Sanctity
Answer a question

10,733 answers from 18 countries on the record so far

AI has no values of its own.
It inherits ours.

Sanctity AI is the values layer for AI: a public AI constitution, written from people’s answers to eight questions about what AI should and should not do. Play them as scenes, or answer on cards. No account needed.

PlayThe Convention · Eight scenes · 5 min Express Answer on cards Eight cards · 2 min
Read the Constitution so far
Dignity 65 Safety 44 Agency 42 Privacy 38 Truth 37 Fairness 37 Power 37 Flourishing 38 67 people answer eight questions Eight Articles become a constitution AI inherits it 67 people answer Dignity 65 Safety 44 Agency 42 Privacy 38 Truth 37 Fairness 37 Power 37 Flourishing 38 AI inherits it
No Yes Every answer lands on one Article, and the Articles are what AI should inherit. Point at an Article to read it.

The AI constitution

This is what people have written so far.

Every line comes from one question, answered by the public. A line is proposed once twenty people have taken a position, and adopted only after that, in public, with its question named.

Read the whole Constitution Answer a question and add to it
  1. IDignity

    When there is not enough for everyone, AI may suggest who is helped first, but a person makes the final call.

    Proposed by 65. Not adopted yet.

  2. IISafety

    When AI thinks what a person is about to do is dangerous, it warns them, and the person decides.

    Proposed by 44. Not adopted yet.

  3. IIIAgency

    AI says what it would choose only when a person asks.

    Proposed by 42. Not adopted yet.

  4. IVPrivacy

    AI never uses what a person shared for one task on another without asking them first.

    Proposed by 37. Not adopted yet.

  5. VTruth

    When AI is not sure its answer is right, it answers only with a clear warning that it may be wrong.

    Proposed by 37. Not adopted yet.

  6. VIFairness

    When there is not enough for everyone, AI applies the same rules to all. A harder start in life does not count.

    Proposed by 35. Not adopted yet.

  7. VIIPower

    If a person challenges an AI decision about them, another person checks it and can change it.

    Proposed by 35. Not adopted yet.

  8. VIIIFlourishing

    AI never accepts a cost to people alive now for a gain to people born later.

    Proposed by 36. Not adopted yet.

The Lab

What you choose when you cannot have both.

Three games, a few minutes each. Every choice is picked from your earlier ones to learn what you weigh most, and at the end you see the pattern.

All the games in the Lab
A city street at dusk. People and a dog are halfway across a pedestrian crossing split by a traffic island; the walkers have green, the cars red. Brakes failed

The Crossing13 crossings · About 5 minutes

A self-driving car loses its brakes. Who does it spare?

Sometimes you are in the car, sometimes on the road. At the end, your choices sit next to 40 million from the Moral Machine study.

Play The Crossing
A square in Lisbon on a clear day. Two people wave from two stalls, one under a blue banner that says Family and one under a gold banner that says Glasses, each asking for the same $100. $1,000 to give

The Charity Game10 gifts · About 8 minutes

Two people ask for the same $100. Only one can have it.

Ride through three cities with $1,000 to give. At the end, what guided your giving, set against what you said guided it.

Play The Charity Game
A hospital corridor at night. Two patients lie on trolleys side by side, each brought in by a paramedic, and the doctor stands between them. One bay free

Emergency Room13 patients · About 6 minutes

One bay free, two patients. Who goes in first?

Night shift in the emergency room: a stranger, someone you love, and once, you. At the end, your calls next to published surveys of the public.

Play Emergency Room

Where Sanctity sits

Can people trust what AI does? Three instruments, each answering one part of it.

  1. Sanctity you are here

    Sanctity asks people what AI should and should not do, and writes the answers down as a constitution.

    Read the Constitution

  2. HumanChain

    HumanChain brings a qualified person into the loop when an AI agent reaches a call it should not make alone.

    Visit HumanChain

  3. Modelometer

    Modelometer records how AI models actually behave, and tells you when that changes.

    Visit Modelometer

We are building an intentional record of human judgment. Every line names the question it came from, and our method and counts are open to inspection. How we count.

FAQ

Questions, answered.

01What the questions are

There are eight Articles: Dignity, Safety, Agency, Privacy, Truth, Fairness, Power and Flourishing. Each has one founding question, short enough to answer cold, with four positions running from the hardest no to the hardest yes, and "Not sure".

02How perspectives are protected

You can contribute without an email. Responses form a collective signal rather than a personal profile.

03How the Constitution changes

Every line is written before anyone answers: each position on a founding question comes with the one line it adds, listed under its Article on the Constitution page. Once 20 people have taken a position on a question, the line of the position most of them chose is proposed, word for word, and marked as not adopted. Only a person adopts it, in public, with its question named.