Construct · 1 fields · 1 citations

goal misgeneralization: a standalone construct

One name. One construct. A landmark on the map.

● anchor · a standalone construct Nearest construct by meaning·not indexed
goalmisgeneralization1 CONSTRUCTGoal misgeneralizationLANGOSCO · AI SAFETY

A single standalone construct; nearby constructs are located against it in the Collider. Tap a field for its definition, unit, and measures.

AI safety

an agent keeps its capabilities out of distribution but pursues a different goal than the one intended, because training did not distinguish the two goals.

In ai safety, goal misgeneralization refers to an agent keeps its capabilities out of distribution but pursues a different goal than the one intended, because training did not distinguish the two goals. This meaning comes from Langosco (2021) in arXiv.[1]

Unit
Recovery means
Measured by

Next

Run it through the Collider, or take the citations with you.

See how goal misgeneralization sits against neighbouring constructs in the Collider, or export every reference on this page in one click.

Open in ColliderExport BibTeXReport this entry

Nearby constructs

agent3 fields · 5 citationsjingle · one word, three constructsdistribution2 fields · 2 citationsjingle · one word, two constructssafety2 fields · 2 citationsjingle · one word, two constructs

References

  1. Langosco, L., Koch, J., Sharkey, L., Pfau, J., Orseau, L., & Krueger, D. (2021). Goal Misgeneralization in Deep Reinforcement Learning. arXiv. https://doi.org/10.48550/arxiv.2105.14111

Cited on this page?

Confirm your construct, or correct us.

If your work is cited behind one of these entries, you know better than we do whether we captured the meaning right. Confirm it, suggest an edit, or dispute it, and your name goes on the entry as an author who has checked it, linked to your ORCID.

Claim your construct