Evidence: Falsification Dashboard
The Falsification Dashboard: Why I Invite You To Prove Me Wrong
Independent AI safety researcher, London. Verify any claim.
Most research programmes defend their claims. I invite you to attack mine. The falsification dashboard lists every claim in the Eden programme, and for each one, states exactly what evidence would prove it wrong. 37 claims. 34 standing. 3 contested. 0 falsified. The invitation is open.
Why Falsification Is the Hardest Page to Write
It is easy to make claims. It is hard to specify in advance what would prove them wrong. It is hardest of all to do this publicly, in real time, with cryptographic proof that the claims existed on a specific date and the falsification conditions were stated before anyone tried to meet them.
This is the opposite of marketing. Marketing amplifies strengths and hides weaknesses. Falsification amplifies weaknesses and invites their discovery. The falsification dashboard is not a confidence display. It is a vulnerability disclosure. Here is everything I think is true. Here is what would prove each one wrong. Go.
What Happens When a Claim Falls
When (not if) a claim is falsified, three things happen: (1) the claim is marked FALSIFIED on the dashboard, (2) the research record is updated with the evidence that disproved it, and (3) the person who provided the evidence is credited, unless they prefer anonymity.
This is how science is supposed to work. Claims are published. Evidence accumulates. Wrong claims are discarded. Right claims get stronger. The traditional system does this slowly, through journals and retractions. I do it in public, with cryptographic proof, on a website anyone can check.
Zero claims have been falsified so far. I do not expect this to last. When the first claim falls, it will be the strongest possible evidence that the system works, because the system exists to make claims falsifiable, not to make them invulnerable.