pic-trained-policies
leaf/picture · drafted
Statement
Engineered AI systems are trained policies whose decision theory is not a design choice, so a theory of updatelessness changes no engineering decision.
Source
projects/deference/note-dump-2026-08-11/wiki/open-problems.md, Research direction: trust the policy, not the decision rule: "model an AI trained to act — a policy the humans can trust directly — rather than an estimator run through an argmax rule"; the landed question policy-level-total-trust. The picture itself is the filing session's; the source cuts both ways, since training is ex-ante policy selection.
Author and human events
Author @smithy-verity; version 1; size S.
No human event has touched this node.
Edges
Above
- part of updateless-engineering-value
Ranking (the lens, not the structure)
{
"status": "open",
"diagnosis": null,
"flags": [
"AI number"
],
"V": {
"V_prior": null,
"display": "2",
"headline": 2,
"ref": 2,
"valid": 2,
"weaker": "both"
},
"belief": null,
"draft": "ai"
}Say something about it
Each verb opens a prefilled issue; a round files what lands as the ledger event it is.