Agent, Learning, Reinforcement learning, Machine learning, Natural environment, Artificial intelligence

REALab: Conceptualising the Tampering Problem

On Nov 18, 2020
@DeepMind shared
Building safe AI requires accounting for the possibility of feedback corruption. The REALab platform provides new insights by studying tampering in simulation: https://t.co/t2rjxjnpJT More reading on REALab & Decoupled Approval: https://t.co/bg2sDKXAFz & https://t.co/2zjtslE63O https://t.co/cC25m3hOuY
Open

By Tom Everitt, Ramana Kumar, Jonathan Uesato, Victoria Krakovna, Richard Ngo, Shane Legg

medium.com
On Nov 18, 2020
@DeepMind shared
Building safe AI requires accounting for the possibility of feedback corruption. The REALab platform provides new insights by studying tampering in simulation: https://t.co/t2rjxjnpJT More reading on REALab & Decoupled Approval: https://t.co/bg2sDKXAFz & https://t.co/2zjtslE63O https://t.co/cC25m3hOuY
Open

REALab: Conceptualising the Tampering Problem

REALab: Conceptualising the Tampering Problem

By Tom Everitt, Ramana Kumar, Jonathan Uesato, Victoria Krakovna, Richard Ngo, Shane Legg

Click here to read the article

Click here to read the article

Scalable agent alignment via reward modeling: a research direction Jan Leike DeepMind David Krueger∗ DeepMind Mila Tom Everitt DeepMind Miljan Martic DeepMind Vishal Maini DeepMind Shane ...

POET: Endlessly Generating Increasingly Complex and Diverse Learning Environments and their Solutions through the Paired Open-Ended Trailblazer

POET: Endlessly Generating Increasingly Complex and Diverse Learning Environments and their Solutions through the Paired Open-Ended Trailblazer

Uber AI Labs introduces the Paired Open-Ended Trailblazer (POET), an algorithm that leverages open-endedness to push the bounds of machine learning.

Learning human objectives by evaluating hypothetical behaviours

Learning human objectives by evaluating hypothetical behaviours

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

Learning from Human Preferences

Learning from Human Preferences

One step towards building safe AI systems is to remove the need for humans to write goal functions, since using a simple proxy for a complex goal, or getting the complex goal a bit wrong, ...

Measuring Safety in Reinforcement Learning

Measuring Safety in Reinforcement Learning

Weights & Biases, developer tools for machine learning

MetaGenRL: Improving Generalization in Meta Reinforcement Learning

MetaGenRL: Improving Generalization in Meta Reinforcement Learning

Biological evolution has distilled the experiences of many learners into the general learning algorithms of humans. Inspired by this process, MetaGenRL distills the experiences of many ...

Accepted Papers

Accepted Papers

Language in Reinforcement Learning Workshop

Artificial Intelligence in Oil & Gas Production

Artificial Intelligence in Oil & Gas Production

Solution Seeker, a Norwegian tech start-up and spin-off from the Norwegian University of Science and Technology, is developing the world’s first AI for real-time production optimisation.

Behaviour Suite for Reinforcement Learning (bsuite)

Behaviour Suite for Reinforcement Learning (bsuite)

bsuite is a collection of carefully-designed experiments that investigate core capabilities of a reinforcement learning (RL) agent - deepmind/bsuite

Training intelligent adversaries using self-play with ML-Agents

Training intelligent adversaries using self-play with ML-Agents

In the latest release of the ML-Agents Toolkit (v0.14), we have added a self-play feature that provides the capability to train competitive agents in adver...

Google’s ML-fairness-gym lets researchers study the long-term effects of AI’s decisions

Google’s ML-fairness-gym lets researchers study the long-term effects of AI’s decisions

Google's ML-fairness-gym, which was released in open source, allows AI practitioners and data scientists to study the fairness of AI systems.