Essays · Human values

Human values

Essays and plain-English guides on human oversight, AI governance and the values AI should hold.

Human values3 min read Whose values should AI hold? Every AI model encodes values. The honest question is not whether, but whose. The case for a values layer that is chosen in the open rather than smuggled in by default. 25 June 2026

Human values2 min read Can one AI hold values for everyone? No single set of values fits every person and context. Value pluralism is the honest stance for AI: a values layer that can hold genuine differences rather than imposing one default on all. 25 June 2026

Human values2 min read What is the 'whose values' problem in AI alignment? AI alignment usually asks how to align a model to human values, as if there were one set. The harder, neglected half is whose values. Why the 'whose' question is the one that matters most. 25 June 2026

Human values2 min read What is the values layer for AI? The values layer for AI is the part of a system where what it should care about is made explicit and contestable, instead of inferred and hidden inside the model. 25 June 2026

Human values2 min read Human values in AI: whose values should AI hold? Every AI model encodes values; the honest question is whose, and who decides. Human values in AI is the problem of choosing and legitimising those values, with a human in the loop. 25 June 2026

Human values2 min read What does legitimacy mean for AI? Legitimacy for AI means the people affected by its values have a say in them. Why accuracy is not the same as legitimacy, and what giving the governed a vote could mean for the values layer. 25 June 2026

Human values2 min read Is every AI model an opinion? Every AI model encodes a point of view through the data and choices that made it. Treating models as neutral hides the opinions inside them. The honest move is to ask whose. 25 June 2026

Human values2 min read Should experts get more say over AI's values? On questions of value, everyone should weigh equally; expertise belongs to questions with right answers. Why the distinction between facts and values decides who gets a say over AI. 25 June 2026

Human values2 min read Can dignity be engineered into AI? Dignity is usually an AI slogan. It can be a design constraint instead: a line decisions must not cross, built into the values layer rather than printed in the marketing. 25 June 2026

Human values2 min read Does AI need a constitution? A constitution for AI is a metaphor for making a system's values explicit, written, and contestable, rather than buried in its weights. Why explicit values beat implicit ones, including the human-oversight duty as the law comes into force. 25 June 2026