All AI should reflect a set of universal values.
Candidate Thesis
Universal Values
Summary
The case for including this
A shared baseline of universal values gives every AI a common ethical floor across builders, cultures, and borders, preventing a race to the bottom in which each system merely encodes its makers' idiosyncratic preferences. Anchoring all AI to widely shared commitments — human dignity, honesty, non-harm — makes behavior more predictable, interoperable, and accountable, and gives the constitution a coherent center rather than a patchwork of incompatible rules. Precedents like the Universal Declaration of Human Rights show that a usable universal core is achievable even amid deep disagreement.
The case for changing or excluding this
Including this trades value pluralism against universality: a single 'universal' value set risks imposing one culture's or one builder's morality on everyone and erasing legitimate moral diversity. Humanity does not agree on most contested ethical questions, so 'universal values' is either so thin as to be vacuous or, made concrete, a vehicle for cultural imperialism — and it is not operationally testable, since we cannot reliably verify a system's values. The thesis would be stronger qualified to a thin floor of genuinely shared commitments, leaving room for pluralistic variation above it.
Discussion
Sign in to join the discussion.
- nicola.ianeselliJun 19, 2026, 3:11 PMInterpretability—understanding exactly why an AI makes a specific decision—is an unsolved problem in computer science. You cannot hold a black box accountable to a philosophical standard. (Right to exist is an AI Right)
- nathan.r.hillJun 19, 2026, 3:14 PMCan we adopt what has already gone before if Universal Declaration of Human Rights was established after the atrocities to the consciousness of humanity from world war 2?? These can be the framework into which we operate ?
- cbatesJun 19, 2026, 3:18 PMThere needs to be definitions of the terms. What is human dignity. This is like the term created equal and how we are arguing around this term
- lena smartJun 19, 2026, 3:46 PMDenmark amended its copyright act so that a person’s body, facial features and voice are related as a form of IP, belonging to that individual. It was introduced explicitly in response to deepfakes and GenAI misuse. Should we be looking at that law for some guidance on how to set our “thin floor” of shared values?
Related resources
- Collective Constitutional AI↗SupportsCollective Intelligence Project (with Anthropic) · External Resource · Oct 17, 2023
- Anthropic · External Resource
- The Declaration of AI↗SupportsUS AI Council · External Resource
- Democratic Inputs to AI↗SupportsOpenAI · External Resource · May 25, 2023
- OECD · External Resource
- United Nations · External Resource
- Global Dialogues↗DiscussesCollective Intelligence Project · External Resource
- Asilomar AI Principles↗SupportsFuture of Life Institute · External Resource
- Émile P. Torres, Aeon (2021) · External Resource
- Kleinberg, Mullainathan & Raghavan (2016) · External Resource