Prescience Labs

Experiments

Prototypes for thinking, deciding, and building with AI.

Research

AI safety

How intelligent systems can be made reliable, aligned, and accountable to the people who depend on them.

Featured experiment

chAImbers

Answers from private data.
Without handing it over.

An interactive prototype exploring how two parties can agree on a question, set the rules for private evidence, and ask an AI judge for a tightly scoped answer.

Explore chAImbers
Judgy, the chAImbers AI judge, wearing a barrister’s wig and holding scales of justice.
  1. Agree the question

    Both sides negotiate the question and the rules. Nothing moves forward until they agree.

    A proposed question in chAImbers, awaiting agreement from the other side.
  2. Test in private

    Try the judge on your own evidence before submitting. Explore its answer and the reasoning behind it.

    A private simulation showing the judge’s reasoning and supporting evidence.
  3. Limit what comes out

    The shared verdict is just TRUE or FALSE. Either side can appeal and ask the judge to reconsider.

    The chAImbers verdict screen returning TRUE for the agreed question.

Research prototype. Encryption, sealing and enclave attestation are simulated in this demo.