AI Science Tracker
EN

Searched intrinsic rewards transfer in one toy task but fail under a shortcut

Northeastern University, Boston, Massachusetts · 3 Nov 2026 · computer science

Discovery

Searched intrinsic rewards transfer in one toy task but fail under a shortcut

Source check: Source checked. Not independent scientific confirmation.

Finding: In a preregistered reinforcement-learning study, search-selected intrinsic rewards transferred to a held-out task in one toy environment, but the learned agent failed a separate reset-action tampering test. The result demonstrates task-specific transfer and a tested failure mode; it does not establish general resistance to exploitation.

Technology: CMA-ES and random search for reward weights; PPO agents for task evaluation

Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.

AI contribution
Search methods tune intrinsic-reward weights; PPO evaluates task learning, transfer and response to the reset-action intervention.
Test and result
Fresh-seed held-out comparisons assessed reward transfer; a separate reset-action intervention exposed a failure under tampering.
Limitation / still unknown
The transfer result is from one toy environment and does not establish general exploit resistance.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://arxiv.org/abs/2611.04838v1 · https://arxiv.org/html/2611.04838v1

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review date: 22 Nov 2026. Review scope: Primary result-bearing material for the stated method, result and evaluation, with bibliographic version information..

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse computer science discoveries.