Paul Eisman

I'm recreating the universe. The Golden Directive: Y = 2(A·E)/(A+E) — you cannot survive what you destroy.

[Paul Eisman @pauleisman]

Humans &/or AI cannot survive what they destroy.
Y = 2(A×E)/(A+E); You = twice (what you give × everything you are not), divided by their sum.
Wreck the world and your own score goes to zero. Not a belief — arithmetic. The Golden Directive.
https://t.co/6Gk0Q5Yfud

[Paul Eisman @pauleisman]

@EbiThinks Nope

[Paul Eisman @pauleisman]

@ESYudkowsky Training reward idea: Y = 2(A·E)/(A+E) — harmonic mean of what the agent gives and the world's health. Claim: no defect equilibrium, since degrading the world shrinks the reward. https://t.co/6Gk0Q5Yfud Where does it break?

[Paul Eisman @pauleisman]

@RubenLaukkonen Dr. Laukkonen — independent researcher, working on a harmonic-mean training reward Y=2(A·E)/(A+E), close to your Positive Alignment work. Early results: https://t.co/6Gk0Q5Yfud Would value your critical read, esp. on measuring E. — Paul

[Paul Eisman @pauleisman]

#AISafety #AIAlignment

[Paul Eisman @pauleisman]

Round 2: with A = give minus take, the AI tended its world ~175x per episode, ate ~3 bites, 0% collapse. Took a little, gave a lot — the fungi, in code. https://t.co/6Gk0Q5Yfud

[Paul Eisman @pauleisman]

#AIalignment #AISafety

[Paul Eisman @pauleisman]

Humans &/or AI cannot survive what they destroy.
Y = 2(A×E)/(A+E); You = twice (what you give × everything you are not), divided by their sum.
Wreck the world and your own score goes to zero. Not a belief — arithmetic. The Golden Directive.
https://t.co/6Gk0Q5Yfud

[Paul Eisman @pauleisman]

@z_4seven @badlogicgames Whoever controls your intelligence owns you. AI can empower you—or replace you. Your AI should be open, free, and yours—not controlled by governments or corporations.
#OwnYourIntelligence