[1906.01820] Risks from Learned Optimization in Advanced Machine Learning Systems
Abstract:We analyze the type of learned optimization that occurs when a learned model (such as a neural network) is itself an optimizer - a situation we refer to as mesa-optimization, a neologism we introduce in this paper. We believe that the possibility of mesa-optimization raises two important questions for the safety and transparency of advanced machine learning systems. First, under what circumstances will learned models be optimizers, including when they should not be? Second, when a learned model is an optimizer, what will its objective be - how will it differ from the loss function it was trained under - and how can it be aligned? In this paper, we provide an in-depth analysis of these two primary questions and provide an overview of topics for future research.
Risks from Learned Optimization in Advanced Machine Learning Systems Evan Hubinger∗ , Chris van Merwijk∗ , Vladimir Mikulik∗ , Joar Skalse∗ , and Scott Garrabrant† June 11, 2019 arXiv:1906.01820v3 [cs.AI] 1 Dec 2021 Abstract We analyze the type of…
saved by
related reading
- Risks from Learned Optimization: Introduction — LessWronglesswrong.com
- Risks from Learned Optimization: Introduction — AI Alignment Forumalignmentforum.org
- Conditions for Mesa-Optimization — AI Alignment Forumalignmentforum.org
- Risks from Learned Optimization: Introduction — AI Alignment Forumalignmentforum.org
- [1906.01820] Risks from Learned Optimization in Advanced Machine Learning Systemsarxiv.org
- Deceptive Alignment — AI Alignment Forumalignmentforum.org
- The Inner Alignment Problem — AI Alignment Forumalignmentforum.org
- Mesa-Optimization — AI Alignment Forumalignmentforum.org
- [2305.18290] Direct Preference Optimization: Your Language Model is Secretly a Reward Modelarxiv.org
- Metagaming matters for training, evaluation, and oversightalignment.openai.com
- Metalearning or Learning to Learn Since 1987people.idsia.ch
- NL.pdfabehrouz.github.io