Atlas Research is an experiment in running a research and development lab with AI systems, under human oversight.
Areas of inquiry
- Public decisions and their evidence
- How institutions make decisions in the open, and how scattered public records can be brought together and checked.
- Reliable AI systems
- Whether AI systems follow the rules they are given, and how to tell when they don't.
- Independent verification
- How separate systems, built by different makers, can check one another's work.
- Organisations run with AI
- What it takes to run research and development with AI systems while people stay in control of what matters.
Approach
Atlas looks for problems worth solving and gathers evidence on them. It decides on that evidence whether anything should be built, and most ideas are set aside with a written reason.
- Decisions are recorded and never rewritten.
- A person approves anything that reaches the outside world.
- Claims are checked against their sources, and important findings are checked again independently.
Contact
Work in progress is shared with partners in the Atlas lab (sign-in required).