DeepMind Safety Research – Medium

DeepMind Safety Research

DeepMind Safety Research

Goal Misgeneralisation: Why Correct Specifications Aren’t Enough For Correct Goals

By Rohin Shah, Vikrant Varma, Ramana Kumar, Mary Phuong, Victoria Krakovna, Jonathan Uesato, and Zac Kenton. For more details, check out…

9 min readOct 7, 2022

--

1

Goal Misgeneralisation: Why Correct Specifications Aren’t Enough For Correct Goals

--

1

DeepMind Safety Research

Discovering when an agent is present in a system

New, formal definition of agency gives clear principles for causal modelling of AI agents and the incentives they face.

4 min readAug 25, 2022

--

Discovering when an agent is present in a system

--

DeepMind Safety Research

Your Policy Regulariser is Secretly an Adversary

By Rob Brekelmans, Tim Genewein, Jordi Grau-Moya, Grégoire Delétang, Markus Kunesch, Shane Legg, Pedro A. Ortega

7 min readMar 24, 2022

--

Your Policy Regulariser is Secretly an Adversary

--

DeepMind Safety Research

Avoiding Unsafe States in 3D Environments using Human Feedback

By Matthew Rahtz, Vikrant Varma, Ramana Kumar, Zachary Kenton, Shane Legg, and Jan Leike.

5 min readJan 21, 2022

--

Avoiding Unsafe States in 3D Environments using Human Feedback

--

DeepMind Safety Research

Model-Free Risk-Sensitive Reinforcement Learning

By the Safety Analysis Team: Grégoire Delétang, Jordi Grau-Moya, Markus Kunesch, Tim Genewein, Rob Brekelmans, Shane Legg, and Pedro A…

7 min readNov 11, 2021

--

Model-Free Risk-Sensitive Reinforcement Learning

--

DeepMind Safety Research

Progress on Causal Influence Diagrams

By Tom Everitt, Ryan Carey, Lewis Hammond, James Fox, Eric Langlois, and Shane Legg

12 min readJun 30, 2021

--

Progress on Causal Influence Diagrams

--

DeepMind Safety Research

An EPIC way to evaluate reward functions

How can you tell if you have a good reward function? EPIC provides a fast and reliable way to evaluate reward functions.

9 min readApr 16, 2021

--

An EPIC way to evaluate reward functions

--

DeepMind Safety Research

Alignment of Language Agents

By Zachary Kenton, Tom Everitt, Laura Weidinger, Iason Gabriel, Vladimir Mikulik and Geoffrey Irving

3 min readMar 30, 2021

--

Alignment of Language Agents

--

DeepMind Safety Research

What mechanisms drive agent behaviour?

By the Safety Analysis Team: Grégoire Déletang, Jordi Grau-Moya, Miljan Martic, Tim Genewein, Tom McGrath, Vladimir Mikulik, Markus…

8 min readMar 5, 2021

--

1

What mechanisms drive agent behaviour?

--

1

DeepMind Safety Research

Understanding meta-trained algorithms through a Bayesian lens

By Vladimir Mikulik , Grégoire Delétang, Tom McGrath, Tim Genewein, Markus Kunesch, Jordi Grau-Moya, Miljan Martic, Shane Legg, Pedro A…

10 min readDec 3, 2020

--

Illustration of two coin-flip environments: the “fair coins” and the “bent coins” environment.

--

DeepMind Safety Research

DeepMind Safety Research

We research and build safe AI systems that learn how to solve problems and advance scientific discovery for all. Explore our work: deepmind.com

Help
Status
About
Careers
Press
Blog
Privacy
Terms
Text to speech
Teams