Sine Qua Non EstInstitute for Human–AI ResearchGitHub

Institute for Human–AI Research

Sine Qua Non Est

We develop tools and methods to help researchers check AI-assisted work and share the evidence behind their conclusions.

Iudicium humanum est condicio sine qua non est scientia.

Human judgment is the condition without which there is no science.

Concordia hominis et machinae condicio sine qua non est.

Harmony between human and machine is the indispensable condition.

Explore the projectsRead the mission

Human × AI

Vision

Towards Responsible and Collaborative AI-Assisted Science

AI should help researchers explore more questions without weakening the evidence needed to answer them.

AI can help search literature, write code, and compare explanations. Those outputs still need to be checked against sources and observations. We want that checking to be part of the research process, with enough detail for other teams to challenge a finding or build on it.

Concordia hominis et machinae condicio sine qua non est.

Harmony between human and machine is the indispensable condition.
A researcher looks through a telescope beneath a sky of connected stars. The telescope and a computational grid point toward the same bright star.
A shared horizon for discovery

Explore new questions. Test the answers against evidence.

Mission

Keep the human in the loop

Researchers need to see how an AI result was produced before they rely on it.

We are developing tools and methods that expose the sources an agent used, the steps it took, and unresolved questions. When an agent proposes an explanation, the researcher should be able to trace it to observations and test alternatives.

Researchers must retain the authority to change the method or stop the work.

A large inspection lens enlarges marks on a research sheet. One mark differs from the others. A pencil rests beside the sheet for the researcher's review.
A closer look at the evidence

An explanation is a starting point for investigation.

Research agenda

What we plan to study and build

Our research agenda has four priorities.

Research methods

Study how AI assistance affects hypothesis formation and the review of evidence.

Repeatable experiments

Develop software to rerun experiments with recorded inputs, code, and outputs.

Reporting practices

Document how to report AI involvement, uncertainty, and negative results.

Sharing across teams

Develop controls for sharing research across organizations and recording contributions.

An open research notebook holds an observation sketch on one page and a connected set of experiment marks on the other. Loose sheets and a bookmark keep the supporting work together.
An open record of discovery

A finding is more useful when others can repeat the method.

Collaboration

Help the next team build on your work

Use these steps to prepare AI-assisted work for another team to review.

  1. 01

    Agree on the question

    Before delegating to an agent, define the question, the evidence needed, and the reviewer.

  2. 02

    Share with context

    Include sources, methods, and software versions so another team can assess the result.

  3. 03

    Keep what you learned

    Record corrections and negative results so others can avoid repeating failed approaches.

  4. 04

    Make responsibility clear

    State who reviews the conclusions and who may use or release the work.

Two distinct research buildings stand on opposite banks, joined by a gold bridge. People meet along the bridge beneath a shared constellation.
A bridge between research communities

Shared evidence gives teams a basis for working together.