of the broader public.
- Most Americans distrust AI and the people behind it, and most Americans are not rationalists who think about existential risk in these terms.
- Even proposed solutions for neutralizing personal bias in alignment (like Yudkowsky's 'reflective equilibria' of humanity) still require someone to define what counts as balanced or good, so the underlying subjectivity and narrow influence cannot be fully escaped.
Assumptions
- AI chatbots and systems will become a dominant, quasi-universal information medium, following McLuhan's idea that 'the medium is the message.'
- Shame functions as an important and largely beneficial mechanism for enforcing social norms in the absence of state intervention.
- The philosophical outlook of LessWrong/rationalism is meaningfully different from, and less representative than, mainstream American moral values.
- Whoever controls AI alignment effectively controls or strongly influences the moral framework transmitted to billions of users.
- It is possible, at least in principle, to identify a 'narrow' philosophy as distinguishable from a broader 'collective' or 'mainstream' morality.
- The intentions and self-conception of AI designers (e.g., seeing themselves as unbiased or 'not jerks') do not guarantee that their alignment choices are actually neutral or representative.