AI ALIGNMENT FORUM
AF

IndraG
030
Message
Dialogue
Subscribe

Posts

Sorted by New

Wikitag Contributions

Comments

Sorted by
Newest
No wikitag contributions to display.
An overview of 11 proposals for building safe advanced AI
IndraG3mo00

A simple criticism of idea six is that physics is not independent from human behavior, an argument I heard made at least by David Deutsch and I think also Anders Sandberg. In other words, to be really capable of predicting the future trajectories of planets, (our) civilization will have to predicted as well.

Reply
Sections 1 & 2: Introduction, Strategy and Governance
IndraG2y00

threats being harmful to their target, the execution of threats seems to constitute an inefficiency:

also when the threats are targeted at (Pareto-dominated) inefficiency, i.e. at (conditional on) any actions other than (coordinating on) the most efficient?

As a concrete counter-example, there are productivity/self-control tools, wherewith people elect to target and/or execute threats on themselves to help elicit better behavior. The legal system is basically also collective threats that help us behave better, but is it inefficient such that we should do better without? I think the opposite, such that any one threat can not only be harmful, but also neutral or beneficial.

Reply
Sections 1 & 2: Introduction, Strategy and Governance
IndraG2y00

and a defecting equilibrium which does not.

Why doesn't this also require coordination? Also, there also seems to be a mixed equilibrium where both players randomize their strategies 50/50.

For the Chicken game, the mixed strategy equilibrium is not 50/50 but more specifically 90/10. And a mutual defection can also not constitute crash, but instead allow for further and stable repeated play, particularly if a mixed strategy equilibrium is coordinated and acted upon.

Reply
No posts to display.