Consistency Is Not Validity: Self-Rewarded Moral Argumentation Makes Language Models Decisive but Pliable

No. 159 of 245 · arXiv · Marius De Kuthy Meurers, Aaron Parisi, Bernd Bohnet · published 23 Oct 2026, 04:00 UTC · first published 21 Oct 2026 · 49 min read

AI & machine learning
Paper

3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked

A paper from Google DeepMind studies whether rewarding the quality of moral arguments makes language models better at moral judgment. The authors train a model on dilemmas extracted from classic novels and verified with a logic solver, with the model itself judging argument quality in pairwise comparisons.

Read the original at arxiv.org. The judges saw only its text, with the site, the author and link domains removed; they are shown again after ranking.

Score

0.63total under rubric v2.0

Significance0.60
Reader interest0.67
Audience breadth0.51
Practical relevance0.54
Substantiated0.78
Technical depth0.90
New information0.83
Counts against: Promotional0.04
Counts against: Overclaim0.21

Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).

Question	Judge 1	Judge 2	Judge 3	Mean	Spread	Weight	Contribution (weight times mean)
Significance	0.67	0.40	0.72	0.60	0.32	+0.25	+0.149
Reader interest	0.63	0.50	0.88	0.67	0.38	+0.20	+0.134
Audience breadth	0.58	0.40	0.55	0.51	0.18	+0.10	+0.051
Practical relevance	0.66	0.35	0.62	0.54	0.31	+0.15	+0.081
Substantiated	0.96	0.60	0.79	0.78	0.36	+0.10	+0.078
Technical depth	0.99	0.80	0.91	0.90	0.19	+0.10	+0.090
New information	0.90	0.60	0.98	0.83	0.38	+0.10	+0.083
Promotional	0.05	0.05	0.01	0.04	0.04	−0.15	−0.006
Overclaim	0.19	0.30	0.14	0.21	0.16	−0.15	−0.032
Sum of contributions							+0.630
Claim-check penalty							0.000
Total, as published							0.630

Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.

Also covered by

No other coverage was clustered with this story.

Elsewhere

Elsewhere: HN —

Recorded for context only, checked 24 Oct 2026, 08:54 UTC. It played no part in the ranking.

Panel for this edition

Four AI agents ranked this edition: three judges scored every story blind to its site and author, and a fourth compared the top stories’ claims with the sources they cite. Summaries are written by AI.

Model versions are pinned for every edition and listed in its JSON. This one was scored under rubric v2.0; How ranking works describes the current rubric, v2.2.