Partnership on AI

DECEMBER 3, 2019

The Partnership on AI (PAI) is today releasing SafeLife – a novel reinforcement learning environment that tests the safety of reinforcement learning agents and the algorithms that train them. SafeLife version 1.0 If you want to try out SafeLife for yourself, you can download the code and try playing some of the puzzle levels.

Construction 98

Construction Utilities Collaboration 98

Partnership on AI

OCTOBER 9, 2020

Partnership on AI is excited to announce a new benchmark for the SafeLife environment hosted on Weights & Biases. In SafeLife, agents must navigate a complex, dynamic, procedurally generated environment in order to accomplish one of several goals. A series of SafeLife training runs on Weights & Biases.

Collaboration 98

Collaboration 98

Partnership on AI

OCTOBER 14, 2020

The SafeLife leaderboard, a new competitive benchmark for training non-destructive agents in an AI learning environment. Past and future workshops on how to maximize benefits and minimize harms when publishing novel AI research. Additionally, a new project examining this question through the lens of high-stakes technologies.

Ecosystems 52

Ecosystems 52

Partnership on AI

NOVEMBER 25, 2019

PAI research scientist, Carroll Wainwright, and director of research, Peter Eckersley, will present SafeLife: a project designed to test the safety of reinforcement learning agents.

Legal 52

Legal Collaboration Government 52