arg-miri-2015-e-channel
leaf/step · drafted
Statement
Utility mixing and utility indifference fail the corrigibility desiderata of Soares et al. 2015, and indifference leaves an incentive routed through the estimator that computes the compensation, for an agent whose utility on the correction algebra depends on the corrections.
Source
Soares et al. 2015, sections 3 and 4, and Armstrong 2016 (double indifference); projects/deference/corrigibility-note-dump-2026-10-01/corrigibility-discussion-outline.md, item 4 (desiderata 2 to 5 and the section 4.2 pathologies are its remaining parts); the clean-room package projects/deference/faf-cleanroom-2026-10-02/packages/corr-indifference (one text item flagged; Lean statements correct). The author's reply, projects/deference/corrigibility-note-dump-2026-10-01/position-statement.md, section 2.7: "I agree with your idea that indifference was simply the wrong target -- I can be far from indifferent about a fact, but still eager to learn."
Notes
Outline item 4. Its edge from indifference-wrong-target records the author's reply: if indifference was the wrong target, the e-channel objection to indifference does not reach an agent that is eager to learn.
Author and human events
Author @smithy-verity; version 1; size M.
No human event has touched this node.
Edges
Above
- part of anti-naturality-arguments-as-stated
- given of indifference-wrong-target
- given indifference-wrong-target
Ranking (the lens, not the structure)
{
"status": "open",
"diagnosis": null,
"flags": [
"AI number",
"given indifference-wrong-target"
],
"V": {
"V_prior": null,
"display": "2",
"headline": 2,
"ref": 2,
"valid": 2,
"weaker": "both"
},
"belief": {
"J": 0,
"mid": "7/20",
"range": [
"0/1",
"17/20"
],
"width": "1/1"
},
"draft": "ai"
}Say something about it
Each verb opens a prefilled issue; a round files what lands as the ledger event it is.