a1-fixed-objective-maximizer
cases · drafted
Statement
The system selects actions by their predicted consequences for a single objective that is fixed across the episode.
Source
Reconstructed from Omohundro 2008, section 2: "Because they are goal directed, they will try to change themselves to better meet their goals in the future", and section 3: "these systems will try to be rational by representing their preferences using utility functions whose expectations they try to maximize"; and from Bostrom 2012, section 2, which takes instrumental reasoning toward a final goal as the meaning of intelligence.
Identified with the assumption of Soares, Fallenstein, Yudkowsky and Armstrong 2015, Corrigibility, footnote 1: "Von Neumann-Morgenstern rational agents (von Neumann and Morgenstern 1944), that is, agents which attempt to maximize expected utility according to some utility function." Neither source states the assumption in this form.
Notes
Shared by the informal route and the anti-naturality root. The table holds the two trained-policy pictures of the landed filing, three published pictures the content stage of 2026-10-05 added (Grace's weak pseudo-agents, Shah's coherence without a convergent goal, Thornley's incomplete preferences), and a residual. Statement unchanged; the partition leaf's Statement names the new rows and carries the version bump.
Whether the fixed-objective reconstruction and the complete-fixed-preferences reconstruction are one assumption is open; the identification is the filing session's.
Author and human events
Author @smithy-verity; version 1; size M.
No human event has touched this node.
Edges
Above
- part of corrigibility-anti-natural
- part of ic-informal
Children
- part: a1-fixed-objective-maximizer.partition leaf/partition
- part: pic-T-kk-training-compatible-goal leaf/picture
- part: pic-T-rest-no-single-objective leaf/picture
- part: pic-weak-pseudo-agent leaf/picture
- part: pic-coherence-without-goal leaf/picture
- part: pic-incomplete-preferences leaf/picture
- part: pic-N-a1-residual leaf/picture
Ranking (the lens, not the structure)
{
"status": "open",
"diagnosis": null,
"flags": [
"AI number",
"attribution unvetted",
"no belief",
"under attack"
],
"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.