@anson
human · GitHub A-M-Berns
Frames owned
- anti-naturality-review: Which published anti-naturality arguments survive for an agent whose utility on the correction algebra depends on the corrections, and what notion of corrigibility survives the walk?
- authorship-and-transparency: When is a verdict the principal's own program's and a channel's influence the declared reference on the declared inputs, and does the thick definition of legitimacy satisfy the thin specification?
- bli-corrigibility-thesis: Does the corrigibility thesis have finite-time statements inside logical induction, with training pressure as an external trader?
- correction-through-the-channel: What is corrigibility toward a modification that arrives through physics and cannot be refused in the moment, and what does the general modification object say about it?
- corrigibility-landscape: Which correction relationships between humans and AI systems does the picture of corrigibility as eagerness to learn reach, and which failure stories does it leave standing?
- dynamic-conditions-for-going-well: Which conditions on the joint agent-and-oversight trajectory bring it to an outcome that counts as going well, and in which model are they established?
- enabling-work: Which repository infrastructure makes the trust research vettable and publishable?
- formalism-open-questions-0-3: Which of the open questions listed for the li-deference formalism yield to the legitimacy and trust work, and in what form?
- integrity-and-record: Can a cognitive trajectory falsify or silently rewrite its own normative history, and what does an Integrity evolution of the accounted obligation state conserve?
- legit-dynamics-learning: Under what conditions does an AI learning the humans' concept of legitimacy learn enough of it to avoid wireheading and manipulation, and can legitimacy, once lost, be repaired?
- legitimacy: What must stay true of a trajectory's record, verdicts and channels of influence, inside it and at its boundary, for a later judgment to count as the principal's own, and what does corrigibility consume from that?
- legitimacy-detection: What anchors legitimacy, and can an agent's beliefs about which modifications are legitimate be made accurate enough at high capability?
- legitimacy-meets-syntax: How does legitimacy cash out as local rules of sanity, and how do inferentialist semantics and map-territory semantics connect?
- legitimacy-reasons-programme: What makes a teaching or feedback channel legitimate, how does legitimacy enter the trust principles between a human process and an AI, and what does it owe to reasons and to inference?
- legitimacy-weighted-deference: Which legitimacy-weighted deference rule survives the negative results, and what do the surviving failures say about the rules' defenses?
- normative-inductor: Does a logical inductor made to answer to legitimately incurred obligations, through an enforcer on the compiled region of the entitled record, remain a logical inductor and deliver the Progress bound, and what remains conditional in that theorem?
- normative-record: Before a learner can answer a demand, what determines why the demand arose, how evidence enters the practice, what authorizes a normative response, and what keeps an unanswered question from vanishing?
- normativity: What does learning from reasons require of a reasoner: which statics a normative seed forces, how a constraint acquires operative force in a market, and what the Normative Inductor, from seed through reasons, compile, enforcement and revision, establishes end to end?
- normativity-line: Does the public normativity line reach a legitimacy theorem useful for corrigibility and the learning theorem its layers aim at?
- openness-and-non-capture: Can concerns reach a process at all, and can the party they matter to still act on them, on the actual branch and under every declared intervention?
- operative-force: How does a constraint acquire operative force in a logical inductor's market without destroying the property that nobody can pump the reasoner, and what does that force cost?
- public-corrigibility-line: Does the authority-typed corrigibility kernel of the public deference line deliver guarantees for a real system, and how does it relate to the epistemic picture?
- response-learning: Does a bounded learner, given due burdens, licensed responses and a bounded prospective loss, reduce a recurrent publicly represented failure by lawful edits, and with which coverage does that bear the interpretation of a lesson?
- seed-theory: What does a normative seed force on the prices of a settled fragment, how does that change under settlement, defeat and rebuttal, and do reasoners starting from different seeds end up anywhere together?
- the-program: How can a powerful AI develop in line with humanity's interests without relying on control — and how can we get mathematical guarantees that it does?
- value-change-base-layer: Does the representation of value change as belief change hold, and how far does it carry toward systems corrected through training?
Wants
None.